CN1207683C - Ordered picture splicing method in seed grain-space detection - Google Patents

Abstract

The present invention relates to a method for detecting grain spaces of seeds, particularly to a method for splitting sequential images in the detection of grain spaces of seeds. The method for splitting sequential images in the detection of grain spaces of seeds is characterized in that the method comprises the following steps that (1), a plurality of marks with different sizes and the same interval are made on a recording adhesive tape; (2), a first complete mark of the overlapping region of two adjacent images of the sequential images acquired by a computer is used as a reference so as to find the form center of the first complete mark; (3), pixels of the former image are taken up to the form center of the first complete mark of the image overlapping region; (4), pixels of the latter image are taken from the form center of the same mark of the image overlapping region; (5), in the two adjacent images, grain spaces of adjacent seeds which are not in the overlapping region are the sum of horizontal distance between seeds in the first image and the center form of the image mark, and horizontal distance between the form center of the image mark in the second image and seeds in the image.

Description

Sequence Image Mosaic Method in Seed Distance Detection

technical field

The invention relates to a method for detecting the distance between seeds, in particular to a splicing method for sequence images in the detection of the distance between seeds.

Background technique

The detection technology based on image processing and analysis is a comprehensive detection method integrating computer technology, image technology and automatic control technology. It has the characteristics of direct, fast, real and reliable, and is a non-contact direct detection method. With the specialization of image processing technology and the improvement of computer performance and price ratio, this technology has been widely used in the field of agricultural engineering.

The detection of seed particle distance is a key step in the performance detection of precision seeders. At present, the detection of seed particle distance at home and abroad is mainly based on photoelectric detection methods. The seed is calculated by recording the time interval between two seeds falling and the speed of the ground wheel. Pitch. Its method belongs to indirect detection and cannot directly reflect the actual spacing of seeds.

Contents of the invention

The object of the present invention is to provide a kind of seed particle distance detection method based on image processing and analysis technology, this method can realize three sowing single monomer test at the same time, shorten test time; The measured data is processed in real time, and a precision seeder is given. The qualified index, replay index, missed seed index and other performance indicators can automatically draw the particle distance distribution histogram.

The sequence image mosaic method in the detection of seed particle distance is characterized in that:

The method comprises the steps of:

(1) Make several equidistant and alternate marks on the recording tape;

(2) Find its centroid as the criterion of the first complete mark in the overlapping area of two frames of adjacent images of the sequence images collected by the computer;

(3) The pixels in the previous frame image are taken to the centroid of the first complete mark in the overlapping area of the image;

(4) The pixels in the next frame image start to get from the centroid of the same mark in the image overlap region;

(5) The grain distance of adjacent seeds that are not in the overlapping area in two adjacent images is the horizontal distance from the seed in the first frame image to the centroid of the image mark and the centroid of the image mark in the second frame image and the The sum of the horizontal distances of the seeds in the image.

This method is a seed distance detection method based on image processing and analysis technology. It studies the dynamic detection technology of precision seeder seeding when the detection test bench is running at different speeds, and realizes the simultaneous test of three sowing units, which greatly shortens the test time. time. Through the processing and analysis of the collected sequence images, the measurement of the distance between multiple seeds is realized by using the marker-based image matching technology. And the measured data is processed in real time, and performance indicators such as the qualified index, reseeding index, and missed seeding index of the precision seeder are given, and the histogram of the particle distance distribution is automatically drawn.

Description of drawings

Fig. 1 is a structural block diagram of the test system of the present invention

Fig. 2 is the connection schematic diagram of camera head and image card of the present invention

Fig. 3 is the original collection image of the present invention

Fig. 4 is the image after thresholding of the present invention

Fig. 5 is the image splicing principle of the present invention

Fig. 6 is a histogram of particle distance distribution of the present invention

Detailed ways

Please refer to Figure 1, the principle of seed particle distance detection is: the planter or the sowing unit (seeding device) is statically installed above the rubber belt, and the rubber belt moves horizontally relative to the planter under the drag of the speed-regulating motor. During the test, the seeds in the seed box fell on the tape through the seed meter and the seed pipe (the tape is coated with lubricating oil to stick the seeds so that there is no relative displacement between the tape and the tape), and the tape is measured by the CCD. When using the system camera, the images in a certain format are collected in real time, and the images are divided and converted by the image card, so that the computer can analyze, calculate and process, so as to measure the distance between seeds or the number of seeds within a unit distance. The evaluation results of the seeding uniformity of the given seeder. The system program implements coordinated control on the conveyor belt speed and image ranging speed at the same time.

The composition of the detection system module:

1. Control command module: main equipment sequence switch control, rubber belt moving speed, seed meter speed command control.

2. Image acquisition module: according to the specified tape speed, the size of the overlapping area and the measurement accuracy are weighed, and the sequence image is acquired in the image buffer area of the computer.

3. Image pre-segmentation module: process the image, separate the seeds from the background, and identify the squeezed seeds.

4. Data analysis and calculation module: calculate the distance between seeds or the number of seeds within a certain length interval. Analyze various performance parameters according to the requirements of the seed distance detection.

5. Post-processing module: store the test results in the database, generate test result reports and print them as required.

6. System automatic control module: receive instructions from the central computer, and complete the coordinated control of all parts of the entire system.

System hardware structure:

The measurement system needs to test three sowing monomers at the same time and complete the detection of the distance between three rows of seeds. Therefore, a color image card that supports independent acquisition of RGB three-component video signals is selected, and three synchronized black and white independent video sources are used as images. The three standard R, G, and B component inputs of the card complete data storage according to 8×3=24 bits.

·Color image acquisition card: OK_RGB10;

· CCD black and white industrial camera: WAT-505EX;

·One to four video splitter

Lens: SE1614 (16mm, 1/10000s);

·Computer: Lenovo Benyue 2000 PIII 933.

Fig. 2 is a schematic diagram of the connection between the camera and the image card of the present invention.

Image acquisition and analysis:

Image Acquisition:

The detection system is a dynamic measurement system. During the test, the tape speeds are 0.5m/s, 1m/s, 1.5m/s, 2m/s, 2.5m/s, 3m/s respectively. In order to minimize the overlap between the two frames before and after the sequence image without affecting the accuracy of splicing, it is necessary to solve the problem of dynamic image acquisition.

The camera acquires images and forms video signals in a scanning manner, line by line. The interlaced scan is divided into two fields, odd and even, and the scan lines of the odd field are evenly inserted between the adjacent scan lines of the even field. Odd and even occasions are one frame. When the linear speed of the tape is up to 3m/s, the general camera takes 40ms to capture a frame. During this period, the target will move 3×40=120mm. In the collected image, the seed becomes a deformed strip, which will seriously affect the future image. processing and analysis. This system uses a camera with an electronic shutter to eliminate image blur caused by motion by shortening the exposure time.

The images collected by the camera are divided into odd and even fields. The time interval between the odd and even fields is 20ms. In one frame of images composed of the odd and even images, there are two images of the same target. Which image is used as the seed? is indistinguishable. The OK_RGB10 image acquisition card can capture single field, single frame, several frames at intervals, continuous frames, etc., accurate to the scene. Using this feature of the image card, the image is stored in the order of field-by-field acquisition, and the field is used as the unit for processing, so that it can be guaranteed One seed has only one image in one frame of image.

Since the image acquisition speed of the camera is 25 frames/s, the maximum speed of tape movement is 3m/s, and the tape moves through 120mm during the time of capturing one frame of image. If you shoot at the inherent frequency of the camera, there will be a lot of redundant information between the two frames of images. Use the frame interval setting function okSetCaptureParam(hBoard, CAPTURE_INTERVAL, INTERVAL) provided by the OK-RGB10 image acquisition card to set different frame intervals according to different speeds, so as to reduce the amount of data and ensure the correct splicing of the two frames before and after the goal of. The formula for calculating the frame interval is as follows:

$\mathrm{INTERVAL}=\frac{200}{V\×20}-1,$ Among them, INTERVAL——frame interval, V——tape speed

Seed separation:

This system uses image thresholding segmentation method to realize the separation of background and seeds. Pixels whose gray values are greater than the threshold are the seeds, and those whose gray values are less than the threshold are the background. Based on the fact that the collected image has a fixed light source and the collection environment does not change, a fixed threshold is adopted. According to the results of many experiments, the threshold is set to 50 to achieve a good effect. Figure 3 is the original image collected, and Figure 4 is the image after two thresholding.

In Figure 3 and Figure 4, the black part is the background, the brightest point is the green channel, the darkest point is the blue channel, and the brightness is in the middle is the red channel. The dots of different sizes on the top of the image are markers, and the markers are located in the green channel.

It can be seen that in the image after thresholding, the background and seeds have been completely separated. The dots above the image are image splicing marks, and the large areas of red, green and blue in the middle are broadcast by three sowing units. Seeds, here, for the convenience of manual measurement, paper chips are used instead of seeds.

In order to identify different seeds, it is necessary to mark the connected regions in the above figure. Assigning different marks to different regions achieves the purpose of seed identification. Here, the method of pixel marking is adopted: the image is scanned from left to right and from top to bottom (the scanning area does not include the marking area), and if the gray value of the current pixel is 0, move to the next scanning position. If the grayscale value of the current pixel is 255, check its two neighbors to the left and above (according to the scan order used, these two neighbors have been processed when the scan reaches the current pixel). Four situations need to be considered: (1) their gray values are all 0, and a new mark is given to the current pixel; (2) only one gray value is 255, and the mark of the pixel is assigned to the current pixel; (3 ) their gray value is 255 and have the same mark, just assign this mark to the current pixel; (4) their gray value is 255 and have different marks, just assign one of the marks to the current pixel pixels, and make a notation indicating that the two notations are equivalent ^[2][3] . Finally the image is rescanned, replacing each marker with the marker of its equivalent pair. The three channels of red, green and blue are scanned separately, and the seeds of each channel are a group of their own, which has nothing to do with other channels.

Determination of the seed center:

After the seeds are marked, pixels with the same mark are considered to belong to the same seed. For different regions, only its centroid is required, and this point is taken as the center of the seed. We only care about the projection distance of the two seeds on the center line of the broadcast, so we only need to find the coordinates in this direction. Take the upper left corner of the image as the coordinate origin, the right direction is the positive direction of the X axis, and the downward direction is the positive direction of the Y axis. The X coordinate of the center of each seed is the desired one. The calculation formula is as follows ^[4] :

$\stackrel{\‾}{x}=\frac{\underset{t=0}{\overset{\mathrm{no}-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{m-1}{\mathrm{\Σ}}}\mathrm{jB}[i,j]}{A}$ in $A=\underset{i=0}{\overset{\mathrm{no}-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{m-1}{\mathrm{\Σ}}}B[i,j],$ B[I, j] is the label of the point

record value

The X coordinate values of the centroids of all the seeds are obtained by performing the above operations on different ones respectively.

Sequence image stitching principle:

Since the detection system is dynamic and real-time detection, the speed of the conveyor belt is continuously variable within the range of 0.5m/s---3m/s, and the detection test bench is 16m long and 1.97m wide. ) real-time detection of seeding accuracy, each seeder sows 250 grains, the distance between the grains is 30mm-150mm, and the moving length of the conveyor belt is 7500mm--37500mm, if the imaging range of one frame image It is 270mm long x 250mm wide, and the number of collected images is 30-150. In order to make the overlap between the two frames before and after the sequence image as little as possible without affecting the accuracy of splicing, solving the problem of dynamic image collection and sequence image splicing becomes The essential.

A key issue in sequence image detection is the stitching between images, and whether the image matching is accurate has an important impact on the accuracy of the measurement. This system adopts the method of marking on the adhesive tape and using this mark as the standard method for matching the two frames of images before and after. This is done on the basis of the following principles:

Figure 5(a) shows two frames of adjacent images collected by the computer. To measure the distance between two seeds at the boundary, the two frames of images are stitched together as in Figure 5(b), but the amount of calculation is quite large. Because a large number of pixels need to be moved, the algorithm is also complicated. As shown in Figure 5(c), the center of the mark is found in the two frames of the picture respectively based on the mark. In the first frame, the pixel acquisition ends here, and in the second frame, the pixel is taken from the center of the mark, which is equivalent to Stitching two frames of images saves a lot of work of moving pixels for image stitching, and achieves the effect of image stitching. In order to find L, L1 and L2 can be measured respectively, and L=L1+L2 can be obtained. For the drill, a certain distance is also taken to measure the number of seeds in the section. For the overlapping part of the boundary, as long as the two frames of images are subject to the same mark.

The image overlapping area depends on the frame interval set during acquisition, and its calculation formula is as follows:

The overlapping area between the image and the previous frame image: [0, (270-(INTERVAL+1)×20×V)×768÷270]

The overlapping area between the image and the next frame image: [768-(270-(INTERVAL+1)×20×V)×768÷270, 768]

Program implementation algorithm:

a. Find the first complete marker of the overlapping area in the previous frame image

b. Find the first complete marker in the overlapping area in the next image frame

c. Determine whether the sizes of the two marks are consistent

d. If consistent, the tag is considered to match

e. If they are inconsistent, take the second mark of the previous frame or the next frame to match until the match is complete. The influence of the tape speed is considered in the above mark matching algorithm, as long as the change of the tape moving distance within 20 ms caused by the change of the tape speed does not exceed one mark interval, the above algorithm is valid.

Measurement result:

Table 1 intercepts a part of a group of measurements when the tape speed is 0.5m/s, using the results measured by this system and manual measurement values. We used this system to conduct two experiments on the same set of seeds in order to compare the consistency of the detection results using this system.

The qualified index of this experiment is 72.0, the replay index is 17.89, the missed broadcast index is 10.53, the average value is 0.98, and the standard deviation is 0.23. It can be seen from Table 1 that the measurement results of this system are in good agreement with the manual measurement results, and the deviation is within ±2mm, which meets the requirements of the seeder performance test. Moreover, it is further found that the consistency of the results of the two comparative experiments carried out using this system is also good. The drawn histogram 6 is completely consistent with the measured results and the actual distribution of seeds. Manual measurement result(mm) The measurement result of this system 1 (mm) Deviation(mm) The measurement result of this system 2 (mm) Deviation(mm) 26.90 27.07 -0.17 27.07 -0.17 91.00 90.352 0.65 90.352 0.65 120.90 120.586 0.31 120.938 -0.04 7.50 6.68 0.82 7.031 0.47 124.80 124.102 0.70 124.102 0.70 115.50 115.313 0.19 115.313 0.19 73.00 73.477 -0.48 73.125 -0.13 54.80 53.789 1.01 54.141 0.66 45.50 45.352 0.15 45.352 0.15 90.50 90.352 0.15 90 0.50

Claims (1)

1, a kind of seed grain sequence image joining method in detect is characterized in that: this method comprises the steps:

(1) on the record adhesive tape, makes the alternate mark of several equidistant sizes, the static adhesive tape top that is installed in of seeder or planter, drag down at buncher, the relative seeder of adhesive tape moves horizontally, and the seed of planting in the case drops on adhesive tape through feed mechanism for seed, discharging tube, scribble lubricating oil on the adhesive tape, so that seed is clung, make itself and adhesive tape that relative displacement not take place, adhesive tape is by the measuring system camera time, camera and image card cooperate, in real time images acquired;

(2) adopt the image threshold split plot design to realize separating of background and seed, the gray-scale value of pixel is a seed greater than threshold value, and less than threshold value is background;

(3) adopt the method for element marking that mark is carried out in the zone that is communicated with among the figure, give different marks to discern different seeds different zones; Image is from left to right carried out four from top to bottom be communicated with scanning, scanning area does not comprise the mark zone, is 0 as the gray-scale value of current pixel, just moves on to next scanning position; Gray-scale value as current pixel is 255, checks two neighbour's pixels of its left side and top, comprises that four kinds of their gray-scale values of situation: A. all are 0, gives new mark of current pixel; B. having only a gray-scale value is 255, just the mark of this pixel is composed to current pixel; C. their gray-scale value all is 255 and has identical mark, just this mark is composed to current pixel; D. their gray-scale value all is 255 and has different marks, just one of them mark is composed to current pixel, and marking shows this two mark equivalences; Rescan image at last, each mark is replaced with the of equal value right mark in its place;

(4) pixel that mark is the same belongs to same grain seed, obtaining the center of its centre of form as seed for different zones, as true origin, is to the right the X-axis forward with the image upper left corner, be the Y-axis forward downwards, the X coordinate Calculation formula that calculates each kind subcenter is as follows:

$\stackrel{\‾}{x}=\frac{\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{m-1}{\mathrm{\Σ}}}\mathrm{jB}[i,j]}{A}$

Wherein $A=\underset{i=0}{\overset{n-1}{\mathrm{\Σ}}}\underset{j=0}{\overset{m-1}{\mathrm{\Σ}}}B[i,j],$ B[i, j] be the mark value of this point;

(5) centre of form of getting first complete mark in this doubling of the image zone of the pixel in the former frame image is ended;

(6) pixel in one two field picture of back begins to get from the centre of form of the same tag in this doubling of the image zone;

In (7) the two frame adjacent images not at the grain of the adjacent seed of overlapping region apart from the horizontal range sum that is seed in the centre of form of the seed image tagged in the horizontal range of the centre of form of this image tagged and second two field picture in first two field picture and this image;

The frame period that described doubling of the image zone is provided with when gathering and deciding, its computing formula is as follows:

The overlapping region of image and former frame image:

[0，(270-(INTERVAL+1)*20*V)*768/270]

The overlapping region of image and back one two field picture:

[768-(270-(INTERVAL+1)*20*V)*768/270，768]

Wherein: INTERVAL is a frame period, and V is the adhesive tape travelling speed

The program implementation algorithm:

A finds first complete mark of overlapping region in the previous frame image;

B finds first complete mark of overlapping region in the next frame image;

C judges whether two mark size are consistent;

If the d unanimity is just thought indicia matched;

If e is inconsistent, second mark just getting previous frame or next frame mates, till matching.

Images