CN118015770A - Electronic fence hit method, terminal equipment and computer readable storage medium - Google Patents

Abstract

The application discloses an electronic fence hit method, terminal equipment and a computer readable storage medium, which are characterized in that an electronic fence monitoring picture containing a potential invader is monitored and collected in real time through a camera, an image processing and analyzing algorithm is introduced into the rear end to analyze the electronic fence monitoring picture, so that the type of a moving target object is identified, whether the moving target object is of an unauthorized type or not is judged based on the type of the moving target object, and therefore an electronic fence hit early warning prompt is automatically generated when an unauthorized target object is detected. Therefore, the accurate identification capability of the electronic fence system to an intruder can be improved, the occurrence of false alarm conditions is reduced, and the intelligent level of the electronic fence is further improved.

Description

Electronic fence hit method, terminal equipment and computer readable storage medium

Technical Field

The application relates to the field of intelligent early warning, and more particularly, to an electronic fence hit method, terminal equipment and a computer readable storage medium.

Background

An electronic fence is a security protection system that utilizes electronic devices and sensors to protect a particular area. The electronic fence can effectively prevent or warn unauthorized persons or objects from entering the protected area, thereby improving security. However, the electronic fence has some problems, for example, in the conventional intrusion detection system based on image recognition, in the process of detecting an intruder based on image recognition, the intruder may be an animal such as a cat or a dog which is wrongly intruded, and false alarms may occur, which affect the accuracy and stability of the system. This is because conventional image recognition algorithms have certain limitations in processing complex scenes and dynamic objects.

Accordingly, an optimized electronic fence hit scheme is desired.

Disclosure of Invention

The present application has been made to solve the above-mentioned technical problems. The embodiment of the application provides an electronic fence hit method, terminal equipment and a computer readable storage medium, which are used for monitoring and collecting an electronic fence monitoring picture containing a potential invader in real time through a camera, introducing an image processing and analyzing algorithm at the rear end to analyze the electronic fence monitoring picture so as to identify the type of a moving target object, judging whether the moving target object is an unauthorized type or not based on the type of the moving target object, and automatically generating an electronic fence hit early warning prompt when an unauthorized target object is detected. Therefore, the accurate identification capability of the electronic fence system to an intruder can be improved, the occurrence of false alarm conditions is reduced, and the intelligent level of the electronic fence is further improved.

According to one aspect of the present application, there is provided an electronic fence hit method, comprising:

Acquiring an electronic fence monitoring picture which is acquired by a camera and contains potential intruders;

The electronic fence monitoring picture containing the potential invader passes through a moving target detection network to obtain a moving target region-of-interest image;

extracting features of the moving target region-of-interest image through an object feature extractor based on a convolutional neural network model to obtain a moving target object feature map;

Inputting the moving target object feature map into a mixed strengthening module comprising a re-parameterization layer and a self-adaptive attention layer to obtain a multiple strengthening moving target object feature map;

Performing object feature global perception on the multiple enhanced moving object feature map to obtain a receptive field enhanced moving object feature map;

fusing the multiple enhanced moving object feature map and the receptive field enhanced moving object feature map to obtain final moving object features;

And determining a type label of the moving target object based on the final characteristics of the moving target, and determining whether to generate an electronic fence hit early warning prompt.

According to another aspect of the present application, there is provided an electronic fence hit terminal device, comprising:

The monitoring picture acquisition module is used for acquiring an electronic fence monitoring picture which is acquired by the camera and contains a potential invader;

the moving target region of interest detection module is used for enabling the electronic fence monitoring picture containing the potential invader to pass through a moving target detection network so as to obtain a moving target region of interest image;

The feature extraction module is used for extracting features of the moving target region-of-interest image through an object feature extractor based on a convolutional neural network model so as to obtain a moving target object feature map;

the multi-enhancement module is used for inputting the moving target object feature map into the mixed enhancement module comprising a re-parameterization layer and a self-adaptive attention layer so as to obtain a multi-enhancement moving target object feature map;

the feature global perception module is used for carrying out object feature global perception on the multiple enhanced moving object feature images so as to obtain a receptive field enhanced moving object feature image;

The fusion module is used for fusing the multiple enhanced moving object feature images and the receptive field enhanced moving object feature images to obtain final moving object features;

and the result generation module is used for determining the type label of the moving target object based on the final characteristics of the moving target and determining whether to generate an electronic fence hit early warning prompt.

According to another aspect of the present application there is provided a computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, cause the processor to perform the electronic fence hit method as described above.

Compared with the prior art, the electronic fence hit method, the terminal equipment and the computer readable storage medium provided by the application have the advantages that the electronic fence monitoring picture containing potential intruders is monitored and collected in real time through the camera, and the image processing and analysis algorithm is introduced into the rear end to analyze the electronic fence monitoring picture, so that the type of a moving target object is identified, whether the moving target object is of an unauthorized type or not is judged based on the type of the moving target object, and therefore, an electronic fence hit early warning prompt is automatically generated when an unauthorized target object is detected. Therefore, the accurate identification capability of the electronic fence system to an intruder can be improved, the occurrence of false alarm conditions is reduced, and the intelligent level of the electronic fence is further improved.

Drawings

The above and other objects, features and advantages of the present application will become more apparent by describing embodiments of the present application in more detail with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of embodiments of the application and are incorporated in and constitute a part of this specification, illustrate the application and together with the embodiments of the application, and not constitute a limitation to the application. In the drawings, like reference numerals generally refer to like parts or steps.

FIG. 1 is a flow chart of an electronic fence hit method according to an embodiment of the present application;

FIG. 2 is a system architecture diagram of an electronic fence hit method according to an embodiment of the present application;

FIG. 3 is a flowchart of substep S4 of an electronic fence hit method according to an embodiment of the present application;

FIG. 4 is a flowchart of substep S7 of an electronic fence hit method according to an embodiment of the present application;

Fig. 5 is a block diagram of an electronic fence hit terminal device according to an embodiment of the present application.

Detailed Description

Hereinafter, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be apparent that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and it should be understood that the present application is not limited by the example embodiments described herein.

As used in the specification and in the claims, the terms "a," "an," "the," and/or "the" are not specific to a singular, but may include a plurality, unless the context clearly dictates otherwise. In general, the terms "comprises" and "comprising" merely indicate that the steps and elements are explicitly identified, and they do not constitute an exclusive list, as other steps or elements may be included in a method or apparatus.

Although the present application makes various references to certain modules in a system according to embodiments of the present application, any number of different modules may be used and run on a user terminal and/or server. The modules are merely illustrative, and different aspects of the systems and methods may use different modules.

A flowchart is used in the present application to describe the operations performed by a system according to embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed in order precisely. Rather, the various steps may be processed in reverse order or simultaneously, as desired. Also, other operations may be added to or removed from these processes.

Hereinafter, exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be apparent that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and it should be understood that the present application is not limited by the example embodiments described herein.

In the conventional intrusion detection system based on image recognition, during the process of detecting an intruder based on image recognition, the intruder may be an animal such as a cat or a dog which is wrongly intruded, and false alarm can occur, so that the accuracy and stability of the system are affected. This is because conventional image recognition algorithms have certain limitations in processing complex scenes and dynamic objects. Accordingly, an optimized electronic fence hit scheme is desired.

In the technical scheme of the application, an electronic fence hit method is provided. FIG. 1 is a flow chart of an electronic fence hit method according to an embodiment of the present application. Fig. 2 is a system architecture diagram of an electronic fence hit method according to an embodiment of the present application. As shown in fig. 1 and 2, the electronic fence hit method according to the embodiment of the present application includes the steps of: s1, acquiring an electronic fence monitoring picture which is acquired by a camera and contains a potential invader; s2, passing the electronic fence monitoring picture containing the potential invader through a moving target detection network to obtain a moving target region-of-interest image; s3, extracting features of the moving target region-of-interest image through an object feature extractor based on a convolutional neural network model to obtain a moving target object feature map; s4, inputting the moving target object feature map into a mixed strengthening module comprising a re-parameterization layer and a self-adaptive attention layer to obtain a multiple strengthening moving target object feature map; s5, performing object feature global perception on the multiple enhanced moving object feature map to obtain a receptive field enhanced moving object feature map; s6, fusing the multiple enhanced moving object feature images and the receptive field enhanced moving object feature images to obtain final moving object features; and S7, determining a type tag of the moving target object based on the final characteristics of the moving target, and determining whether to generate an electronic fence hit early warning prompt.

In particular, the S1 and the S2 acquire an electronic fence monitoring picture which is acquired by a camera and contains a potential intruder; and the electronic fence monitoring picture containing the potential intruder passes through a moving target detection network to obtain a moving target region-of-interest image. It should be appreciated that in the electronic fence monitoring screen, there may be a plurality of moving objects, including intruders and other non-intruders, such as animals, trees, etc. In order to accurately identify an intruder, it is necessary to first distinguish moving objects in a picture. In particular, since the moving object detection network is a deep learning model specifically for detecting moving objects in images or videos, it can separate moving objects from static backgrounds and other non-moving objects by learning the moving patterns and appearance characteristics of the moving objects. Based on the above, in the technical scheme of the application, the electronic fence monitoring picture containing the potential intruder passes through a moving target detection network to obtain a moving target region-of-interest image. By inputting the electronic fence monitoring picture into the moving target detection network, a moving target region-of-interest image can be obtained, wherein the moving target region-of-interest image comprises a region considered as a moving target by the network, so that the region possibly having an intruder can be more focused, and the accuracy of intrusion detection can be improved.

The object detection network is a deep learning model for detecting and identifying objects in images or videos. Its main objective is to accurately locate and identify objects of different categories in a given input image. The object detection network typically employs a Convolutional Neural Network (CNN) as an infrastructure to learn the feature representation from the input image. Through supervised learning on the training set, the target detection network can learn the visual characteristics and the context information of the object, so that the object can be accurately detected and classified. The object detection network has wide application in the field of computer vision, including object recognition, pedestrian detection, vehicle detection, face recognition, and the like.

In particular, the step S3 is to perform feature extraction on the moving target region of interest image through an object feature extractor based on a convolutional neural network model so as to obtain a moving target object feature map. The method comprises the steps of performing feature mining on the moving target region-of-interest image through an object feature extractor based on a convolutional neural network model to extract abstract feature representation of the moving target in the moving target region-of-interest image, so as to obtain a moving target object feature map. Therefore, key characteristic information of the moving target can be captured, subsequent analysis and recognition are facilitated, and recognition accuracy and robustness of the moving target are improved. Specifically, each layer based on the convolutional neural network model is used for respectively carrying out input data in forward transfer of the layer: carrying out convolution processing on input data to obtain a convolution characteristic diagram; pooling the convolution feature images based on the local feature matrix to obtain pooled feature images; performing nonlinear activation on the pooled feature map to obtain an activated feature map; the output of the last layer based on the convolutional neural network model is the moving target object feature map, and the input of the first layer based on the convolutional neural network model is the moving target region-of-interest image.

Convolutional neural networks (Convolutional Neural Network, CNN) are a deep learning model that is specifically used to process data with a grid structure, such as images and speech. The core idea of CNN is to extract the features of the input data by convolution operation and to perform high-level representation and abstraction of the features by layer-by-layer stacking. The following are the basic components and working principles of CNN: convolution layer: the convolutional layer is the core component of the CNN for extracting features of the input data. It performs a convolution operation on the input data by applying a set of learnable convolution kernels (filters). The convolution operation may capture local patterns and features in the input data and generate a series of feature maps; activation function: after the convolutional layer, a nonlinear activation function, such as ReLU, is typically applied. The activation function introduces nonlinear features that enable the network to learn more complex patterns and representations; pooling layer: the pooling layer is used to reduce the size and number of parameters of the feature map and extract the most important features. Common pooling operations include maximum pooling and average pooling; full tie layer: after passing through a series of convolution and pooling layers, some fully connected layers are typically added. The fully connected layer converts the feature mapping of the previous layer into an output result, such as classification or regression; dropout: to prevent overfitting, dropout techniques are often used in CNNs. Dropout discards a part of neurons randomly in the training process so as to reduce the dependency relationship among the neurons and improve the generalization capability of the model. Through a back propagation algorithm, the CNN can automatically learn and extract the characteristics in the input data and optimize according to the training target. During training, the CNN adjusts the network parameters by minimizing the loss function so that the output results are as close as possible to the real labels.

In particular, the step S4 is to input the motion target object feature map into a hybrid enhancement module including a re-parameterization layer and an adaptive attention layer to obtain a multiple enhancement motion target object feature map. In particular, in one specific example of the present application, as shown in fig. 3, the S4 includes: s41, enabling the moving target object feature map to pass through a re-parameterization layer of the mixed reinforcement module to obtain a re-parameterized moving target object feature map; s42, the moving target object feature map passes through the self-adaptive attention layer of the hybrid reinforcement module to obtain a self-adaptive reinforcement moving target object feature map; s43, fusing the re-parameterized moving object feature map and the self-adaptive reinforcement moving object feature map to obtain the multi-reinforcement moving object feature map.

Specifically, the step S41 is to pass the moving object feature map through a re-parameterization layer of the hybrid reinforcement module to obtain a re-parameterized moving object feature map. In order to enhance and enrich the feature expression capability of the moving target object so as to make the moving target object more distinguishable and robust, in the technical scheme of the application, a re-parameterization layer is further used for carrying out strengthening treatment on the moving target object feature map so as to obtain the re-parameterized moving target object feature map. By processing the re-parameterization layer, randomness can be introduced, and the original feature map is re-parameterized into richer feature representation, so that the expression capability of the feature map of the moving target object is enhanced. In this process, the mean and variance of the moving object feature map are extracted and used to generate a new feature map. The re-parameterized form can be regarded as a mode for enhancing data in a semantic feature space, which is helpful for improving the perception and recognition capability of the classifier on different types of target objects, so that the classifier is better suitable for the characteristics of different types of moving target objects, and the classification and early warning accuracy is improved. More specifically, passing the moving target object feature map through a re-parameterization layer of the hybrid reinforcement module to obtain a re-parameterized moving target object feature map, including: performing feature expression reinforcement on each feature matrix of the moving target object feature map along the channel dimension by using a re-parameterization layer of the mixed reinforcement module according to the following association reinforcement formula so as to obtain the re-parameterized moving target object feature map; wherein, the association strengthening formula is: ; wherein, For the global average value of each feature matrix along the channel dimension of the moving target object feature graph,/>Variance of each feature matrix along channel dimension for the moving object feature map,/>Is obtained by randomly sampling the Gaussian distribution of each feature matrix of the feature map of the moving target object along the channel dimensionPersonal value,/>Is the eigenvalue of each position in each eigenvector of the re-parameterized moving object eigenvector graph along the channel dimension.

Specifically, the step S42 is to pass the moving object feature map through the adaptive attention layer of the hybrid reinforcement module to obtain an adaptive reinforcement moving object feature map. Considering that each channel of the moving object feature map represents a different feature, some of these features have an important role in the type detection of the moving object, and some are irrelevant disturbance features. Therefore, in order to concentrate attention on important channel characteristics and weaken the influence of irrelevant characteristics on detection of a moving target object, so that the identification capability of a classifier on key characteristics of the moving target object is improved. Specifically, in the technical scheme of the application, the moving target object feature map is further subjected to an adaptive attention layer to obtain an adaptive strengthening moving target object feature map. It should be appreciated that the adaptive attention layer converts the feature map of each channel into a weight value by using a meta-weight generator. The weight values are multiplied by the moving object feature map channel by channel, so that each channel in the feature map can be focused to different degrees, and important channel feature information is highlighted. In this way, attention can be focused on important features of the moving object in the moving object region-of-interest image, and influence of irrelevant features is weakened, so that the recognition capability and classification capability of the classifier are improved. More specifically, passing the moving target object feature map through an adaptive attention layer of the hybrid reinforcement module to obtain an adaptive reinforcement moving target object feature map, including: processing the moving target object feature map through the self-adaptive attention layer of the hybrid reinforcement module by using the following self-adaptive reinforcement formula to obtain the self-adaptive reinforcement moving target object feature map; wherein, the self-adaptive strengthening formula is: ; wherein/> Is the characteristic diagram of the moving target object,/>Representing global mean pooling of feature matrices along a channel dimension in a feature map,/>Is the channel feature vector of the moving target object feature map,/>And/>Is the weight and bias of the convolutional layer,/>To activate the function,/>Is a convolution eigenvector of the channel eigenvector,/>Is the eigenvalue of each position in the convolution eigenvector,/>Is a weight vector,/>Is multiplied by the position point,/>Is the self-adaptive intensified motion target object characteristic diagram.

Specifically, the step S43 fuses the re-parameterized moving object feature map and the adaptive reinforcement moving object feature map to obtain the multi-reinforcement moving object feature map. It should be appreciated that the multiple enhanced moving object feature map obtained by fusing the re-parameterized moving object feature map and the adaptive enhanced moving object feature map may capture more details and associated information of moving object objects, improve adaptability to complex scenes and dynamic objects, better distinguish between different types of moving objects, reduce sensitivity to noise and interference, and provide more accurate and robust feature representation for subsequent processing and recognition tasks.

It should be noted that, in other specific examples of the present application, the moving object feature map may be input into a hybrid enhancement module including a re-parameterization layer and an adaptive attention layer in other manners to obtain a multiple enhancement moving object feature map, for example: inputting the moving target object feature map; inputting the input feature map into a re-parameterization layer; inputting the re-parameterized feature map into an adaptive attention layer; and multiplying the output of the self-adaptive attention layer by the output of the re-parameterization layer to obtain the multi-enhancement moving target object characteristic diagram.

Particularly, the step S5 is to perform global object feature perception on the multiple enhanced moving object feature map to obtain a receptive field enhanced moving object feature map. In the technical scheme of the application, the multiple enhanced moving target object feature map is further input into an object feature global perceptron based on a non-local neural network model to obtain a receptive field enhanced moving target object feature map, so as to enhance the receptive field of the moving target object feature map and improve the global context understanding capability of the moving target object. In this way, long-range dependencies between different locations in the input feature map can be captured, thereby enhancing the global perception and context awareness capabilities of the features. The method is favorable for improving the understanding and recognition accuracy of the targets, and particularly under the condition of complex scenes and shielding, different types of moving target objects can be better distinguished, so that the recognition accuracy and the robustness of the moving targets are further improved.

The Non-local neural network (Non-local Neural Network) is a neural network model for modeling long-range dependencies. It enables the network to establish global associations between different locations of the input by introducing non-local operations. The non-local neural network is capable of modeling the relationship between any two locations in the input by introducing non-local operations. This operation obtains global dependencies by computing similarities between different locations in the input feature map. Specifically, the non-local operation calculates the similarity with other locations for each location in the input feature map, and derives a weighted sum based on the similarity calculation. In this way, each location can obtain information from other locations, thereby establishing a global association. The advantage of a non-local neural network is that it is able to capture long range dependencies in the input, thereby improving the characterization capabilities of the network. It achieves good results in tasks such as image segmentation, video motion recognition, and video generation.

In particular, the step S6 is to fuse the multiple enhanced moving object feature map and the receptive field enhanced moving object feature map to obtain a final moving object feature. It should be understood that the multiple enhanced moving object feature map and the receptive field enhanced moving object feature map have different characteristics receptive fields and expression capabilities due to different enhancement and adjustment operations, respectively. Therefore, in order to comprehensively utilize the advantages of the multiple enhanced moving object feature images and the receptive field enhanced moving object feature images to obtain more comprehensive and rich moving object feature representations so as to improve the recognition and understanding capability of the moving objects, the multiple enhanced moving object feature images and the receptive field enhanced moving object feature images are further fused in the technical scheme of the application to obtain a moving object final feature image. In particular, in the technical solution of the present application, the multiple enhanced moving object feature map expresses a hybrid enhanced local image semantic feature of the moving object region of interest image, and after the multiple enhanced moving object feature map is input to an object feature global perceptron based on a non-local neural network model, the receptive field enhanced moving object feature map may further express a global image semantic feature, so when the multiple enhanced moving object feature map and the receptive field enhanced moving object feature map are fused, the applicant optimizes the multiple enhanced moving object feature map and the receptive field enhanced moving object feature map, specifically expressed as: ; wherein/> Is a first eigenvector obtained by expanding the characteristic diagram of the multiple enhanced moving target object, and/>Is a second eigenvector obtained by expanding the receptive field enhanced moving target object eigenvector diagram,/>And/>Respectively representing squares of 1-norm and 2-norm of the feature vector, the first feature vector/>And a second eigenvector/>Having the same eigenvector length/>And/>Is a weight superparameter,/>A base 2 logarithmic operation is shown. Here, in order to promote the consistency of the distribution information representation of the multiple strengthening moving object feature images and the receptive field strengthening moving object feature images in the feature fusion scene, the feature scale and the structural representation of the feature vectors to be fused are used for predefining the absolute coordinates of the distribution regression as the reference of feature value cross geometric registration, so that the consistency of the rigid grid of the information distribution can be maintained, and the concept of probability chamfering loss is utilized for punishing the misalignment and the incomplete overlapping based on the distance between the feature distribution information representations, so that the feature fusion of the multiple strengthening moving object feature images and the receptive field strengthening moving object feature images is realized. Thus, will be defined by/>The formed fusion feature vector is used as the feature vector which is unfolded when the final feature map of the moving object is classified by the classifier, so that the expression effect of the feature vector to be classified can be improved, and the accuracy of the classification result output by the classifier is improved. Therefore, the type of the moving target object can be accurately identified through analysis of the electronic fence monitoring picture, and whether the moving target object is of an unauthorized type or not is judged, so that an electronic fence hit early warning prompt is automatically generated when an unauthorized target object is detected, the accurate identification capability of an electronic fence system to an intruder is improved, and the occurrence of false alarm conditions is reduced.

Specifically, the step S7 is to determine a type tag of the moving target object based on the final feature of the moving target, and determine whether to generate an electronic fence hit early warning prompt. In particular, in one specific example of the present application, as shown in fig. 4, the S7 includes: s71, enabling the final feature map of the moving target to pass through a moving target object identifier based on a classifier to obtain an identification result, wherein the identification result is used for representing a type label of the moving target object; and S72, responding to the type label of the moving target object as an unauthorized type label, and generating an electronic fence hit early warning prompt.

Specifically, the step S71 is to pass the final feature map of the moving object through a moving object identifier based on a classifier to obtain a recognition result, where the recognition result is used to represent a type tag of the moving object. That is, specifically, in the technical solution of the present application, the tag of the classifier is a type tag of the moving target object, so after the classification result is obtained, the type of the moving target object may be identified based on the classification result, and whether the moving target object is an unauthorized type may be determined based on the type of the moving target object. More specifically, passing the moving object final feature map through a moving object identifier based on a classifier to obtain an identification result, including: expanding the final feature map of the moving object into classification feature vectors based on row vectors or column vectors; performing full-connection coding on the classification feature vectors by using a plurality of full-connection layers of the classifier to obtain coded classification feature vectors; and passing the coding classification feature vector through a Softmax classification function of the classifier to obtain the classification result.

Specifically, in S72, in response to the type tag of the moving target object being an unauthorized type tag, an electronic fence hit warning prompt is generated. In one specific example, in response to the classification result being an unauthorized type tag, an electronic fence hit pre-warning cue is generated. Therefore, the accurate identification capability of the electronic fence system to an intruder can be improved, the occurrence of false alarm conditions is reduced, and the intelligent level of the electronic fence is further improved.

It should be noted that, in other specific examples of the present application, the type tag of the moving target object may be determined based on the final feature of the moving target in other manners, and whether to generate an electronic fence hit early warning prompt is determined, for example: extracting a useful feature representation from the final features of the moving object; training a classifier model using the extracted features and corresponding tag data; and predicting the type label of the new moving target object by using the trained classifier model. Inputting the extracted features into a classifier model, and outputting predicted moving target type labels by the model; determining the boundary and the area of the electronic fence; matching the position coordinates of the moving target object with the electronic fence; and determining whether to generate an electronic fence hit early warning prompt according to the matching result. And if the moving target object is in the electronic fence, generating a corresponding early warning prompt.

In summary, the electronic fence hit method according to the embodiment of the application is explained, which monitors and collects an electronic fence monitoring picture including a potential intruder in real time through a camera, and introduces an image processing and analyzing algorithm at the rear end to analyze the electronic fence monitoring picture, so as to identify the type of a moving target object, and judges whether the moving target object is an unauthorized type based on the type of the moving target object, thereby automatically generating an electronic fence hit early warning prompt when an unauthorized target object is detected. Therefore, the accurate identification capability of the electronic fence system to an intruder can be improved, the occurrence of false alarm conditions is reduced, and the intelligent level of the electronic fence is further improved.

Further, an electronic fence hit terminal device is provided.

Fig. 5 is a block diagram of an electronic fence hit terminal device according to an embodiment of the present application. As shown in fig. 5, the electronic fence hit terminal device 300 according to the embodiment of the present application includes: a monitoring screen acquisition module 310, configured to acquire an electronic fence monitoring screen including a potential intruder acquired by a camera; the moving target region of interest detection module 320 is configured to obtain a moving target region of interest image by passing the electronic fence monitoring screen including the potential intruder through a moving target detection network; the feature extraction module 330 is configured to perform feature extraction on the moving target region of interest image by using an object feature extractor based on a convolutional neural network model to obtain a moving target object feature map; the multiple strengthening module 340 is configured to input the moving object feature map into a hybrid strengthening module including a re-parameterization layer and an adaptive attention layer to obtain a multiple strengthening moving object feature map; the feature global perception module 350 is configured to perform object feature global perception on the multiple enhanced moving object feature map to obtain a receptive field enhanced moving object feature map; a fusion module 360, configured to fuse the multiple enhanced moving object feature map and the receptive field enhanced moving object feature map to obtain a final moving object feature; the result generating module 370 is configured to determine a type tag of the moving target object based on the final feature of the moving target, and determine whether to generate an electronic fence hit pre-warning prompt.

As described above, the electronic fence-hit terminal device 300 according to the embodiment of the present application can be implemented in various wireless terminals, such as a server or the like having an electronic fence-hit algorithm. In one possible implementation, the electronic fence-hit terminal device 300 according to an embodiment of the present application can be integrated into a wireless terminal as one software module and/or hardware module. For example, the electronic fence-hit terminal device 300 may be a software module in the operating system of the wireless terminal, or may be an application developed for the wireless terminal; of course, the electronic fence-hit terminal device 300 could equally be one of many hardware modules of the wireless terminal.

Alternatively, in another example, the electronic fence-hit terminal device 300 and the wireless terminal may be separate devices, and the electronic fence-hit terminal device 300 may be connected to the wireless terminal through a wired and/or wireless network and transmit interaction information in a agreed data format.

Furthermore, embodiments of the present application may also be a computer-readable storage medium, having stored thereon computer program instructions, which when executed by a processor, cause the processor to perform steps in the functions of the electronic fence hit method according to various embodiments of the present application described in the "exemplary method" section of the present specification.

The computer readable storage medium may employ any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may include, for example, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium would include the following: an electrical connection having one or more wires, a portable disk, a hard disk, random Access Memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

The foregoing description of the embodiments of the present disclosure has been presented for purposes of illustration and description, and is not intended to be exhaustive or limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the various embodiments described. The terminology used herein was chosen in order to best explain the principles of the embodiments, the practical application, or the improvement of technology in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

Claims (10)

1. An electronic fence hit method, comprising:

Acquiring an electronic fence monitoring picture which is acquired by a camera and contains potential intruders;

The electronic fence monitoring picture containing the potential invader passes through a moving target detection network to obtain a moving target region-of-interest image;

extracting features of the moving target region-of-interest image through an object feature extractor based on a convolutional neural network model to obtain a moving target object feature map;

Inputting the moving target object feature map into a mixed strengthening module comprising a re-parameterization layer and a self-adaptive attention layer to obtain a multiple strengthening moving target object feature map;

Performing object feature global perception on the multiple enhanced moving object feature map to obtain a receptive field enhanced moving object feature map;

fusing the multiple enhanced moving object feature map and the receptive field enhanced moving object feature map to obtain final moving object features;

And determining a type label of the moving target object based on the final characteristics of the moving target, and determining whether to generate an electronic fence hit early warning prompt.

2. The method of claim 1, wherein inputting the moving object feature map into a hybrid enhancement module comprising a re-parameterization layer and an adaptive attention layer to obtain a multi-enhancement moving object feature map, comprises:

passing the moving target object feature map through a re-parameterization layer of the hybrid reinforcement module to obtain a re-parameterized moving target object feature map;

The moving target object feature map passes through the self-adaptive attention layer of the hybrid reinforcement module to obtain a self-adaptive reinforced moving target object feature map;

And fusing the re-parameterized moving object feature map and the self-adaptive strengthening moving object feature map to obtain the multi-strengthening moving object feature map.

3. The method of claim 2, wherein passing the moving object feature map through a re-parameterization layer of the hybrid reinforcement module to obtain a re-parameterized moving object feature map comprises: performing feature expression reinforcement on each feature matrix of the moving target object feature map along the channel dimension by using a re-parameterization layer of the mixed reinforcement module according to the following association reinforcement formula so as to obtain the re-parameterized moving target object feature map;

Wherein, the association strengthening formula is: ; wherein/> For the global average value of each feature matrix along the channel dimension of the moving target object feature graph,/>Variance of each feature matrix along channel dimension for the moving object feature map,/>Is obtained by randomly sampling the Gaussian distribution of each feature matrix of the feature map of the moving target object along the channel dimensionPersonal value,/>Is the eigenvalue of each position in each eigenvector of the re-parameterized moving object eigenvector graph along the channel dimension.

4. The method of claim 3, wherein passing the moving object feature map through an adaptive attention layer of the hybrid reinforcement module to obtain an adaptive reinforcement moving object feature map, comprises: processing the moving target object feature map through the self-adaptive attention layer of the hybrid reinforcement module by using the following self-adaptive reinforcement formula to obtain the self-adaptive reinforcement moving target object feature map;

wherein, the self-adaptive strengthening formula is: ; wherein/> Is the characteristic diagram of the moving target object,/>Representing global mean pooling of feature matrices along a channel dimension in a feature map,/>Is the channel feature vector of the moving target object feature map,/>And/>Is the weight and offset of the convolutional layer,To activate the function,/>Is a convolution eigenvector of the channel eigenvector,/>Is the eigenvalue of each position in the convolution eigenvector,/>Is a weight vector,/>Is multiplied by the position point,/>Is the self-adaptive intensified motion target object characteristic diagram.

5. The method of claim 4, wherein performing global object feature perception on the multiple enhanced moving object feature map to obtain a receptive field enhanced moving object feature map comprises: and inputting the multiple enhanced moving target object feature map into an object feature global sensor based on a non-local neural network model to obtain the receptive field enhanced moving target object feature map.

6. The method of claim 5, wherein fusing the multiple enhanced moving object feature map and the receptive field enhanced moving object feature map to obtain a moving object final feature comprises: and carrying out fusion optimization on the multiple enhanced moving object feature map and the receptive field enhanced moving object feature map to obtain a moving object final feature map serving as the moving object final feature.

7. The method of claim 6, wherein determining a type tag of a moving target object based on the moving target final feature and determining whether to generate an electronic fence hit pre-warning cue comprises:

The final feature map of the moving target passes through a moving target object identifier based on a classifier to obtain an identification result, wherein the identification result is used for representing a type label of the moving target object;

and responding to the type label of the moving target object as an unauthorized type label, and generating an electronic fence hit early warning prompt.

8. The electronic fence hit method of claim 7, wherein passing the moving target final feature map through a classifier-based moving target object identifier to obtain a recognition result, the recognition result being used to represent a type tag of a moving target object, includes:

expanding the final feature map of the moving object into classification feature vectors based on row vectors or column vectors;

performing full-connection coding on the classification feature vectors by using a plurality of full-connection layers of the classifier to obtain coded classification feature vectors; and

And the coding classification feature vector is passed through a Softmax classification function of the classifier to obtain the classification result.

9. An electronic fence hit terminal device, comprising:

The monitoring picture acquisition module is used for acquiring an electronic fence monitoring picture which is acquired by the camera and contains a potential invader;

the moving target region of interest detection module is used for enabling the electronic fence monitoring picture containing the potential invader to pass through a moving target detection network so as to obtain a moving target region of interest image;

The feature extraction module is used for extracting features of the moving target region-of-interest image through an object feature extractor based on a convolutional neural network model so as to obtain a moving target object feature map;

the multi-enhancement module is used for inputting the moving target object feature map into the mixed enhancement module comprising a re-parameterization layer and a self-adaptive attention layer so as to obtain a multi-enhancement moving target object feature map;

the feature global perception module is used for carrying out object feature global perception on the multiple enhanced moving object feature images so as to obtain a receptive field enhanced moving object feature image;

The fusion module is used for fusing the multiple enhanced moving object feature images and the receptive field enhanced moving object feature images to obtain final moving object features;

and the result generation module is used for determining the type label of the moving target object based on the final characteristics of the moving target and determining whether to generate an electronic fence hit early warning prompt.

10. A computer readable storage medium having stored thereon computer program instructions which, when executed by a processor, cause the processor to perform the method of any of claims 1 to 8.