Vision-based target biological stacking identification and dispersion method, system and device

Abstract

The present invention discloses a method, system and device for identifying and breaking up target biological stacks based on vision, the method comprising acquiring a target biological image; detecting the target biological region image to obtain a surface target region; filling the surface target region to obtain a surface target visible region, and then obtaining a lower layer obscured region; detecting the target biological in the lower layer obscured region, and if the target biological outline is detected, it is determined that the target biological is stacked; breaking up the stacked target biological until no stacking occurs in the target biological image; and performing statistical analysis on the target biological in the non-stacked image. The present invention utilizes a region filling segmentation method and a deep learning algorithm to automatically determine whether there is a target biological stack in the collected image, and automatically triggers a vibration device to break up the stacked target biological insect bodies, greatly improving the detection efficiency and accuracy.

Description

Technical Field

[0001] The present invention belongs to the technical field of image recognition, and in particular relates to a method, system and device for identifying and breaking up target biological stacks based on vision. Background Art

[0002] In the prior art, in order to classify and count target biological specimens, it is first necessary to count the occurrence frequencies of various target organisms in the area where the equipment is located, so as to achieve fast and accurate classification and counting. This method basically relies on professionals to manually classify and count target biological specimens, which is time-consuming and labor-intensive and prone to misjudgment due to visual fatigue. During the period of target biological population explosion, relying solely on manpower will lead to low efficiency in identification and counting, and it is impossible to obtain key effective data in the first place.

[0003] At present, there are also devices that use high-definition cameras embedded in the device for image acquisition, but the whole system is expensive, and the accumulation of insect bodies is a pain point and difficulty of current equipment. The accumulation of insect bodies makes it impossible to accurately identify and count the target organisms, which is also an important reason that currently restricts the intelligent identification and accurate counting of target organisms. The solution provided by Chinese patent CN104918007A uses a simple and crude method to mechanically drive the overlapping insect bodies to scatter the insects, but it is unable to judge the degree of accumulation of the insect bodies, resulting in the scattering of insects even when there is no stacking or the scattering of insects has reached the standard. The detection efficiency and accuracy cannot meet the actual needs. Summary of the invention

[0004] In view of the deficiencies of the prior art, the present invention discloses a method, system and device for identifying and breaking up target biological stacks based on vision.

[0005] A method for identifying and breaking up target biological stacks based on vision comprises the following steps:

[0006] Acquire a target biological image, and process the target biological image to obtain a target biological region image;

[0007] Performing target detection on the target biological region image to obtain a surface target region;

[0008] Filling is performed based on the surface target area to obtain a surface target visible area, and then obtaining a lower layer blocked area;

[0009] Performing target organism detection on the blocked area of ​​the lower layer, and if the outline of the target organism is detected, it is determined that the target organisms are stacked;

[0010] The stacked target organisms are broken up and the target organism images are recollected until there is no stacking in the target organism images, and a non-stacked image is obtained;

[0011] Detect, count and analyze the target organisms in the non-stacked image to obtain statistical analysis results.

[0012] As an implementable method, acquiring the target biological image and processing the target biological image to obtain the target biological region image includes the following steps:

[0013] Performing grayscale binarization processing on the target biological image to obtain a binarized image;

[0014] Each connected domain in the binary image is extracted, and the minimum circumscribed rectangle of each connected domain is calculated to segment the target biological region image.

[0015] As an implementable method, the binarization method is the Otsu threshold method, which sets a first preset threshold so that the maximum inter-class variance is maximized. The calculation formula of the Otsu threshold method is:

[0016]

[0017] in, represents the maximum between-class variance, represents the proportion of foreground image pixels under the first preset threshold, represents the pixel ratio of the background image under the first preset threshold, represents the average grayscale of the foreground under the first preset threshold, represents the average grayscale of the background under the first preset threshold, Represents the average grayscale of the target biological image.

[0018] As an implementable method, target detection is performed on the target biological region image to obtain the surface target region, and target detection is performed on the target biological region image through an image detection model, and the image detection model at least includes: multiple convolution layers, a C3 structure layer, a detection layer and an SPPF layer;

[0019] The image detection model includes a first type detection model and a second type detection model.

[0020] As an implementable method, performing target detection on the target biological region image to obtain the surface target region includes the following steps:

[0021] Obtain a first type of training sample data set and a second type of training sample data set, and divide the first type of training sample data set and the second type of training sample data set into a training set, a validation set, and a test set;

[0022] The first type of training sample data set is marked and directly input into the first type of pre-training model for training; the second type of training sample data set is image-blocked and clustered and then input into the second type of pre-training model for training, the target biological area image is detected, and the surface target area is obtained.

[0023] As an implementable method, the image detection model uses a CIOU loss function for error calibration, and the calculation formula of the CIOU loss function is:

[0024]

[0025] in:

[0026]

[0027]

[0028]

[0029] in, Represents the Euclidean distance between the network prediction box and the real box, Represents the center point coordinates of the network prediction box and the real box, Represents the information of the real box annotation, represents the weight function, It is used to measure the consistency of aspect ratio. Represent the width and height of the real box and the width and height of the predicted box respectively. Indicates the diagonal length of the outer frame of the real box and the predicted box, is the intersection-and-union ratio, is the loss function.

[0030] As an implementable method, the filling based on the surface target area to obtain the surface target visible area, and then obtain the underlying obscured area, includes the following steps:

[0031] Based on the surface target area, performing median filtering on the target biological area image to obtain a filtered image;

[0032] Based on the filtered image, edge detection is performed on the filtered image to obtain an image containing edge information;

[0033] Extracting all endpoints of the image containing edge information to obtain an endpoint data set, and obtaining the target biological contour based on the endpoint data set;

[0034] Based on the target biological area image, the minimum enclosing rectangle is extracted, the target biological outline and the pixel points within the minimum enclosing rectangle are used as background sampling points, the pixel values ​​of the background sampling points are calculated, and the target biological outline is segmented by the pixel values ​​to segment the visible area of ​​the surface target, and then the underlying obscured area is obtained.

[0035] As an implementation method, the edge detection calculates the gradient size and direction of each pixel point of the filtered image.

[0036] The gradient size is obtained by the following method:

[0037]

[0038] The gradient direction is obtained by:

[0039]

[0040] in, and Represents the gradient components in the x and y directions, where x and y represent the horizontal and vertical directions on the two-dimensional image, respectively. ‌ represents the gradient direction of the current point, ‌Indicates the gradient size of the current point.

[0041] A vision-based target biological stacking recognition and breaking-up system includes a target biological image acquisition module, a surface target area acquisition module, an image filling and segmentation module, a stacking detection module, a vibration breaking-up module, and a statistical analysis module.

[0042] A target biological image acquisition module, used to acquire a target biological image, process the target biological image, and obtain a target biological region image;

[0043] A surface target region acquisition module, used for performing target detection on the target biological region image to obtain a surface target region;

[0044] An image filling and segmentation module is used to fill in the surface target area to obtain the surface target visible area, and then obtain the underlying blocked area;

[0045] A stacking detection module is used to detect target organisms in the lower blocked area, and if the outline of the target organism is detected, it is determined that the target organisms are stacked;

[0046] A stack vibration breaking module is used to break up the stacked target organisms and re-collect the target organism image until there is no stacking in the target organism image, and then obtain a non-stacked image;

[0047] The statistical analysis module is used to detect, count and analyze the target organisms in the non-stacked image to obtain statistical analysis results.

[0048] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following method is implemented:

[0049] Acquire a target biological image, and process the target biological image to obtain a target biological region image;

[0050] Performing target detection on the target biological region image to obtain a surface target region;

[0051] Filling is performed based on the surface target area to obtain a surface target visible area, and then obtaining a lower layer blocked area;

[0052] Performing target organism detection on the blocked area of ​​the lower layer, and if the outline of the target organism is detected, it is determined that the target organisms are stacked;

[0053] The stacked target organisms are broken up and the target organism images are recollected until there is no stacking in the target organism images, and a non-stacked image is obtained;

[0054] Detect, count and analyze the target organisms in the non-stacked image to obtain statistical analysis results.

[0055] A device for identifying and breaking up target biological stacks based on vision includes a memory, a processor, and a computer program stored in the memory and running on the processor. When the processor executes the computer program, the following method is implemented:

[0056] Acquire a target biological image, and process the target biological image to obtain a target biological region image;

[0057] Performing target detection on the target biological region image to obtain a surface target region;

[0058] Filling is performed based on the surface target area to obtain a surface target visible area, and then obtaining a lower layer blocked area;

[0059] Performing target organism detection on the blocked area of ​​the lower layer, and if the outline of the target organism is detected, it is determined that the target organisms are stacked;

[0060] The stacked target organisms are broken up and the target organism images are recollected until there is no stacking in the target organism images, and a non-stacked image is obtained;

[0061] Detect, count and analyze the target organisms in the non-stacked image to obtain statistical analysis results.

[0062] The present invention has significant technical effects due to the adoption of the above technical solution:

[0063] 1. The present invention uses the area filling segmentation method and deep learning algorithm to automatically determine whether there is a target biological stack in the collected image, and automatically triggers the vibration device to break up the accumulated target biological insects, which greatly improves the detection efficiency;

[0064] 2. The present invention trains the first type detection model and the second type detection model separately, thereby realizing targeted identification and counting of target organisms of different sizes, improving identification accuracy, and achieving accurate counting;

[0065] 3. The vision-based target biological stack identification and breakup device described in the present invention has low hardware requirements, low cost and high detection efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0067] Figure 1 It is a schematic diagram of the overall structure of the device for identifying and breaking up target biological stacks based on vision according to the present invention;

[0068] Figure 2 It is a schematic diagram of the overall process of the method for visually identifying and breaking up target biological stacks according to the present invention;

[0069] Figure 3 It is an overall schematic diagram of the vision-based target biological stacking identification and dispersion system of the present invention;

[0070] Figure 4 In the embodiment of the present invention, an image acquisition device is used to acquire an image of a target organism;

[0071] Figure 5 Schematic diagram of the C3 structural layer in an embodiment of the present invention. DETAILED DESCRIPTION

[0072] The present invention will be further described below in conjunction with the accompanying drawings and embodiments. The following embodiments are explanations of the present invention and the present invention is not limited to the following embodiments.

[0073] Embodiment 1:

[0074] A device for identifying and breaking up target biological stacks based on vision, such as Figure 1 As shown, including:

[0075] A base 11, a camera assembly 12, a vibration assembly 13, a flat plate 14, and a storage table 15. The camera assembly 12 is placed perpendicular to the side of the base 11. The vibration assembly 13 includes four vibration columns, which are respectively placed perpendicularly at the four corners of the base 11. The flat plate 14 is placed on the vibration assembly 13, and a storage table 15 is provided on the flat plate 14.

[0076] The camera assembly 12 includes a plurality of camera modules. The vibration column moves up and down along an axis perpendicular to the flat plate 14 to drive the flat plate 14 to vibrate.

[0077] The vision-based target biological stack identification and breaking-up device has low hardware requirements and has the advantages of low cost and high detection efficiency.

[0078] Embodiment 2:

[0079] A vision-based target biological stacking recognition and dispersion method, such as Figure 2 As shown, the following steps are included:

[0080] S100, acquiring a target biological image, and processing the target biological image to obtain a target biological region image;

[0081] S200, performing target detection on the target biological region image to obtain a surface target region;

[0082] S300, filling based on the surface target area to obtain a surface target visible area, and then obtain a lower layer blocked area;

[0083] S400, performing target organism detection on the blocked area of ​​the lower layer, and if the outline of the target organism is detected, determining that the target organisms are stacked;

[0084] S500, breaking up the stacked target organisms and recollecting the target organism images until there is no stacking in the target organism images, thereby obtaining a non-stacked image;

[0085] S600: Detect, count and analyze the target organisms in the non-stacked image to obtain statistical analysis results.

[0086] The present invention utilizes the area filling segmentation method and the deep learning algorithm to automatically determine whether there is a target biological stack in the collected image, and automatically triggers a vibration device to break up the stacked target biological insect bodies, thereby greatly improving the detection efficiency; by training the first type detection model and the second type detection model respectively, targeted identification and counting of target organisms of different sizes are achieved, thereby improving the recognition accuracy; the vision-based target biological stack identification and breaking up device has low hardware requirements and has the advantages of low cost and high detection efficiency.

[0087] In this embodiment, step S100 acquires a target biological image, performs grayscale binarization processing on the target biological image, and segments a target biological region image, which specifically includes the following steps:

[0088] S110: Use a camera or other image acquisition device to capture the target biological image on the object platform 15, such as Figure 4 As shown, the Exif information of the target biological image is deleted.

[0089] In the step of collecting the target biological image, the pest identification image can be obtained through the camera component on the scattered insect structure, and the image can also be obtained through an external mobile terminal, wherein the external mobile terminal can be a mobile phone, tablet, etc. with a camera.

[0090] This solution preprocesses the collected target biological image and deletes the Exif information in the target biological image. By calling the remove function of the piexif library, the Exif information can be deleted.

[0091] S120: performing grayscale binarization processing on the target biological image to obtain a binary image.

[0092] Performing grayscale processing on the target biological image to obtain a grayscale image, and performing binarization processing on the grayscale processed image to obtain a binarized image;

[0093] The binarization method is the Otsu threshold method, where the threshold is selected from 0-255, and the maximum inter-class variance is maximized by setting the threshold. The calculation formula of the Otsu threshold method is:

[0094]

[0095] in, represents the maximum between-class variance, represents the percentage of foreground image pixels under the threshold, Indicates the percentage of background image pixels under the threshold, represents the average grayscale of the foreground under the threshold, represents the average grayscale of the background under the threshold, Represents the average grayscale of the target biological image.

[0096] S130: extracting each connected domain in the binary image, and calculating the minimum circumscribed rectangle of each connected domain to segment the target biological region image.

[0097] In step S200, target detection is performed on the target biological region image to obtain a surface target region table.

[0098] S210: Classify the target organisms into a first type (body size greater than 3 mm) and a second type (body size less than or equal to 3 mm), and train the first type detection model and the second type detection model respectively;

[0099] S220: Obtain a training data set of a first type of detection model and a training data set of a second type of detection model;

[0100] S230: directly input the training data set of the first type detection model into the image detection model for training after marking; perform image segmentation and clustering on the training data set of the second type detection model and input it into the image detection model for training to obtain the upper target biological region position, the image detection model includes: multi-layer convolution layer, C3 structure layer (such as Figure 5 as shown), detection layer and SPPF layer.

[0101] The image detection model uses the CIOU loss function for error calibration, and the calculation formula of the CIOU loss function is:

[0102]

[0103] in:

[0104]

[0105]

[0106]

[0107] in, Represents the Euclidean distance between the network prediction box and the real box, Represents the center point coordinates of the network prediction box and the real box, Represents the information of the real box annotation, represents the weight function, It is used to measure the consistency of aspect ratio. Represent the width and height of the real box and the width and height of the predicted box respectively. Indicates the diagonal length of the outer frame of the real box and the predicted box, is the intersection-and-union ratio, is the loss function.

[0108] In step S300, filling is performed based on the surface target area to obtain the surface target visible area, and then the underlying blocked area is obtained, including the following steps:

[0109] S301, based on the surface target area, performing median filtering on the target biological area image to obtain a filtered image;

[0110] S302, performing edge detection on the filtered image based on the filtered image to obtain an image containing edge information;

[0111] S303, extracting all endpoints of the image containing edge information to obtain an endpoint data set, and obtaining the target biological contour based on the endpoint data set;

[0112] S304, extracting the minimum bounding rectangle based on the target biological area image, taking the target biological outline and the pixel points within the minimum bounding rectangle as background sampling points, calculating the pixel values ​​of the background sampling points, performing area filling and segmentation on the target biological outline with the pixel values, segmenting the surface target visible area, and then obtaining the lower layer obscured area.

[0113] The edge detection calculates the gradient size and direction of each pixel point of the filtered image.

[0114] The gradient size is obtained by the following method:

[0115]

[0116] The gradient direction is obtained by:

[0117]

[0118] in, and Represents the gradient components in the x and y directions, where x and y represent the horizontal and vertical directions on the two-dimensional image, respectively. ‌ represents the gradient direction of the current point, ‌Indicates the gradient size of the current point.

[0119] S400, performing target organism detection on the blocked area of ​​the lower layer, and if the outline of the target organism is detected, determining that the target organisms are stacked;

[0120] S500, breaking up the stacked target organisms and recollecting the target organism images until there is no stacking in the target organism images, thereby obtaining a non-stacked image;

[0121] S600: Detect, count and analyze the target organisms in the non-stacked image to obtain statistical analysis results.

[0122] Compared with the traditional pest observation method, the main improvements are as follows: the present invention uses the area filling segmentation method and the deep learning algorithm to automatically determine whether there is a target biological stack in the collected image, and automatically triggers the vibration device to break up the accumulated target biological insect bodies, thereby greatly improving the detection efficiency; by training the first type detection model and the second type detection model respectively, targeted identification and counting of target organisms of different sizes are achieved, thereby improving the recognition accuracy; the vision-based target biological stack identification and breaking up device has low hardware requirements and has the advantages of low cost and high detection efficiency.

[0123] Embodiment 3:

[0124] A vision-based target biological stack recognition and breakup system, such as Figure 3 As shown, it includes a target biological image acquisition module 100, a surface target area acquisition module 200, an image filling and segmentation module 300, a stacking detection module 400, a vibration and dispersion module 500, and a statistical analysis module 600.

[0125] The target biological image acquisition module 100 is used to acquire a target biological image and process the target biological image to obtain a target biological region image;

[0126] A surface target region acquisition module 200 is used to perform target detection on the target biological region image to obtain a surface target region;

[0127] An image filling and segmentation module 300 is used to fill in the surface target area to obtain the surface target visible area, and then obtain the underlying blocked area;

[0128] The stacking detection module 400 is used to detect target organisms in the lower blocked area, and if the outline of the target organisms is detected, it is determined that the target organisms are stacked;

[0129] The stack vibration breaking module 500 is used to break up the stacked target organisms and re-collect the target organism image until there is no stacking in the target organism image, and then obtain a non-stacked image;

[0130] The statistical analysis module 600 is used to detect, count and analyze the target organisms in the non-stacked image to obtain statistical analysis results.

[0131] Various changes and modifications can be made without departing from the spirit and scope of the present invention, and all equivalent technical solutions also belong to the scope of the present invention.

[0132] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0133] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, devices or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0134] The present invention is described with reference to the flowcharts and / or block diagrams of the method, terminal device (system), and computer program product according to the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of the processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing terminal device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing terminal device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0135] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing terminal device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0136] These computer program instructions can also be loaded onto a computer or other programmable data processing terminal device so that a series of operating steps are executed on the computer or other programmable terminal device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable terminal device to implement the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0137] It should be noted that:

[0138] The "one embodiment" or "embodiment" mentioned in the specification means that the specific features, structures or characteristics described in conjunction with the embodiment are included in at least one embodiment of the present invention. Therefore, the phrases "one embodiment" or "embodiment" appearing in various places throughout the specification do not necessarily refer to the same embodiment.

[0139] In addition, it should be noted that the shapes and names of the parts of the specific embodiments described in this specification may be different. Any equivalent or simple changes made based on the structure, features and principles described in the patent concept of the present invention are included in the protection scope of the patent of the present invention. The technicians in the technical field of the present invention can make various modifications or supplements to the specific embodiments described or replace them in a similar manner, as long as they do not deviate from the structure of the present invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A method for identifying and breaking up target biological stacks based on vision, characterized in that: The following steps are involved: Acquire a target biological image, and process the target biological image to obtain a target biological region image; Performing target detection on the target biological region image to obtain a surface target region; Filling is performed based on the surface target area to obtain a surface target visible area, and then obtaining a lower layer blocked area; The following steps are involved: Based on the surface target area, performing median filtering on the target biological area image to obtain a filtered image; Based on the filtered image, edge detection is performed on the filtered image to obtain an image containing edge information; Extracting all endpoints of the image containing edge information to obtain an endpoint data set, and obtaining the target biological contour based on the endpoint data set; Extracting the minimum bounding rectangle based on the target biological region image, taking the target biological contour and the pixel points within the minimum bounding rectangle as background sampling points, calculating the pixel values ​​of the background sampling points, performing regional filling segmentation on the target biological contour with the pixel values, segmenting the surface target visible area, and then obtaining the lower layer blocked area; The edge detection calculates the gradient size and direction of each pixel point of the filtered image. The gradient size is obtained by: The gradient direction is obtained by: in, and Represents the gradient components in the x and y directions, where x and y represent the horizontal and vertical directions on the two-dimensional image, respectively. Indicates the gradient direction of the current point, Indicates the gradient size of the current point; Performing target organism detection on the blocked area of ​​the lower layer, and if the outline of the target organism is detected, it is determined that the target organisms are stacked; The stacked target organisms are broken up and the target organism images are recollected until there is no stacking in the target organism images, and a non-stacked image is obtained; Detect, count and analyze the target organisms in the non-stacked image to obtain statistical analysis results.

2. The method for identifying and breaking up target biological stacks based on vision according to claim 1, characterized in that: The step of acquiring a target biological image and processing the target biological image to obtain a target biological region image comprises the following steps: Performing grayscale binarization processing on the target biological image to obtain a binarized image; Each connected domain in the binary image is extracted, and the minimum circumscribed rectangle of each connected domain is calculated to segment the target biological region image.

3. The method for identifying and breaking up target biological stacks based on vision according to claim 2, characterized in that: The binarization method is the Otsu threshold method, which sets a first preset threshold so that the maximum inter-class variance is maximized. The calculation formula of the Otsu threshold method is: in, represents the maximum between-class variance, represents the proportion of foreground image pixels under the first preset threshold, represents the pixel ratio of the background image under the first preset threshold, represents the average grayscale of the foreground under the first preset threshold, represents the average grayscale of the background under the first preset threshold, Represents the average grayscale of the target biological image.

4. The method for identifying and breaking up target biological stacks based on vision according to claim 1, characterized in that: Performing target detection on the target biological region image through an image detection model, wherein the image detection model at least includes: a multi-layer convolution layer, a C3 structure layer, a detection layer, and an SPPF layer; The image detection model includes a first type detection model and a second type detection model.

5. The method for identifying and breaking up target biological stacks based on vision according to claim 4, characterized in that: The method of performing target detection on the target biological region image to obtain the surface target region includes the following steps: Obtain a first type of training sample data set and a second type of training sample data set, and divide the first type of training sample data set and the second type of training sample data set into a training set, a validation set, and a test set; The first type of training sample data set is marked and directly input into the first type of pre-training model for training; the second type of training sample data set is image-blocked and clustered and then input into the second type of pre-training model for training, the target biological area image is detected, and the surface target area is obtained.

6. The method for identifying and breaking up target biological stacks based on vision according to claim 4, characterized in that: The image detection model uses the CIOU loss function for error calibration, and the calculation formula of the CIOU loss function is: in: in, Represents the Euclidean distance between the network prediction box and the real box, Represents the center point coordinates of the network prediction box and the real box, Represents the information of the real box annotation, represents the weight function, It is used to measure the consistency of aspect ratio. Represent the width and height of the real box and the width and height of the predicted box respectively. Indicates the diagonal length of the outer frame of the real box and the predicted box, is the intersection-and-union ratio, is the loss function.

7. A vision-based target biological stacking identification and disintegration system, characterized in that: It includes target biological image acquisition module, surface target area acquisition module, image filling and segmentation module, stacking detection module, vibration and dispersion module, and statistical analysis module. A target biological image acquisition module, used to acquire a target biological image, process the target biological image, and obtain a target biological region image; A surface target region acquisition module, used for performing target detection on the target biological region image to obtain a surface target region; An image filling and segmentation module is used to fill in the surface target area to obtain the surface target visible area, and then obtain the underlying blocked area; The following steps are involved: Based on the surface target area, performing median filtering on the target biological area image to obtain a filtered image; Based on the filtered image, edge detection is performed on the filtered image to obtain an image containing edge information; Extracting all endpoints of the image containing edge information to obtain an endpoint data set, and obtaining the target biological contour based on the endpoint data set; Extracting the minimum bounding rectangle based on the target biological region image, taking the target biological contour and the pixel points within the minimum bounding rectangle as background sampling points, calculating the pixel values ​​of the background sampling points, performing regional filling segmentation on the target biological contour with the pixel values, segmenting the surface target visible area, and then obtaining the lower layer blocked area; The edge detection calculates the gradient size and direction of each pixel point of the filtered image. The gradient size is obtained by: The gradient direction is obtained by: in, and Represents the gradient components in the x and y directions, where x and y represent the horizontal and vertical directions on the two-dimensional image, respectively. Indicates the gradient direction of the current point, Indicates the gradient size of the current point; A stacking detection module is used to detect target organisms in the lower blocked area, and if the outline of the target organism is detected, it is determined that the target organisms are stacked; A vibration breaking module is used to break up the stacked target organisms and re-collect the target organism image until there is no stacking in the target organism image, and then obtain a non-stacked image; The statistical analysis module is used to detect, count and analyze the target organisms in the non-stacked image to obtain statistical analysis results.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

9. A device for identifying and breaking up target biological stacks based on vision, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 6 is implemented.

Images