A method for monitoring the boll opening rate of a cotton field at the sample scale

Abstract

The application discloses a kind of sample scale cotton field shedding rate monitoring methods, it is related to cotton field shedding rate monitoring field.Method includes: obtaining the image before and after cotton field spraying ripening agent, and obtaining contemporaneous shedding rate;Image is segmented into sub-image and is labeled, and a variety of models are used to identify the boll of sub-image, determine optimal model;Image is resegmented into the sub-image of multiple pixel sizes, and the optimal model is used to identify and label the sub-image of all pixel sizes, and reorganize into original image and keep boll label frame;Reorganized original image is grouped according to pixel size, and image is selected in grouping to be combined and overlapped two by two, and new boll label frame is formed and the shedding rate of overlapping image is calculated by judging boll label frame;The fitting effect of the shedding rate of all overlapping images is compared, the best pixel size combination is obtained, and the best shooting scheme is obtained using the best pixel size combination, so as to realize the accurate and rapid monitoring of cotton field shedding rate.

Description

Technical Field

[0001] This invention relates to the agricultural field, and in particular to a method for monitoring the boll opening rate of cotton fields at the sample point scale. Background Technology

[0002] Cotton is an important economic crop, widely cultivated globally. Applying harvesting aids for defoliation and ripening is a prerequisite for mechanical cotton harvesting, and determining the timing and dosage of these aids requires comprehensive consideration of various factors. Among these, the boll opening rate is a key factor in determining the timing and dosage of defoliation and ripening aids, evaluating their effectiveness, and determining the harvest time. Currently, in China's cotton-producing areas, a boll opening rate of 30-40% is required before applying defoliation and ripening aids. Furthermore, the boll opening rate is also an important indicator for evaluating the effectiveness of harvesting aids and determining the timing of mechanical harvesting. Traditional manual methods for investigating the boll opening rate involve selecting sample points and determining the number of plants to be surveyed, then counting the number of opened and unopened bolls on each plant, and finally calculating the boll opening rate by dividing the number of opened bolls by the total number of bolls (the sum of opened and unopened bolls). Traditional methods are labor-intensive, time-consuming, and inefficient, with a limited number of sample points surveyed per unit time, resulting in poor representativeness. Therefore, there is an urgent need to develop a rapid and accurate method for monitoring the boll opening rate in cotton fields.

[0003] In existing technologies, with the upgrading of deep learning algorithms and the improvement of computing power of personal computing devices, research on boll opening rate is mainly based on image recognition and the construction of 3D point clouds of cotton fields to achieve boll segmentation and counting. Researchers have introduced multi-scale residual modules and attention modules based on YOLOX to enhance the extraction of feature details in cotton field images. In order to reduce the recognition loss of small targets, a multi-receptive field extraction module was added to achieve image recognition of unopened cotton bolls. Other researchers have used self-created drone flight paths to obtain RGB images of cotton fields and constructed 3D point clouds of cotton fields based on them. They then used histograms of the RGB and HSV color space channels and cotton boll shape and spatial position for feature extraction, and used deep forest classification to finally achieve the recognition of opened and unopened cotton bolls.

[0004] However, the aforementioned existing technologies require compression or segmentation when processing images. Compression leads to the loss of detailed features of cotton bolls, while segmentation results in misidentification or omission of boundary cotton bolls. Furthermore, cotton fields experience severe shading during the later stages of growth, a problem that is difficult to address with the frontal images of cotton bolls acquired by drones, resulting in a large number of missed identifications of cotton bolls in the lower canopy of these images.

[0005] In summary, existing technologies suffer from the loss of detailed features of cotton bolls, as well as issues of misidentification and omission, which reduce the ability to identify cotton bolls and make the monitoring of cotton boll opening rate inaccurate and slow. Summary of the Invention

[0006] Therefore, it is necessary to provide a method for monitoring cotton boll opening rate at the sample point scale to address the above-mentioned technical problems.

[0007] The present invention adopts the following technical solution:

[0008] A method for monitoring cotton boll opening rate includes the following steps:

[0009] Multiple original images of cotton fields before and after the application of ripening agents were obtained, and the actual value of the cotton boll opening rate during the same period was obtained.

[0010] The original image is segmented into a first sub-image according to a specified pixel size and the cotton boll is labeled for the first time; multiple deep learning models for object detection are used to identify cotton bolls in the first sub-image and the identification results are obtained; the optimal deep learning model is determined based on the identification results.

[0011] The original image is re-segmented into multiple sub-images of different pixel sizes. The optimal deep learning model is used to identify cotton bolls and perform a second cotton boll annotation on the sub-images. The sub-images of each pixel size are then stitched together to obtain the reconstructed original image of each pixel size. Each reconstructed original image retains the cotton boll annotation bounding box of its sub-image.

[0012] The original images of the reconstruction of all pixel sizes are divided into two groups according to pixel size. One original image of the reconstruction of pixel size is selected from each group and combined and overlapped to obtain the overlapping image of each pixel size combination.

[0013] Determine the overlap of the cotton boll annotation boxes of the two reconstructed original images in the overlapping images of each pixel size combination; based on the determination result, replace the cotton boll annotation boxes in the overlapping images with new cotton boll annotation boxes, obtain each overlapping image with new cotton boll annotation boxes, and calculate the boll opening rate of each overlapping image through the cotton bolls in the new cotton boll annotation boxes.

[0014] By comparing the fitting effect of the boll opening rate of each overlapping image with the actual boll opening rate of cotton fields during the same period, the optimal pixel size combination of the two reconstructed original images in the overlapping images is obtained; the boll opening rate of cotton fields is monitored using the obtained optimal pixel size combination.

[0015] Preferably, when acquiring original images of cotton fields before and after the application of ripening agents via ground-based photography, the method further includes:

[0016] By comparing the fitting effect of the boll opening rate of the overlapping image with the actual boll opening rate of the cotton field at the same time under different shooting scheme parameters, the optimal shooting scheme parameters of the original image are obtained.

[0017] The cotton boll opening rate was monitored using the obtained optimal pixel size combination and optimal shooting scheme parameters.

[0018] Preferably, the shooting scheme parameters are based on the area above the cotton field canopy as a reference point, and the shooting scheme parameters include the height, angle, and shooting direction set during shooting.

[0019] Preferably, the investigation of the true value of the ground-level cotton-like particle opening rate at each sampling point during the same period specifically includes:

[0020] Cotton plants with uniform growth were selected at each sampling point. After the photos were taken, the total number of bolls and the number of bolls that opened were counted for each plant. The true value of the ground boll opening rate is the ratio of the number of bolls that opened to the total number of bolls.

[0021] Preferably, obtaining the overlapping image of each pixel size combination includes the following steps:

[0022] Using the median value of multiple pixel sizes as a dividing line, the reconstructed original image of all pixel sizes is divided into a first group and a second group;

[0023] One pixel-sized original image is selected from each of the first and second groups, and then combined and overlapped to obtain an overlapping image with one pixel-sized combination.

[0024] Repeatedly select and recombine the original images from the two groups, combining and overlapping them until an overlapping image with each pixel size combination is obtained; wherein the pixel size combinations of the two recombine original images in each overlapping image are different.

[0025] Preferably, acquiring all overlapping images with new cotton boll marking boxes specifically includes:

[0026] For overlapping images of each pixel size combination, determine the overlap of the cotton boll bounding boxes of the reconstructed original image for two pixel sizes in the overlapping image, and form new cotton boll bounding boxes for the overlapping image, including:

[0027] If, in the overlapping images, the cotton boll label frame of the reconstructed original image of one pixel size in the first group does not overlap with the cotton boll label frame of any reconstructed original image of any pixel size in the second group, then the cotton boll label frame of the reconstructed original image of one pixel size in the first group is retained to form a new cotton boll label frame in the overlapping image.

[0028] If, in the overlapping images, the cotton boll label box of the reconstructed original image of one pixel size in the second group does not overlap with the cotton boll label box of any reconstructed original image of any pixel size in the first group, then the cotton boll label box of the reconstructed original image of one pixel size in the second group is retained to form a new cotton boll label box in the overlapping image.

[0029] If the cotton boll annotation boxes of the two reconstructed original images overlap, then calculate the outermost position of the cotton boll annotation boxes of the two reconstructed original images and annotate the cotton bolls to form new cotton boll annotation boxes in the overlapping images.

[0030] By replacing the cotton boll label boxes of the reconstructed original image in the overlapping image with the new cotton boll label boxes formed above, an overlapping image with new cotton boll label boxes is obtained.

[0031] Preferably, the formation of the new cotton boll label box in the overlapping image specifically includes:

[0032] With the bottom left corner of the image as the origin, the positive Y-axis is vertically upward, and the positive X-axis is horizontally to the right. The coordinates of the four corner points of the cotton boll annotation box of the reconstructed original image of one pixel size in the first group are recorded as bottom left (xa1, ya1), top left (xa2, ya2), top right (xa3, ya3), and bottom right (xa4, ya4). The cotton boll annotation boxes of the reconstructed original image of one pixel size in the second group follow the same order, and their four corner point coordinates are recorded as (xb1, yb1), (xb2, yb2), (xb3, yb3), and (xb4, yb4).

[0033] Compare the values ​​of the eight sets of coordinates in each of the four sets of coordinates. Keep the smaller values ​​among xa1 and xb1, ya1 and yb1, xa2 and xb2, and ya4 and yb4, and denote them as xc1, yc1, xc2, and yc4, respectively. Keep the larger values ​​among ya2 and yb2, xa3 and xb3, ya3 and yb3, and xa4 and xb4, and denote them as yc2, xc3, yc3, and xc4, respectively.

[0034] Through the above steps, four new sets of coordinates are obtained, including: lower left corner coordinates (xc1, yc1), upper left corner coordinates (xc2, yc2), upper right corner coordinates (xc3, yc3), and lower right corner coordinates (xc4, yc4).

[0035] The quadrilateral formed by the four new sets of coordinates is the new cotton boll annotation box of the overlapping image.

[0036] The above-mentioned at least one technical solution adopted in this invention can achieve the following beneficial effects:

[0037] This invention segments the original image to obtain sub-images of multiple pixel sizes. These sub-images are then labeled using model recognition and reconstructed into original images of varying pixel sizes. The original images are then grouped, combined, and overlapped according to pixel size to obtain overlapping images with different pixel combinations. By judging the cotton boll annotation boxes in the overlapping images, incomplete cotton bolls at the image edges can be pieced together into a complete cotton boll. Larger cotton boll annotation boxes in the overlapping images may cover the entire cotton boll, while smaller cotton boll annotation boxes can be used to capture the image details of smaller cotton bolls. This solves the problems of cotton boll detail loss caused by image compression and segmentation, as well as misidentification and omission of cotton boll boundaries.

[0038] In addition, the ground-based photography method used in this invention makes it easier to obtain image information of the lower part of the cotton boll canopy, resulting in more complete image data and avoiding shading in cotton fields during the later stages of growth.

[0039] In summary, this invention improves the identification capability of cotton bolls by combining and overlapping different pixel sizes to obtain the optimal pixel size combination and determining the optimal shooting scheme, thereby achieving accurate and rapid monitoring of cotton boll opening rate. Attached Figure Description

[0040] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this invention, illustrate exemplary embodiments of the invention and are used to explain the invention, but do not constitute an undue limitation of the invention. In the drawings:

[0041] Figure 1 A schematic flowchart illustrating a method for monitoring cotton boll opening rate at the sample point scale provided by the present invention;

[0042] Figure 2 A schematic diagram of an image acquisition scheme for a cotton field boll opening rate monitoring method at the sample point scale provided by the present invention;

[0043] Figure 3 The YOLOv5m model used in this invention to identify sub-images of all pixel sizes is shown in the figure.

[0044] Figure 4 A comparison of the recognition effects before and after the merging operation of 400 and 700 pixel sub-images in a cotton field boll opening rate monitoring method at the sample point scale provided by the present invention.

[0045] Figure 5 The fitting effect of the boll opening rate of an overlapping image of all pixel sizes of a cotton field boll opening rate monitoring method at the sample point scale provided by the present invention on the boll opening rate with the true value. Detailed Implementation

[0046] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0047] The technical solutions provided by the various embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0048] Figure 1 This is a schematic diagram of a method for monitoring cotton boll opening rate at the sample point scale according to the present invention, which specifically includes the following steps:

[0049] S101: Obtain multiple original images of cotton fields before and after the application of ripening agents, and obtain the true value of the cotton boll opening rate during the same period;

[0050] The acquisition of the true value of the cotton boll opening rate during the same period includes: selecting 20 cotton plants with uniform growth at each sampling point, and after taking pictures, investigating the total number of bolls and the number of bolls that have opened for bolls for each plant. The true value of the ground boll opening rate is the ratio of the number of bolls that have opened for bolls to the total number of bolls.

[0051] S102: Divide the original image into a first sub-image according to a specified pixel size and perform the first cotton boll labeling; use multiple deep learning models for object detection to identify cotton bolls in the first sub-image and obtain the recognition results; determine the optimal deep learning model based on the recognition results.

[0052] S103: The original image is re-segmented into second sub-images of multiple pixel sizes. The optimal deep learning model is used to identify cotton bolls and perform a second cotton boll annotation on the second sub-images. The second sub-images of each pixel size are then stitched together to reconstruct the original image, resulting in a reconstructed original image of each pixel size. Each reconstructed original image of each pixel size retains the cotton boll annotation box of its second sub-image.

[0053] S104: Divide the reconstructed original images of all pixel sizes into two groups. Select one pixel-sized reconstructed original image from each group, combine them pairwise, and overlap them to obtain an overlapping image for each pixel-sized combination. Specifically, this includes:

[0054] Using the median value of multiple pixel sizes as a dividing line, the reconstructed original image of all pixel sizes is divided into a first group and a second group;

[0055] One pixel-sized original image is selected from each of the first and second groups, and then combined and overlapped to obtain an overlapping image with one pixel-sized combination.

[0056] Repeatedly select and recombine the original images from the two groups, combining and overlapping them until an overlapping image with each pixel size combination is obtained; wherein the pixel size combinations of the two recombine original images in each overlapping image are different.

[0057] S105: Determine the overlap of the cotton boll annotation boxes in the two reconstructed original images in the overlapping images of each pixel size combination, and based on the determination result, replace the cotton boll annotation boxes in the overlapping images with new cotton boll annotation boxes, obtain each overlapping image with a new cotton boll annotation box, and calculate the boll opening rate of each overlapping image through the cotton bolls within the new cotton boll annotation boxes, specifically including:

[0058] For overlapping images of each pixel size combination, determine the overlap of the cotton boll bounding boxes of the reconstructed original image for two pixel sizes in the overlapping image, and form new cotton boll bounding boxes for the overlapping image, including:

[0059] For overlapping images of each pixel size combination, determine the overlap of the cotton boll bounding boxes of the reconstructed original image for two pixel sizes in the overlapping image, and form new cotton boll bounding boxes for the overlapping image, including:

[0060] If, in the overlapping images, the cotton boll label frame of the reconstructed original image of one pixel size in the first group does not overlap with the cotton boll label frame of any reconstructed original image of any pixel size in the second group, then the cotton boll label frame of the reconstructed original image of one pixel size in the first group is retained to form a new cotton boll label frame in the overlapping image.

[0061] If, in the overlapping images, the cotton boll label box of the reconstructed original image of one pixel size in the second group does not overlap with the cotton boll label box of any reconstructed original image of any pixel size in the first group, then the cotton boll label box of the reconstructed original image of one pixel size in the second group is retained to form a new cotton boll label box in the overlapping image.

[0062] If the cotton boll annotation boxes of the two reconstructed original images overlap, then calculate the outermost position of the cotton boll annotation boxes of the two reconstructed original images and annotate the cotton bolls to form new cotton boll annotation boxes in the overlapping images.

[0063] The new cotton boll annotation boxes in the overlapping image formed above replace the cotton boll annotation boxes in the reconstructed original image in the overlapping image, thus obtaining an overlapping image with new cotton boll annotation boxes;

[0064] The formation of the new cotton boll label box in the overlapping images includes:

[0065] With the bottom left corner of the image as the origin, the positive Y-axis is vertically upward, and the positive X-axis is horizontally to the right. The coordinates of the four corner points of the cotton boll annotation box of the reconstructed original image of one pixel size in the first group are recorded as bottom left (xa1, ya1), top left (xa2, ya2), top right (xa3, ya3), and bottom right (xa4, ya4). The cotton boll annotation boxes of the reconstructed original image of one pixel size in the second group follow the same order, and their four corner point coordinates are recorded as (xb1, yb1), (xb2, yb2), (xb3, yb3), and (xb4, yb4).

[0066] Compare the values ​​of the eight sets of coordinates in each of the four sets of coordinates. Keep the smaller values ​​among xa1 and xb1, ya1 and yb1, xa2 and xb2, and ya4 and yb4, and denote them as xc1, yc1, xc2, and yc4, respectively. Keep the larger values ​​among ya2 and yb2, xa3 and xb3, ya3 and yb3, and xa4 and xb4, and denote them as yc2, xc3, yc3, and xc4, respectively.

[0067] Through the above steps, four new sets of coordinates are obtained, including: lower left corner coordinates (xc1, yc1), upper left corner coordinates (xc2, yc2), upper right corner coordinates (xc3, yc3), and lower right corner coordinates (xc4, yc4).

[0068] The quadrilateral formed by the four new sets of coordinates is the new cotton boll annotation box of the overlapping image.

[0069] S106: By comparing the fitting effect of the boll opening rate of each overlapping image with the actual value of the boll opening rate of cotton fields in the same period, the optimal pixel size combination of the two reconstructed original images in the overlapping images is obtained; the boll opening rate of cotton fields is monitored by the obtained optimal pixel size combination.

[0070] S107: By comparing the fitting effect between the boll opening rate of the overlapping image with the actual boll opening rate of cotton fields under different shooting scheme parameters and the optimal pixel size combination, the optimal shooting scheme parameters of the original image are obtained; the boll opening rate of cotton fields is monitored using the obtained optimal pixel size combination and optimal shooting scheme parameters.

[0071] The shooting plan parameters are based on the area above the cotton field canopy as a reference point, and the parameters set during shooting include height, angle, and shooting direction.

[0072] In addition, the above-mentioned methods for capturing original images of cotton fields, such as... Figure 2 As shown, this includes: taking pictures of the cotton field from 4 heights, 5 angles and 2 shooting directions above the cotton field canopy to obtain original images of the cotton field at multiple periods before and after the application of ripening agent;

[0073] The system includes four height settings: 0cm, 10cm, 20cm, and 30cm; five angle settings: 0°, 15°, 30°, 45°, and 60°; and two shooting directions: parallel to the planting row and perpendicular to the planting row. The reference angle of 0° is horizontal to the right, and the five angles are set clockwise according to the direction of the reference angle.

[0074] Taking the Chinese hybrid cotton H318 variety grown in a certain area in 2022 as an example, it was sown on April 25th and field management was carried out according to local routines. On September 23rd, a harvesting aid of 50% thiamethoxam·ethephon suspension (T·E) was sprayed, with four treatments of 750, 1500, 3000, and 4500 mL·hm⁻², and water as the control.

[0075] Images were collected and the boll opening rate was manually investigated on September 15th, September 30th, October 7th, and October 14th. A Canon EOS 550 camera (P mode, aperture f / 13, exposure time 1 / 160s, focal length 16mm, ISO-400) was used. After selecting sample points, photos were taken at different heights above the canopy using different angles and shooting directions, resulting in 912 photos with a resolution of 5184*3456 pixels. Shooting heights were set at 0, 10, 20, and 30cm (4 horizontal levels), shooting angles (angle downwards from the horizontal plane) were set at 0°, 15°, 30°, 45°, and 60° (5 horizontal levels), and shooting directions were set parallel to and perpendicular to the planting rows. Large areas of a single leaf obstructing the view in front of the lens were avoided during shooting.

[0076] According to the method in step S102, the model with the best overall performance in terms of recognition accuracy and recognition time is selected as the optimal deep learning model for object detection. The recognition results are shown in the table below:

[0077]

[0078] Using 8121 images (500*500 pixels) as the training dataset, the object detection models YOLO v3, Faster R-CNN, YOLOv5, and CeInternet all achieved acceptable accuracy levels (precision > 0.8). YOLOv5 and Faster R-CNN showed higher accuracy, with their classification accuracy and mAP approaching 0.9. Considering the overall recognition time, YOLOv5 achieved satisfactory accuracy at a relatively fast speed. While Faster R-CNN ranked first in all accuracy metrics, its recognition time was 3-5 times longer than YOLOv5. Taking all factors into account, YOLOv5 is the optimal deep learning model for object detection.

[0079] The overlapping image of each pixel size combination is obtained by using the methods in steps S103 to S105;

[0080] In step S103, the pixel sizes used in this embodiment include: 100*100, 200*200, 300*300, 400*400, 500*500, 600*600, 700*700, 800*800, 900*900, and 1000*1000.

[0081] In addition, in step S104, the reconstructed original image with a pixel size of 500*500 is used as the dividing line, and the reconstructed original images with pixel sizes of 100*100, 200*200, 300*300, 400*400 and 500*500 are divided into the first group, and the reconstructed original images with other pixel sizes are divided into the second group.

[0082] The fitting effect of the calculated unfurling rate of the overlapping images for each pixel size combination to the true value is as follows: Figure 3 As shown, when the sub-image size is between 100 and 200 pixels, regardless of the combination of image pixel sizes, the R-value of the fitting effect is... 2 The values ​​were all relatively low (46%). <R 2 <72%). This indicates that when the sub-image size is small, the recognition accuracy may be low due to the lack of detailed information or limitations of the recognition algorithm, thus affecting the overall fitting effect. However, in the combination of sub-image sizes of 300 to 400 pixels and 600 to 700 pixels, as well as the combination of 500 pixels and 700 to 900 pixels, the fitting effect is significantly improved, with R... 2The values ​​are generally high, with the best results achieved by overlapping images combining 400*400 and 700*700 pixels. 2 The recognition rate reached 82.80%, with a comparison of the recognition performance before and after combining 400*400 and 700*700 pixels. Figure 4 As shown.

[0083] Cotton bolls were identified using a combination of 400 and 700 pixel sizes. The boll opening rate for various shooting scenarios was then calculated and fitted to the true values. The influence of shooting parameters (shooting height, angle, and direction) on the fitting effect on the boll opening rate was determined, and the fitting results are shown below. Figure 5 As shown in the figure, the closer the data point is to the top left corner, the better the fitting effect. In the entire dataset and the dataset taken perpendicular to the row direction, the RRMSE of the fitting model for shooting angles of 0-30° (angle with the horizontal direction) is all concentrated on the left side of the X-axis, while 45° and 60° are more loosely distributed on the right side; the RRMSE of the fitting model taken at distances of 20 and 30 cm from the canopy is generally concentrated on the left side, with some combinations of 0 and 10 cm also concentrated on the left side. Compared with shooting perpendicular to the row direction, the combination of shooting angle and height taken parallel to the row direction has higher overall recognition accuracy, with 8 combinations showing higher RRMSE. 2 The RRMSE was above 0.9 for all seven combinations and less than 10 for all seven combinations. Specifically, the parameters with better recognition performance were the combination of 30cm with 0, 15, and 30°, and the combination of 20cm with 0, 15, and 30°.

[0084] In summary, this invention provides a method for monitoring cotton boll opening rate at the sampling scale. The method involves segmenting the original image into multi-pixel sub-images, performing model recognition annotation on these sub-images, and reconstructing them into original images of each pixel size. The original images are then grouped, combined, and overlapped according to pixel size to obtain overlapping images with different pixel combinations. By judging the cotton boll annotation boxes in the overlapping images, incomplete cotton bolls at the image edges can be pieced together into a complete cotton boll. Larger cotton boll annotation boxes in the overlapping images may cover the entire cotton boll, while smaller cotton boll annotation boxes can be used to capture image details of smaller cotton bolls, thus solving the problem of misidentification and omission of cotton boll boundaries caused by image segmentation. Furthermore, using ground-based photography makes it easier to obtain image information of the lower part of the cotton boll canopy, resulting in more complete image data and avoiding shading phenomena that may occur in cotton fields during the later stages of growth. In summary, this invention improves the identification capability of cotton bolls by combining and overlapping different pixel sizes to obtain the optimal pixel size combination and determining the optimal shooting scheme, thereby achieving accurate and rapid monitoring of cotton boll opening rate and providing strong scientific guidance for cotton boll opening rate monitoring.

[0085] The technical solutions of the present invention are not limited to the specific embodiments described above. Any technical modifications made in accordance with the technical solutions of the present invention fall within the protection scope of the present invention.

Claims

1. A method for monitoring the boll opening rate in cotton fields, characterized in that, The method includes the following steps: Multiple original images of cotton fields before and after the application of ripening agents were obtained, and the actual value of the cotton boll opening rate during the same period was obtained. The original image is segmented into a first sub-image according to a specified pixel size and the cotton boll is labeled for the first time; multiple deep learning models for object detection are used to identify cotton bolls in the first sub-image and the identification results are obtained; the optimal deep learning model is determined based on the identification results. The original image is re-segmented into multiple sub-images of different pixel sizes. The optimal deep learning model is used to identify cotton bolls and perform a second cotton boll annotation on the sub-images. The sub-images of each pixel size are then stitched together to obtain the reconstructed original image of each pixel size. Each reconstructed original image retains the cotton boll annotation bounding box of its sub-image. The original images of the reconstruction of all pixel sizes are divided into two groups according to pixel size. One original image of the reconstruction of pixel size is selected from each group and combined and overlapped to obtain the overlapping image of each pixel size combination. Determine the overlap of the cotton boll annotation boxes of the two reconstructed original images in the overlapping images of each pixel size combination; based on the determination result, replace the cotton boll annotation boxes in the overlapping images with new cotton boll annotation boxes, obtain each overlapping image with new cotton boll annotation boxes, and calculate the boll opening rate of each overlapping image through the cotton bolls in the new cotton boll annotation boxes. By comparing the fitting effect of the boll opening rate of each overlapping image with the actual boll opening rate of cotton fields during the same period, the optimal pixel size combination of the two reconstructed original images in the overlapping images is obtained; the boll opening rate of cotton fields is monitored using the obtained optimal pixel size combination.

2. The method for monitoring cotton boll opening rate as described in claim 1, characterized in that, When acquiring raw images of cotton fields before and after the application of ripening agents via ground-based photography, the method further includes: By comparing the fitting effect of the boll opening rate of the overlapping image with the actual boll opening rate of the cotton field at the same time under different shooting scheme parameters, the optimal shooting scheme parameters of the original image are obtained. The cotton boll opening rate was monitored using the obtained optimal pixel size combination and optimal shooting scheme parameters.

3. The method for monitoring cotton boll opening rate as described in claim 2, characterized in that, The shooting scheme parameters are based on the area above the cotton field canopy as a reference point, and include the height, angle, and shooting direction set during shooting.

4. The method for monitoring cotton boll opening rate as described in claim 1, characterized in that, Obtaining the true value of the cotton boll opening rate in the same period specifically includes: Cotton plants with uniform growth were selected at each sampling point. After the photos were taken, the total number of bolls and the number of bolls that opened were counted for each plant. The true value of the cotton field opening rate is the ratio of the number of bolls that opened to the total number of bolls.

5. The method for monitoring cotton boll opening rate as described in claim 1, characterized in that, Obtaining the overlapping image of each pixel size combination includes the following steps: Using the median value of multiple pixel sizes as a dividing line, the reconstructed original image of all pixel sizes is divided into a first group and a second group; One pixel-sized original image is selected from each of the first and second groups, and then combined and overlapped to obtain an overlapping image with one pixel-sized combination. Repeatedly select and recombine the original images from the two groups, combining and overlapping them until an overlapping image with each pixel size combination is obtained; wherein the pixel size combinations of the two recombine original images in each overlapping image are different.

6. The method for monitoring cotton boll opening rate as described in claim 1, characterized in that, The acquisition of each overlapping image with a new cotton boll label specifically includes: For overlapping images of each pixel size combination, determine the overlap of the cotton boll bounding boxes of the reconstructed original image for two pixel sizes in the overlapping image, and form new cotton boll bounding boxes for the overlapping image, including: If, in the overlapping images, the cotton boll label frame of the reconstructed original image of one pixel size in the first group does not overlap with the cotton boll label frame of any reconstructed original image of any pixel size in the second group, then the cotton boll label frame of the reconstructed original image of one pixel size in the first group is retained to form a new cotton boll label frame in the overlapping image. If, in the overlapping images, the cotton boll label box of the reconstructed original image of one pixel size in the second group does not overlap with the cotton boll label box of any reconstructed original image of any pixel size in the first group, then the cotton boll label box of the reconstructed original image of one pixel size in the second group is retained to form a new cotton boll label box in the overlapping image. If the cotton boll annotation boxes of the two reconstructed original images overlap, then calculate the outermost position of the cotton boll annotation boxes of the two reconstructed original images and annotate the cotton bolls to form new cotton boll annotation boxes in the overlapping images. By replacing the cotton boll label boxes of the reconstructed original image in the overlapping image with the new cotton boll label boxes formed above, an overlapping image with new cotton boll label boxes is obtained.

7. A method for monitoring cotton boll opening rate as described in claim 1 or 6, characterized in that, The formation of the new cotton boll annotation box in the overlapping image specifically includes: Taking the bottom left corner of the image as the origin, the positive Y-axis is vertically upward, and the positive X-axis is horizontally to the right. The coordinates of the four corner points of the cotton boll annotation box of the reconstructed original image (one pixel in size) in the first group are recorded as bottom left (xa1, ya1), top left (xa2, ya2), top right (xa3, ya3), and bottom right (xa4, ya4). The same order is used for the cotton boll annotation box of the reconstructed original image (one pixel in size) in the second group, and the coordinates of its four corner points are recorded as (xb1, yb1), (xb2, yb2), (xb3, yb3), and (xb4, yb4). Compare the values ​​of the eight sets of coordinates in each of the four sets of coordinates. Keep the smaller values ​​among xa1 and xb1, ya1 and yb1, xa2 and xb2, and ya4 and yb4, and denote them as xc1, yc1, xc2, and yc4, respectively. Keep the larger values ​​among ya2 and yb2, xa3 and xb3, ya3 and yb3, and xa4 and xb4, and denote them as yc2, xc3, yc3, and xc4, respectively. Through the above steps, four new sets of coordinates are obtained, including: bottom left corner coordinates (xc1, yc1), top left corner coordinates (xc2, yc2), top right corner coordinates (xc3, yc3), and bottom right corner coordinates (xc4, yc4). The quadrilateral formed by the four new sets of coordinates is the new cotton boll annotation box of the overlapping image.

Images