CN114004984B - A method and system for comparing drawings of high-voltage cable accessories process library - Google Patents

Abstract

The present invention relates to a drawing comparison method and system based on a high-voltage cable accessory process library, the method comprising: scanning the process drawings of the high-voltage cable to construct a process drawing template; scanning to obtain the drawing file to be compared, performing drawing similarity detection, and obtaining the process drawing template with the highest similarity; segmenting the text area and the view area from the drawing file to be compared, positioning, text detection and recognition of the text area, and then comparing it with the text information corresponding to the process drawing template to determine whether there is a difference; positioning the view area with annotated dimensions, performing character detection and recognition on the positioned annotated dimensions, and then comparing it with the annotated dimension information corresponding to the process drawing template to determine whether there is a difference; if it is determined that a difference occurs, data tracing is performed and the difference position is marked. Compared with the prior art, the present invention does not require a large amount of data sets for training, can specifically determine the difference points in the drawing, and improve the accuracy of the comparison.

Description

A method and system for comparing drawings of high-voltage cable accessories process library

Technical Field

The present invention relates to the technical field of high-voltage cable accessory comparison, and in particular to a method and system for comparing drawings of a high-voltage cable accessory process library.

Background Art

At present, the process drawings of high-voltage cable accessories of cable companies are the most important documents of the company. Their review and storage are basically realized in the form of traditional paper, which has the problems of high storage management cost and low review efficiency. In view of the fact that there are many manufacturers of high-voltage cable accessories, there are many types of cable accessories, with different types of characteristics and limitations, and some manufacturers do not have a certain historical background. The generated high-voltage cable accessories have not been precipitated by long-term operating experience data in some design links, and the process design of the drawings often changes.

Traditional high-voltage cable accessories process review is to manage paper drawings. There is relatively little research on image recognition technology for cable professional high-voltage cable accessories drawings. However, image comparison, recognition, and segmentation are the most studied topics in the field of image processing. However, image recognition technology is still the focus of research.

Drawing comparison is to compare the differences between two drawings. It is an important method to improve the efficiency of drawing review and reduce the cost of high-voltage cable accessories review.

As for the method of drawing comparison, there is currently AI recognition technology, which uses neural networks for comparison and recognition, but AI recognition technology requires a large number of drawing data sets, and the number of data sets determines the accuracy of AI recognition. The accuracy of the network model trained with a small data set is difficult to guarantee. Secondly, since the resolution of the scanned drawings and other documents improved by the manufacturer is different from the resolution of the drawings in the process library, if training is to be carried out, each data set image needs to be pre-processed, and the process is very cumbersome and takes a lot of time. Therefore, in most cases, the process drawings of high-voltage cable accessories that can be provided are not sufficient to support the establishment of AI recognition technology.

Summary of the invention

The purpose of the present invention is to overcome the defects of the above-mentioned prior art and provide a high-voltage cable accessory process library drawing comparison method and system that is conducive to standardizing the drawing review process, improving the drawing review efficiency, ensuring the drawing review quality and the storage management of digital process drawings.

The purpose of the present invention can be achieved by the following technical solutions:

A drawing comparison method based on a high-voltage cable accessory process library comprises the following steps:

Scan the process drawings of high-voltage cables, pre-process them, and use them as process drawing templates, and attach label information to each process drawing;

After scanning and preprocessing the process drawings of the high-voltage cables to be compared, the drawing files to be compared are obtained;

Perform drawing similarity detection on the drawing file to be compared and each process drawing template to obtain the process drawing template with the highest similarity;

Segmenting a text area and a view area from the drawing file to be compared, positioning, detecting and recognizing text in the text area, and comparing the text recognition result of the text area with the text information corresponding to the process drawing template with the highest similarity to determine whether there is a difference;

Positioning the dimension annotation of the view area, performing character detection and recognition on the positioned dimension annotation, and comparing the dimension annotation recognition result of the view area with the dimension annotation information corresponding to the process drawing template with the highest similarity to determine whether there is a difference;

If it is determined that a difference occurs, data tracing is performed and the difference position is marked in the drawing file to be compared. Otherwise, a result is outputted that the drawing file to be compared is similar to the process drawing template.

Furthermore, a text line detection algorithm is used to perform the text detection.

Furthermore, a target detection algorithm based on a deep convolutional neural network is used to segment the text area and the view area from the drawing file to be compared.

Furthermore, a partial binary tree classification method is used to compare the dimension positioning result and dimension recognition result of the view area with the dimension positioning information and dimension information of the view area corresponding to the process drawing template with the highest similarity to determine whether there is a difference.

Furthermore, the drawing similarity detection is specifically to compare the drawing file to be compared with the process drawing template through an active window, calculate the similarity of the active window at the same position based on the pixel value comparison within the active window, traverse through the active window to obtain the overall similarity result of the drawing file to be compared and the process drawing template, and select the process drawing template with the highest similarity result as the most similar process drawing template.

Furthermore, the drawing similarity detection further includes:

Performing feature recognition on the drawing file to be compared to obtain a feature map of the drawing file;

Character recognition is performed on the characteristic graph of the drawing file to enhance the character pattern and enhance the character pattern of the drawing file to be compared.

Furthermore, the LBP algorithm is used to perform feature recognition on the drawing file to be compared.

Furthermore, the process of determining the window size of the LBP algorithm includes the following steps:

Set and initialize parameter k,

An active window of each pixel is set according to the parameter k, and a calculation formula for the average pixel intensity value of the active window of each pixel is set;

For each pixel, the average pixel intensity difference between two non-overlapping windows of the pixel is calculated in the horizontal and vertical directions.

For each pixel, obtain a k value that can maximize the average pixel intensity difference between two non-overlapping windows in the horizontal direction or the average pixel intensity difference between two non-overlapping windows in the vertical direction;

According to the k value determined for each pixel, the size of the active window is determined as the window size of the LBP algorithm.

Furthermore, the calculation expression of the pixel average intensity value is:

Where x is the horizontal coordinate of the pixel, y is the vertical coordinate of the pixel, _Ak (x,y) is the average intensity value of the pixel (x,y) in an active window, and g(i,j) is the pixel value at coordinate (i,j) in the image.

Furthermore, the calculation expression of the pixel average intensity difference E _k,h (x, y) between two non-overlapping windows in the horizontal direction is:

E _{k, h} (x, y) = |A _k (x+2 ^k-1 , y)-A _k (x-2 ^k-1 , y) |

The calculation expression of the pixel average intensity difference E _k,v (x, y) between two non-overlapping windows in the vertical direction is:

E _{k, v} (x, y) = |A _k (x, y+2 ^k-1 )-A _k (x, y-2 ^k-1 )|.

Furthermore, character recognition is performed on the feature map of the drawing file through an optical character recognition method to enhance the character pattern. The optical character recognition method includes preprocessing the feature map of the drawing file, loading it into a pre-trained character recognition engine for character recognition, and finally enhancing the image of the character recognition result.

Furthermore, the character recognition engine adopts the open source OCR engine Tesseract.

Furthermore, the preprocessing includes binarization processing, image sharpening processing and denoising processing.

Furthermore, the scanning includes using a scanner to scan and identify the electronic version of the drawing.

Furthermore, the drawing comparison method also includes: establishing a high-voltage cable accessory process library for storing the process drawing templates and their label information.

The present invention also provides a drawing comparison system based on a high-voltage cable accessory process library, comprising:

A high-voltage cable accessory process library module is used to scan the process drawings of high-voltage cables, and after preprocessing, use them as process drawing templates, attach label information to each process drawing, and establish a high-voltage cable accessory process library for storing the process drawing templates and their label information;

The drawing file scanning module to be compared is used to scan and pre-process the process drawings of the high-voltage cables to be compared, and then obtain the drawing files to be compared;

A similar drawing detection module is used to perform drawing similarity detection on the drawing file to be compared with each process drawing template to obtain the process drawing template with the highest similarity;

A drawing comparison module is used to segment the text area and the view area from the drawing file to be compared, locate the text area, detect and recognize the text, and compare the text recognition result of the text area with the text information corresponding to the process drawing template with the highest similarity to determine whether there is a difference;

Positioning the dimension annotation of the view area, performing character detection and recognition on the positioned dimension annotation, and comparing the dimension annotation recognition result of the view area with the dimension annotation information corresponding to the process drawing template with the highest similarity to determine whether there is a difference;

If it is determined that a difference occurs, data tracing is performed and the difference position is marked in the drawing file to be compared. Otherwise, a result is outputted that the drawing file to be compared is similar to the process drawing template.

Further, a text line detection algorithm is used to perform the text detection;

Using a target detection algorithm based on a deep convolutional neural network, segmenting the text area and the view area from the drawing file to be compared;

A partial binary tree classification method is adopted to compare the dimension positioning result and the dimension recognition result of the view area with the dimension positioning information and the dimension information of the view area corresponding to the process drawing template with the highest similarity to determine whether there is a difference.

Furthermore, the drawing similarity detection is specifically to compare the drawing file to be compared with the process drawing template through an active window, calculate the similarity of the active window at the same position based on the pixel value comparison within the active window, traverse through the active window to obtain the overall similarity result of the drawing file to be compared and the process drawing template, and select the process drawing template with the highest similarity result as the most similar process drawing template.

Furthermore, the drawing similarity detection further includes:

Performing feature recognition on the drawing file to be compared to obtain a feature map of the drawing file;

Perform character recognition on the feature map of the drawing file, enhance the character pattern, and obtain a character enhanced map of the drawing file to be compared; compare the character enhanced map with the corresponding process drawing template, make a consistency judgment, and output the drawing comparison result of the drawing file to be compared.

Further, the LBP algorithm is used to perform feature recognition on the drawing file to be compared;

The process of determining the window size of the LBP algorithm includes the following steps:

Set and initialize parameter k,

An active window of each pixel is set according to the parameter k, and a calculation formula for the average pixel intensity value of the active window of each pixel is set;

For each pixel, the average pixel intensity difference between two non-overlapping windows of the pixel is calculated in the horizontal and vertical directions.

For each pixel, obtain a k value that can maximize the average pixel intensity difference between two non-overlapping windows in the horizontal direction or the average pixel intensity difference between two non-overlapping windows in the vertical direction;

According to the k value determined for each pixel, the size of the active window is determined as the window size of the LBP algorithm; the calculation expression of the pixel average intensity value is:

Where x is the horizontal coordinate of the pixel, y is the vertical coordinate of the pixel, _Ak (x,y) is the average pixel intensity value of the pixel (x,y) in an active window, and g(i,j) is the pixel value with coordinates (i,j) in the image;

The calculation expression of the pixel average intensity difference E _k,h (x, y) between two non-overlapping windows in the horizontal direction is:

E _{k, h} (x, y) = |A _k (x+2 ^k-1 , y)-A _k (x-2 ^k-1 , y) |

The calculation expression of the pixel average intensity difference E _k,v (x, y) between two non-overlapping windows in the vertical direction is:

E _{k, v} (x, y) = |A _k (x, y+2 ^k-1 )-A _k (x, y-2 ^k-1 )|.

Further, character recognition is performed on the feature map of the drawing file by an optical character recognition method to enhance the character pattern. The optical character recognition method includes preprocessing the feature map of the drawing file, loading it into a pre-trained character recognition engine for character recognition, and finally enhancing the character recognition result;

The character recognition engine adopts the open source OCR engine Tesseract.

Furthermore, the preprocessing includes binarization processing, image sharpening processing and denoising processing.

Furthermore, the scanning includes using a scanner to scan and identify the electronic version of the drawing.

Compared with the prior art, the present invention has the following advantages:

(1) After constructing the process drawing template, the present invention first calculates the similarity of the drawing files to be compared, sorts the similarity, and obtains the process drawing template with the highest similarity; then the drawing is segmented into a text area and a view area; the text area is located, and text is detected and recognized; the view area is located for dimension annotation, and the dimension annotation is detected and recognized; the identified dimension information is compared with the information in the database in the form of data to determine the consistency of the drawing; if there is a difference, the data is traced to find the difference point in the drawing;

This solution does not require a large amount of data sets for training. It searches for the most similar process drawing template through similarity judgment. It then detects and identifies the text area and view area separately, and can specifically identify the differences, thereby improving the accuracy of the comparison.

(2) When the present invention uses the LBP algorithm to perform feature recognition on high-voltage cable accessory drawings, it is found that the existing LBP algorithm has limitations when processing complex high-voltage cable accessory drawings. Therefore, an active window selection scheme with an adaptive threshold is proposed to obtain a window size that maximizes the average pixel intensity difference between non-overlapping windows in the horizontal and vertical directions. The combination of this scheme and the LBP algorithm can reduce the error of LBP in extracting primitive features.

(3) The present invention builds a high-voltage cable accessory process library that stores process drawing templates, and provides a standardized and standardized technical protocol and a review and identification process for process drawings.

(4) Compared with traditional drawings, digital process drawings can use image noise reduction and enhancement technology to improve the clarity and display effect of drawings, thereby making the handwriting patterns clearer and better showing the effects of design drawings to staff.

(5) Compared with traditional manual review and management, digital process library management is more convenient, safe and reliable, reducing the work of production personnel manually flipping through paper materials, statistical compilation and data storage management, and truly achieving the goal of improving the management and review efficiency of process drawings and reducing costs. Through long-term use, the costs of production operation maintenance, manual management, etc. have been greatly reduced, creating better economic benefits for the company.

(6) This method can realize dynamic supervision of the entire process of high-voltage cable accessories, ensure the standardization of the process library in real time, and significantly reduce the risk costs and safety hazards brought by non-standard process drawings.

(7) Through the comprehensive review of all high-voltage cable accessories, work efficiency is greatly improved. The review dimensions are comprehensive and scientific, with technical support and logical design, making the review process simple, efficient and reliable.

BRIEF DESCRIPTION OF THE DRAWINGS

FIG1 is a schematic diagram of a drawing comparison process of a drawing comparison method based on a high-voltage cable accessories process library provided in an embodiment of the present invention;

FIG2 is a schematic diagram of a flow chart of comparing the result of dimension recognition provided in an embodiment of the present invention;

FIG. 3 is a recognition result of a drawing file to be compared provided in this embodiment.

DETAILED DESCRIPTION

In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

In the description of the present invention, it should be noted that the terms "center", "up", "down", "left", "right", "vertical", "horizontal", "inside", "outside", etc. indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings, or are the directions or positional relationships in which the inventive product is usually placed when in use. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as a limitation on the present invention.

It should be noted that the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

In addition, the terms "horizontal", "vertical" and the like do not mean that the components are required to be absolutely horizontal or suspended, but can be slightly tilted. For example, "horizontal" only means that its direction is more horizontal than "vertical", and does not mean that the structure must be completely horizontal, but can be slightly tilted.

Example 1

As shown in FIG1 , this embodiment provides a drawing comparison method based on a high-voltage cable accessory process library, comprising the following steps:

Scan the process drawings of high-voltage cables, pre-process them, and use them as process drawing templates, and attach label information to each process drawing;

After scanning and preprocessing the process drawings of the high-voltage cables to be compared, the drawing files to be compared are obtained;

Perform drawing similarity detection on the drawing file to be compared and each process drawing template to obtain the process drawing template with the highest similarity;

Segment the text area and view area from the drawing file to be compared, locate the text area, detect and recognize the text, and compare the text recognition result of the text area with the text information corresponding to the process drawing template with the highest similarity to determine whether there is a difference;

Locate the dimension annotation of the view area, perform character detection and recognition on the located dimension annotation, and compare the dimension annotation information corresponding to the process drawing template with the highest similarity based on the dimension annotation recognition result of the view area to determine whether there is a difference;

If a difference is found, data tracing is performed and the difference position is marked in the drawing file to be compared. Otherwise, the result that the drawing file to be compared is similar to the process drawing template is output.

The drawing comparison process, drawing similarity detection process and other steps are described below.

1. Drawing comparison process

This embodiment uses a text line detection algorithm to perform text detection;

The segmentation deep learning method is used to segment the text area and view area from the drawing file to be compared;

The partial binary tree classification method is used to compare the dimension positioning results and dimension recognition results of the view area with the dimension positioning information and dimension information of the view area corresponding to the process drawing template with the highest similarity to determine whether there are any differences.

Using Halcon machine vision software: HALCON is a complete set of standard machine vision algorithm packages developed by MVtec of Germany, with a widely used machine vision integrated development environment. HALCON supports Windows, Linux and MacOS X operating environments, and the entire function library can be accessed using a variety of common programming languages such as C, C++, C#, Visual basic and Delphi.

The principles of the segmentation deep learning method and the partial binary tree classification method are described below.

Segmentation deep learning method: Use a target detection algorithm based on a deep convolutional neural network to segment drawings and divide the drawing view into multiple sub-views. The target detection algorithm uses a regression-based target detection algorithm, which has a much faster target detection speed than other algorithms.

The detection algorithm used treats target detection as a regression problem. The network model receives the original pixels and directly outputs the boundary regression and category of the target. In terms of image feature extraction, the algorithm trains the global area of the image, which can better distinguish the target and background while speeding up. In terms of network prediction, end-to-end detection is used on the predicted image, and the entire image is divided into S*S grid areas. Then, multiple prior frames are established from the center of each grid. These frames are pre-set by the network. The prediction results of the network will determine whether these frames contain objects and the type of the object. Then, the center of the predicted frame is obtained and the frame height of the prior frame is calculated to obtain the position of the entire predicted frame. After obtaining the final prediction result, score sorting and non-maximum suppression screening are performed. The frames and scores of each category with scores greater than the set threshold are taken out, and non-maximum suppression is performed using the position and score of the frame to finally obtain the result.

Partial binary tree classification method: The partial binary tree method is a method derived from the partial binary tree idea. The method mainly obtains the similarity measure between the features of each array based on the characteristic parameters of the input array samples, such as array length and other features. According to the size of the similarity measure, multiple classifiers are constructed to achieve data comparison and classification.

The specific implementation process is as follows:

S1: Preprocess drawings;

S2: Perform drawing similarity detection to obtain the process drawing template with the highest similarity;

S3: Perform image segmentation to separate regions and generate new images, including text region images and view region images;

S4: Positioning the annotation size in the view area image;

S5: Use the text segmentation and recognition method to identify the annotation size;

S6: Compare the identified size array with the size array corresponding to the process drawing template, as shown in FIG2 , specifically, using a partial binary tree classification method to first determine whether the number of data is consistent, if not, there is a difference, otherwise determine whether the data size is consistent, if not, there is a difference;

S7: obtaining the position coordinates of the dimensions in advance during the dimension positioning and recognition process. If there is a difference, the coordinates are returned and a striking rectangular frame is generated to mark the difference;

S8: If the number and size of the data are consistent, then a data difference comparison is performed and the places where the differences exist are marked, as shown in FIG3 . The box in the upper right corner of FIG3 is the marked difference area.

2. Drawing Similarity Detection Process

The drawing similarity detection is specifically to compare the drawing file to be compared with the process drawing template through the active window, calculate the similarity of the active window at the same position based on the pixel value comparison within the active window, traverse the active window to obtain the overall similarity result between the drawing file to be compared and the process drawing template, and select the process drawing template with the highest similarity result as the most similar process drawing template.

Preferably, in the most similar process drawing template, the active window position whose similarity of the active window is lower than the similarity threshold is selected, and a mark frame is added to the drawing file to be compared. The mark frame is used to inform the auditor, and the auditor will use the comparison report to achieve efficient audit work.

Preferably, the drawing similarity detection further includes:

Perform feature recognition on the drawing file to be compared, and obtain a feature map of the drawing file;

Perform character recognition on the feature map of the drawing file, enhance the character pattern, and obtain the character enhanced map for drawing comparison.

Specifically, this embodiment uses image recognition technology, OCR character recognition technology and consistency detection algorithm, and uses photoelectric technology and digital processing technology to scan high-voltage cable drawings, technical agreements and other process drawings into the process library through high-definition scanning equipment, extract features and character recognition of electronic process drawings or process documents, and compare them with the process drawing templates solidified in the standard process library. Finally, the system determines the comparison results according to the algorithm to form a comparison report for subsequent review. In addition, the modification of high-voltage cable accessory process drawings or documents from different manufacturers is analyzed, and the system prompts the deletion, addition and modification of the drawings, which simplifies the process of design reviewers comparing and reviewing drawings, reduces errors and omissions, and improves work efficiency.

The following is a detailed description of the new steps for drawing similarity detection

2.1. Feature recognition based on image recognition technology

The image recognition technology in the digital process library of high-voltage cable accessories mainly involves feature extraction and edge detection algorithms such as LBP operator and HOG operator. In this embodiment, the entire image recognition process includes image preprocessing (image denoising, image enhancement), image restoration (reconstruction of image, restoration of image), image coding and compression, image segmentation (dividing regions with different features) and final recognition.

Among them, the LBP operator was proposed by Ojala et al. in 1996. It is a classic operator for feature description and is widely used in the field of image analysis. This operator can not only capture rich detail information, but also compress redundant information. However, when the radius of this LBP operator is too large, the sensitivity to noise will be enhanced. The gradient histogram HOG was proposed by French researcher Dalal. The main purpose of the HOG algorithm is to calculate the gradient of the grayscale and normalized image, count the gradient information of the image, divide the image into small cell units to form the unique HOG features of each drawing, so as to realize the comparison of subsequent drawings.

Based on the complex characteristics of high-voltage cable process drawings, such as contrast, color, and density distribution, the existing LBP algorithm is studied in depth to obtain better feature extraction and classification results. In view of the shortcomings of the original algorithm, the global and local pixel grayscale mean difference is used to determine the size of the adaptive threshold, making it highly adaptive to drawing recognition, and an adaptive threshold LBP algorithm is proposed.

The adaptive mode LBP algorithm used in this embodiment combines the window size with the basic LBP algorithm, and has the performance of adaptive analysis characteristics. The window size is determined by the average intensity difference in the horizontal and vertical directions.

Assuming the image is g(x, y), calculate the average pixel intensity value in the active window of size (2*2 ^k )×(2*2 ^k ):

Where x is the horizontal coordinate of the pixel, y is the vertical coordinate of the pixel, _Ak (x,y) is the average pixel intensity value of the pixel (x,y) in an active window, and g(i,j) is the pixel value with coordinates (i,j) in the image;

For each pixel, calculate the average pixel intensity difference between non-overlapping windows in the horizontal and vertical directions:

Where E _k,h (x, y) is the average pixel intensity difference between two non-overlapping windows in the horizontal direction, and E _k,v (x, y) is the average pixel intensity difference between two non-overlapping windows in the vertical direction;

For each pixel, the k value that maximizes the E _k,h (x, y) or E _k,v (x, y) value (regardless of direction) is used to set the best size: S _best (x, y) = (2*2 ^k ) × (2*2 ^k );

From the above, we can see that S _best (x, y) is the approximate size of the feature primitive of the pixel with (x, y) as the coordinate. The combination of this size and the LBP algorithm reduces the error of LBP in primitive feature extraction.

2.2 OCR Character Recognition

The character recognition process of an image is a complete set of procedures, which includes image analysis, preprocessing, character recognition and recognition correction, etc. Each step is related to the accuracy of the final recognition result. For example, the clearer the image to be recognized (that is, the better the preprocessing is), the better the recognition effect is. Character recognition is the most important part of the character recognition process of an image. At present, the most commonly used and most mature character recognition technology is optical character recognition (OCR). OCR is a technology that uses optical methods to convert the text in paper documents into black and white dot matrix image files for printed characters, and converts the text in the image into text format through recognition software for further editing and processing by word processing software.

The OCR recognition process is mainly divided into four parts:

1) Image preprocessing: The main function of this module is to preprocess the sample images by unifying the size, segmenting, graying and binarizing them, in preparation for the subsequent character recognition.

2) Training the character library. Use Tesseract to conduct targeted training on the characters in the sample images to improve recognition accuracy.

3) Character recognition. Use the open source OCR engine Tesseract to perform character recognition on the image. To implement character recognition on an image in the system, you only need to call the image_to_string method in the pytesseract library. The details are shown in the following formula:

text=Pytesseract.image_to_string(img, lang=LANG, config='--psm7--oem3')

Among them, text is the character content returned after recognition; LANG is the self-trained character library or Tesseract's own language package; img is the preprocessed image.

4) Recognition correction. Correct the rejected or misrecognized characters in the image. For grayscale images, grayscale adjustment can be performed, that is, contrast enhancement. Taking one of the grayscale images as an example, the experiment found that it was rejected before enhancement, but correctly recognized after enhancement.

3. Other steps

3.1 Preprocessing and scanning process

Preprocessing includes binarization, image sharpening and denoising.

Scanning includes using a scanner to scan and recognize the drawings as electronic versions.

3.2. High-voltage cable accessories process library

The high-voltage cable accessories process library module uses a scanner to convert all existing standard cable technical protocols, process drawings and related documents into high-definition electronic process drawings through image noise reduction and enhancement technology, and enters them into the high-voltage cable accessories process library for storage and solidification of process library templates. Each process drawing contains important information such as its label, type, time and version.

At the same time, this module can use key information of the process library such as number, name and time to quickly query and retrieve.

In the daily production work of cable companies, a lot of data and drawings are involved. Relevant staff can find the process plan faster based on the queried information to ensure the smooth progress of the work, thereby improving work efficiency and ensuring the progress of the process work. Based on the application of image recognition technology, the process drawings of high-voltage cable accessories are effectively retrieved and stored, and comprehensive networking and information management are realized. All archival resources are fully shared, and a general database is established to ensure that multiple drawing resources are fully managed, uniformly stored, and standardizedly applied, highlighting the application value of the digital process library of high-voltage cables.

3.3 Comparison Report

After the digital process drawings are compared, the comparison results between the drawings to be identified and the standard drawings will be identified in the report, and the auditors will be informed of the comparison report using the identification box. The auditors will use the comparison report to achieve efficient audit work. With the help of this drawing comparison review report, dynamic supervision of the entire process can be achieved, promoting the joint improvement of drawing review services and design quality. The review information can be dynamically shared with the safety supervision department to avoid safety hazards caused by invalid construction drawings.

Example 2

This embodiment provides a drawing comparison system based on a high-voltage cable accessories process library. In order to achieve the purpose of controlling the standard process of high-voltage cable accessories and ensure the safe and stable application of operating cables, control and approval are carried out from the source of signing the technical agreement, a standard high-voltage cable accessories process library is established, and the simulation installation of the first set of cable accessories process is used to realize the identification and solidification of the process. In the process of signing the technical agreement in the later stage, the compliance and consistency of its drawings are compared.

This embodiment mainly studies the establishment of a digital process library system for high-voltage cable accessories, and discovers changes and differences in drawings through image recognition, text recognition, and drawing comparison, thereby improving work efficiency to a certain extent and bringing better economic and social benefits.

Specifically, the drawing comparison system based on the high-voltage cable accessories process library includes:

The high-voltage cable accessory process library module is used to scan the process drawings of high-voltage cables, and after pre-processing, use them as process drawing templates, and attach label information to each process drawing to establish a high-voltage cable accessory process library for storing process drawing templates and their label information;

The drawing file scanning module to be compared is used to scan and pre-process the process drawings of the high-voltage cables to be compared, and then obtain the drawing files to be compared;

A similar drawing detection module is used to perform drawing similarity detection on the drawing file to be compared with each process drawing template to obtain the process drawing template with the highest similarity;

The drawing comparison module is used to segment the text area and view area from the drawing file to be compared, locate the text area, detect and recognize the text, and compare the text recognition results of the text area with the text information corresponding to the process drawing template with the highest similarity to determine whether there is a difference;

Locate the dimension annotation of the view area, perform character detection and recognition on the located dimension annotation, and compare the dimension annotation information corresponding to the process drawing template with the highest similarity based on the dimension annotation recognition result of the view area to determine whether there is a difference;

If a difference is found, data tracing is performed and the difference position is marked in the drawing file to be compared. Otherwise, the result that the drawing file to be compared is similar to the process drawing template is output.

In the high-voltage cable accessories process library module, preprocessing includes binarization, image sharpening and denoising. Scanning includes using a scanner to scan and identify the electronic version of the drawing.

The drawing comparison process, drawing similarity detection process and other steps are described below.

1. Drawing comparison process

This embodiment uses a text line detection algorithm to perform the text detection;

Using a segmentation deep learning method, segmenting the text area and the view area from the drawing file to be compared;

A partial binary tree classification method is adopted to compare the dimension positioning result and the dimension recognition result of the view area with the dimension positioning information and the dimension information of the view area corresponding to the process drawing template with the highest similarity to determine whether there is a difference.

Using Halcon machine vision software: HALCON is a complete set of standard machine vision algorithm packages developed by MVtec of Germany, with a widely used machine vision integrated development environment. HALCON supports Windows, Linux and MacOS X operating environments, and the entire function library can be accessed using a variety of common programming languages such as C, C++, C#, Visual basic and Delphi.

The principles of the segmentation deep learning method and the partial binary tree classification method are described below.

Segmentation deep learning method: Use a target detection algorithm based on a deep convolutional neural network to segment drawings and divide the drawing view into multiple sub-views. The target detection algorithm uses a regression-based target detection algorithm, which has a much faster target detection speed than other algorithms.

The detection algorithm used treats target detection as a regression problem. The network model receives the original pixels and directly outputs the boundary regression and category of the target. In terms of image feature extraction, the algorithm trains the global area of the image, which can better distinguish the target and background while speeding up. In terms of network prediction, end-to-end detection is used on the predicted image, and the entire image is divided into S*S grid areas. Then, multiple prior frames are established from the center of each grid. These frames are pre-set by the network. The prediction results of the network will determine whether these frames contain objects and the type of the object. Then, the center of the predicted frame is obtained and the frame height of the prior frame is calculated to obtain the position of the entire predicted frame. After obtaining the final prediction result, score sorting and non-maximum suppression screening are performed. The frames and scores of each category with scores greater than the set threshold are taken out, and non-maximum suppression is performed using the position and score of the frame to finally obtain the result.

Partial binary tree classification method: The partial binary tree method is a method derived from the partial binary tree idea. The method mainly obtains the similarity measure between the features of each array based on the characteristic parameters of the input array samples, such as array length and other features. According to the size of the similarity measure, multiple classifiers are constructed to achieve data comparison and classification.

The specific implementation process is as follows:

S1: Preprocess drawings;

S2: Perform drawing similarity detection to obtain the process drawing template with the highest similarity;

S3: Perform image segmentation to separate regions and generate new images, including text region images and view region images;

S4: Positioning the annotation size in the view area image;

S5: Use the text segmentation and recognition method to identify the annotation size;

S6: comparing the identified size array with the size array corresponding to the process drawing template, specifically, using a partial binary tree classification method to first determine whether the data quantity is consistent, if not, there is a difference, otherwise determine whether the data size is consistent, if not, there is a difference;

S7: In the process of dimension positioning and identification, the position coordinates of the dimension are obtained in advance. If there is a difference, the coordinates are returned and a striking rectangular frame is generated to mark the difference.

S8: If the number and size of the data are consistent, compare the data differences and mark the places where there are differences.

2. Drawing Similarity Detection Process

The drawing similarity detection is specifically to compare the drawing file to be compared with the process drawing template through the active window, calculate the similarity of the active window at the same position based on the pixel value comparison within the active window, traverse through the active window to obtain the overall similarity result of the drawing file to be compared and the process drawing template, and select the process drawing template with the highest similarity result as the most similar process drawing template.

Preferably, in the most similar process drawing template, the active window position whose similarity of the active window is lower than the similarity threshold is selected, and a mark frame is added to the drawing file to be compared. The mark frame is used to inform the auditor, and the auditor will use the comparison report to achieve efficient audit work.

Preferably, the drawing similarity detection further includes:

Perform feature recognition on the drawing file to be compared, and obtain a feature map of the drawing file;

Perform character recognition on the feature map of the drawing file, enhance the character pattern, and obtain the character enhanced map for drawing comparison.

Specifically, this embodiment uses image recognition technology, OCR character recognition technology and consistency detection algorithm, and uses photoelectric technology and digital processing technology to scan high-voltage cable drawings, technical agreements and other process drawings into the process library through high-definition scanning equipment, extract features and character recognition of electronic process drawings or process documents, and compare them with the process drawing templates solidified in the standard process library. Finally, the system determines the comparison results according to the algorithm to form a comparison report for subsequent review. In addition, the modification of high-voltage cable accessory process drawings or documents from different manufacturers is analyzed, and the system prompts the deletion, addition and modification of the drawings, which simplifies the process of design reviewers comparing and reviewing drawings, reduces errors and omissions, and improves work efficiency.

The following is a detailed description of the new steps for drawing similarity detection

2.1. Feature recognition based on image recognition technology

The image recognition technology in the digital process library of high-voltage cable accessories mainly involves feature extraction and edge detection algorithms such as LBP operator and HOG operator. In this embodiment, the entire image recognition process includes image preprocessing (image denoising, image enhancement), image restoration (reconstruction of image, restoration of image), image coding and compression, image segmentation (dividing regions with different features) and final recognition.

Among them, the LBP operator was proposed by Ojala et al. in 1996. It is a classic operator for feature description and is widely used in the field of image analysis. This operator can not only capture rich detail information, but also compress redundant information. However, when the radius of this LBP operator is too large, the sensitivity to noise will be enhanced. The gradient histogram HOG was proposed by French researcher Dalal. The main purpose of the HOG algorithm is to calculate the gradient of the grayscale and normalized image, count the gradient information of the image, divide the image into small cell units to form the unique HOG features of each drawing, so as to realize the comparison of subsequent drawings.

Based on the complex characteristics of high-voltage cable process drawings, such as contrast, color, and density distribution, the existing LBP algorithm is studied in depth to obtain better feature extraction and classification results. In view of the shortcomings of the original algorithm, the global and local pixel grayscale mean difference is used to determine the size of the adaptive threshold, making it highly adaptive to drawing recognition, and an adaptive threshold LBP algorithm is proposed.

The adaptive mode LBP algorithm used in this embodiment combines the window size with the basic LBP algorithm, and has the performance of adaptive analysis characteristics. The window size is determined by the average intensity difference in the horizontal and vertical directions.

Assuming the image is g(x, y), calculate the average pixel intensity value in the active window of size (2*2 ^k )×(2*2 ^k ):

Where x is the horizontal coordinate of the pixel, y is the vertical coordinate of the pixel, _Ak (x,y) is the average pixel intensity value of the pixel (x,y) in an active window, and g(i,j) is the pixel value with coordinates (i,j) in the image;

For each pixel, calculate the average pixel intensity difference between non-overlapping windows in the horizontal and vertical directions:

Where E _k,h (x, y) is the average pixel intensity difference between two non-overlapping windows in the horizontal direction, and E _k,v (x, y) is the average pixel intensity difference between two non-overlapping windows in the vertical direction;

For each pixel, the k value that can maximize the E _k,h (x, y) or E _k,v (x, y) value (regardless of direction) is used to set the best size: S _best (x, y) = (2*2 ^k ) × (2*2 ^k );

From the above, we can know that S _best (x, y) is the approximate size of the feature primitive of the pixel with (x, y) as the coordinate. The combination of this size and the LBP algorithm reduces the error of LBP in primitive feature extraction.

2.2 OCR Character Recognition

The character recognition process of an image is a complete set of processes, which includes image analysis, preprocessing, character recognition and recognition correction, etc. Each step is related to the accuracy of the final recognition result. For example, the clearer the image to be recognized (that is, the better the preprocessing is), the better the recognition effect is. Character recognition is the most important part of the character recognition process of an image. At present, the most commonly used and most mature character recognition technology is optical character recognition (OCR). OCR is a technology that uses optical methods to convert the text in paper documents into black and white dot matrix image files for printed characters, and converts the text in the image into text format through recognition software for further editing and processing by word processing software.

The OCR recognition process is mainly divided into four parts:

1) Image preprocessing: The main function of this module is to preprocess the sample images by unifying the size, segmenting, graying and binarizing them, in preparation for the subsequent character recognition.

2) Training the character library. Use Tesseract to conduct targeted training on the characters in the sample images to improve recognition accuracy.

3) Character recognition. Use the open source OCR engine Tesseract to perform character recognition on the image. To implement character recognition on an image in the system, you only need to call the image_to_string method in the pytesseract library. The details are shown in the following formula:

text=pytesseract.image_to_string(img,lang=LANG,config='--psm7--oem3′)

Among them, text is the character content returned after recognition; LANG is the self-trained character library or Tesseract's own language package; img is the preprocessed image.

4) Recognition correction. Correct the rejected or misrecognized characters in the image. For grayscale images, grayscale adjustment can be performed, that is, contrast enhancement. Taking one of the grayscale images as an example, the experiment found that it was rejected before enhancement, but correctly recognized after enhancement.

3. Other steps

3.1 Preprocessing and scanning process

Preprocessing includes binarization, image sharpening and denoising.

The scanning includes using a scanner to scan and identify the drawings as electronic versions.

3.2. High-voltage cable accessories process library

The high-voltage cable accessories process library module uses a scanner to convert all existing standard cable technical protocols, process drawings and related documents into high-definition electronic process drawings through image noise reduction and enhancement technology, and enters them into the high-voltage cable accessories process library for storage and solidification of process library templates. Each process drawing contains important information such as its label, type, time and version.

At the same time, this module can use key information of the process library such as number, name and time to quickly query and retrieve.

In the daily production work of cable companies, a lot of data and drawings are involved. Relevant staff can find the process plan faster based on the queried information to ensure the smooth progress of the work, thereby improving work efficiency and ensuring the progress of the process work. Based on the application of image recognition technology, the process drawings of high-voltage cable accessories are effectively retrieved and stored, and comprehensive networking and information management are realized. All archival resources are fully shared, and a general database is established to ensure that multiple drawing resources are fully managed, uniformly stored, and standardizedly applied, highlighting the application value of the digital process library of high-voltage cables.

3.3 Comparison Report

After the digital process drawings are compared, the comparison results between the drawings to be identified and the standard drawings will be identified in the report, and the auditors will be informed of the comparison report using the identification box. The auditors will use the comparison report to achieve efficient audit work. With the help of this drawing comparison review report, dynamic supervision of the entire process can be achieved, promoting the joint improvement of drawing review services and design quality. The review information can be dynamically shared with the safety supervision department to avoid safety hazards caused by invalid construction drawings.

The hardware structure of the above system can be in the following forms:

1. Overall structure

The digital process library system for high-voltage cable accessories includes:

one or more processors;

Memory; and

One or more programs stored in the memory, wherein the one or more programs include data processing steps for executing the above-mentioned modules.

2. Modular structure

The digital process library system for high-voltage cable accessories includes:

High-voltage cable accessories process library module, drawing file scanning module to be compared and drawing comparison module;

Each module has a separate hardware structure, that is, each module includes a memory and a processor.

The memory stores a computer program, and the processor calls the computer program to execute the data processing steps corresponding to each of the above modules.

This embodiment establishes a drawing comparison system based on the high-voltage cable accessories process library, studies the feature processing method of the high-voltage cable digital drawing, realizes the extraction and identification of the process drawing features, and uses SQL database technology to establish a high-voltage cable historical process drawing and technical protocol library, providing standardized and standardized technical protocols and process drawing review and identification procedures.

Compared with the traditional high-voltage cable accessories process data review process, the advantages of this system are:

1. Compared with traditional drawings, digital process drawings can use image noise reduction and enhancement technology to improve the clarity and display effect of drawings, so that the handwriting patterns can be clearer and the effects of design drawings can be better displayed to the staff.

2. Compared with traditional manual review and management, digital process library management is more convenient, safe and reliable, reducing the work of manual review of paper materials, statistical summary and data storage management by production personnel, truly achieving the purpose of improving the management and review efficiency of process drawings and reducing costs. Through long-term use, the costs of production operation maintenance, manual management, etc. have been greatly reduced, creating better economic benefits for the company.

3. This solution can realize dynamic supervision of the entire process of high-voltage cable accessories, ensure the standardization of the process library in real time, and significantly reduce the risk costs and safety hazards brought by non-standard process drawings.

4. The digital process library of high-voltage cable accessories has a certain self-learning function. With the continuous increase of process drawings and technical agreements entered, the process database continues to expand, and the system functions will continue to be improved and improved. By then, it will be possible to promote and enhance the storage and review management of relevant data files of high-voltage cable professionals, and the application prospects are good.

5. Through the comprehensive review of all high-voltage cable accessories, the work efficiency is greatly improved. The review dimensions are comprehensive and scientific, with technical support and logical design, making the review process simple, efficient and reliable.

6. When using the LBP algorithm to perform feature recognition on high-voltage cable accessory drawings, it was found that the existing LBP algorithm has limitations when processing complex high-voltage cable accessory drawings;

Therefore, an active window selection scheme with adaptive threshold is proposed to obtain the window size that maximizes the average pixel intensity difference between non-overlapping windows in the horizontal and vertical directions. Combining it with the LBP algorithm can reduce the error of LBP in primitive feature extraction.

The preferred specific embodiments of the present invention are described in detail above. It should be understood that a person skilled in the art can make many modifications and changes based on the concept of the present invention without creative work. Therefore, any technical solution that can be obtained by a person skilled in the art through logical analysis, reasoning or limited experiments based on the concept of the present invention on the basis of the prior art should be within the scope of protection determined by the claims.

Claims (15)

1. The drawing comparison method based on the high-voltage cable accessory process library is characterized by comprising the following steps of:

Scanning process drawings of the high-voltage cable, preprocessing, taking the process drawings as process drawing templates, and attaching label information to each process drawing;

Scanning and preprocessing a process drawing of the high-voltage cable to be compared, and obtaining a drawing file to be compared;

Carrying out drawing similarity detection on the drawing file to be compared and each process drawing template to obtain a process drawing template with highest similarity;

dividing a text region and a view region from the drawing file to be compared, positioning, detecting and identifying the text region, comparing the text region with text information corresponding to the process drawing template with the highest similarity according to the text identification result of the text region, and judging whether a difference exists or not;

Positioning the marked size of the view area, detecting and identifying characters of the positioned marked size, and comparing the marked size with the marked size information corresponding to the process drawing template with the highest similarity according to the marked size identification result of the view area to judge whether the difference exists or not;

If the difference is judged to occur, carrying out data tracing, marking the difference position in the drawing file to be compared, otherwise, outputting a result similar to the process drawing template of the drawing file to be compared;

comparing the marked size positioning result and the marked size recognition result of the view area with the marked size positioning information and the marked size information of the view area corresponding to the process drawing template with the highest similarity by adopting a biased binary tree classification method, and judging whether the difference exists or not;

The drawing similarity detection specifically comprises the steps of comparing the drawing file to be compared with a process drawing template through a movable window, calculating the similarity of the movable window at the same position according to pixel value comparison in the movable window at the same position, traversing through the movable window to obtain the similarity result of the drawing file to be compared and the whole process drawing template, and selecting the process drawing template with the highest similarity result as the most similar process drawing template;

the drawing similarity detection further comprises:

Performing feature recognition on the drawing files to be compared to obtain a drawing file feature diagram;

And carrying out character recognition on the feature diagram of the drawing file, enhancing the character pattern, and enhancing the character pattern of the drawing file to be compared.

2. The method for comparing drawings based on a high-voltage cable accessory process library according to claim 1, wherein a text line detection algorithm is adopted for text detection.

3. The drawing comparison method based on the high-voltage cable accessory process library of claim 1, wherein a target detection algorithm based on a deep convolutional neural network is adopted to divide a text region and a view region from the drawing file to be compared.

4. The drawing comparison method based on the high-voltage cable accessory process library of claim 1, wherein the feature recognition is carried out on the drawing files to be compared by adopting an LBP algorithm.

5. The drawing comparison method based on the high-voltage cable accessory process library as claimed in claim 4, wherein the window size determination process of the LBP algorithm comprises the following steps:

the parameter k is set and initialized,

Setting an active window of each pixel according to the parameter k, and setting a calculation formula of a pixel average intensity value of the active window of each pixel;

For each pixel, calculating the pixel average intensity difference between two mutually non-overlapping windows of the pixel in the horizontal direction and the vertical direction, respectively,

For each pixel, obtaining a k value which can enable the value of the pixel average intensity difference between two mutually non-overlapping windows in the horizontal direction or the value of the pixel average intensity difference between two mutually non-overlapping windows in the vertical direction to reach the maximum;

The size of the active window is determined as the window size of the LBP algorithm based on the k value determined for each pixel.

6. The drawing comparison method based on the high-voltage cable accessory process library according to claim 5, wherein the calculation expression of the pixel average intensity value is:

Where x is the abscissa of the pixel, y is the ordinate of the pixel, A _k (x, y is the average intensity value of the pixel (x, y) in an active window, g (i, j) is the pixel value of the image with coordinates (i, j).

7. The drawing comparison method based on the high-voltage cable accessory process library according to claim 6, wherein the pixel average intensity difference E _k,h between two windows which are not overlapped with each other in the horizontal direction (the calculation expression of x, y is:

E_k,h(x,y＝|A_k(x+2^k-1,y-A_k(x-2^k-1，y

the pixel average intensity difference E _k,v (x, y) between two mutually non-overlapping windows in the vertical direction is calculated as:

E_k,v(x,y＝|A_k(x,y 2^k-1)-A_k(x,y 2^k-1)|。

8. The drawing comparison method based on the high-voltage cable accessory process library according to claim 1, wherein character recognition is carried out on the drawing file feature diagram through an optical character recognition method, character patterns are enhanced, the optical character recognition method comprises the steps of preprocessing the drawing file feature diagram, loading the preprocessed drawing file feature diagram into a character recognition engine for character recognition, and finally carrying out picture enhancement on character recognition results.

9. The method of claim 8, wherein the character recognition engine is an open source OCR engine TESSERACT.

10. The method for comparing drawings based on a high-voltage cable accessory process library according to claim 1, wherein the preprocessing comprises binarization processing, picture sharpening processing and denoising processing.

11. The method of claim 1, wherein the scanning includes scanning the identified electronic version of the drawing with a scanner.

12. The drawing comparison method based on the high-voltage cable accessory process library according to claim 1, wherein the drawing comparison method further comprises: and establishing a high-voltage cable accessory process library for storing the process drawing template and the label information thereof.

13. Drawing comparison system based on high voltage cable accessory technology storehouse, characterized by comprising:

the high-voltage cable accessory process library module is used for scanning process drawings of the high-voltage cable, taking the process drawings as process drawing templates after pretreatment, attaching tag information to each process drawing, and establishing a high-voltage cable accessory process library for storing the process drawing templates and the tag information thereof;

The drawing file scanning module to be compared is used for scanning and preprocessing the process drawing of the high-voltage cable to be compared to obtain drawing files to be compared;

The similarity drawing detection module is used for detecting drawing similarity between the drawing file to be compared and each process drawing template and obtaining a process drawing template with highest similarity;

The drawing comparison module is used for dividing a text region and a view region from the drawing file to be compared, positioning, detecting and identifying the text region, comparing the text region with text information corresponding to the process drawing template with the highest similarity according to the text identification result of the text region, and judging whether a difference exists or not;

Positioning the marked size of the view area, detecting and identifying characters of the positioned marked size, and comparing the marked size with the marked size information corresponding to the process drawing template with the highest similarity according to the marked size identification result of the view area to judge whether the difference exists or not;

If the difference is judged to occur, carrying out data tracing, marking the difference position in the drawing file to be compared, otherwise, outputting a result similar to the process drawing template of the drawing file to be compared;

performing text detection by adopting a text line detection algorithm;

Dividing a text region and a view region from the drawing file to be compared by adopting a target detection algorithm based on a depth convolution neural network;

comparing the marked size positioning result and the marked size recognition result of the view area with the marked size positioning information and the marked size information of the view area corresponding to the process drawing template with the highest similarity by adopting a biased binary tree classification method, and judging whether the difference exists or not;

The drawing similarity detection specifically comprises the steps of comparing the drawing file to be compared with a process drawing template through a movable window, calculating the similarity of the movable window at the same position according to pixel value comparison in the movable window at the same position, traversing through the movable window to obtain the similarity result of the drawing file to be compared and the whole process drawing template, and selecting the process drawing template with the highest similarity result as the most similar process drawing template;

the drawing similarity detection further comprises:

Performing feature recognition on the drawing files to be compared to obtain a drawing file feature diagram;

Character recognition is carried out on the drawing file feature diagram, character patterns are enhanced, and a character enhancement diagram of the drawing file to be compared is obtained; and comparing the character enhancement map with a corresponding process drawing template, judging consistency, and outputting a drawing comparison result of the drawing file to be compared.

14. The drawing comparison system based on the high-voltage cable accessory process library of claim 13, wherein the feature recognition is performed on the drawing files to be compared by adopting an LBP algorithm;

The window size determination process of the LBP algorithm comprises the following steps:

the parameter k is set and initialized,

Setting an active window of each pixel according to the parameter k, and setting a calculation formula of a pixel average intensity value of the active window of each pixel;

For each pixel, calculating the pixel average intensity difference between two mutually non-overlapping windows of the pixel in the horizontal direction and the vertical direction, respectively,

For each pixel, obtaining a k value which can enable the value of the pixel average intensity difference between two mutually non-overlapping windows in the horizontal direction or the value of the pixel average intensity difference between two mutually non-overlapping windows in the vertical direction to reach the maximum;

determining the size of an active window according to the k value determined by each pixel, wherein the size is used as the window size of an LBP algorithm;

the calculation expression of the pixel average intensity value is as follows:

Where x is the abscissa of the pixel, y is the ordinate of the pixel, a _k (x, y is the average intensity value of the pixel (x, y) in an active window, g (i, j) is the pixel value of the coordinate (i, j) in the image;

the pixel average intensity difference E _k,h between two windows that do not overlap each other in the horizontal direction (the calculation expression of x, y is:

E_k,h(x,y＝|A_k(x+2^k-1,y-A_k(x-2^k-1，y

the pixel average intensity difference E _k,v (x, y) between two mutually non-overlapping windows in the vertical direction is calculated as:

E_k,v(x,y＝|A_k(x,y 2^k-1)-A_k(x,y 2^k-1)|。

15. The drawing comparison system based on the high-voltage cable accessory process library of claim 13, wherein character recognition is carried out on the drawing file feature diagram through an optical character recognition method, character patterns are enhanced, the optical character recognition method comprises the steps of preprocessing the drawing file feature diagram, loading the preprocessed drawing file feature diagram into a character recognition engine for character recognition, and finally carrying out picture enhancement on character recognition results;

the character recognition engine employs an open source OCR engine TESSERACT.