A method for estimating genetic parameters of underwater fish

Abstract

The present invention provides a method for estimating genetic parameters of underwater fish, belonging to the field of marine technology. The method comprises: obtaining original image data of the underwater fish; segmenting the enhanced original image of the underwater fish using an improved binarized level set model; obtaining an underwater fish image by eliminating interference from background grids; predicting the weight of the underwater fish using a weight prediction model; and estimating the genetic parameters of the underwater fish using the obtained predicted weight and image traits. By eliminating interference from background grids, the present invention constructs a method for efficiently and accurately assessing the genetic parameters of fish growth traits using underwater fish images, solving the problem of cage grids interfering with underwater fish image segmentation. The method has the advantages of high throughput, automation, ease of operation, and complete non-damage to the fish, providing a theoretical basis and application reference for the development of automated and intelligent analysis technologies for fish phenotypic information.

Description

Technical Field

[0001] The present invention relates to the field of marine technology, and more particularly to a method for estimating genetic parameters of underwater fish. Background Art

[0002] In underwater fish breeding, weight is a crucial economic growth trait, and selecting for heavier fish is one of the key goals. However, in practical aquaculture environments, such as large-scale deep-sea cage aquaculture, accurately measuring the weight of each individual fish is a tedious and difficult task. Therefore, exploring an easily accessible trait that can serve as a surrogate for weight as an economic growth trait for genomic selection breeding is of great significance.

[0003] Currently, developing effective fish sampling techniques based on underwater images or videos, and using computer vision to automatically monitor and estimate fish biomass in water bodies, is an interesting area of ​​research. Image-based measurement methods, in particular, can be used with the help of vision-based robotics to detect and explore organisms in hazardous areas inaccessible to humans. In measuring the growth traits of aquatic species, more accurate image processing methods are gradually replacing traditional manual measurement methods as a non-invasive measurement method. For example, due to the elliptical shape of sea cucumbers, the length, width, circumference, and area of ​​sea cucumbers can be determined by analyzing and processing their surface images using image processing techniques such as edge segmentation. To determine the shape of objects in images, numerous image segmentation techniques have been proposed to detect the boundary contours of objects. Existing segmentation methods include rule tracking, matched filter response, and topology adaptation. These image segmentation methods have good segmentation effects on water images taken during fishing. However, since fishing photography requires catching fish and anesthetizing them for taking photos, there are problems such as cumbersome image data acquisition process, stress and irreversible damage to fish, and incompatibility with the production environment, resulting in low practical application value.

[0004] When the background environment is complex, especially when taking photos or videos underwater, the captured images are greatly affected by the water environment, such as low light, turbid water, interference from the cage grid, and other fish occlusions. As a result, image segmentation techniques that work well above water cannot effectively process underwater fish images. Furthermore, it is even more difficult to use the area of ​​underwater fish images to predict fish weight for fish breeding. However, underwater images have the advantages of simple data acquisition, no contact with the fish, no damage, and are suitable for production environments, with high practical application value. However, existing methods have difficulty in accurately and effectively segmenting the underwater fish body contours. Therefore, there is an urgent need to invent an effective underwater fish image segmentation technology that can predict fish weight based on the area of ​​the fish, which can be effectively applied to underwater fish breeding. Summary of the Invention

[0005] In response to the above problems, the purpose of the present invention is to provide a method for estimating genetic parameters of underwater fish, which is used to solve the problems of cumbersome image data acquisition process, stress and irreversible damage to fish, inconsistency with the production environment, and low practical application value caused by underwater fish capture and shooting in large-scale breeding processes; by shooting underwater fish images underwater, the edge contours of individual underwater fish are segmented using a binary level set image segmentation method, overcoming the difficulty of existing methods in coping with the individual segmentation of underwater fish shot in complex water environments such as dark underwater light, turbid water, and cage grid obstruction. The present invention realizes high-throughput, automated, standardized, non-destructive and accurate underwater phenotypic measurement and real-time feedback of fish, provides accurate phenotypic information for the breeding of fish and other aquatic species, accelerates the accurate breeding of aquatic species, and shortens the breeding cycle.

[0006] To achieve the above and other related purposes, the present invention provides a method for estimating genetic parameters of underwater fish, the method comprising:

[0007] S1 obtains the original image data of underwater fish;

[0008] S2 uses two-dimensional Gaussian kernel matched filtering to enhance the original image of underwater fish;

[0009] S3 uses the improved binary level set model to segment the enhanced original image of underwater fish to obtain a binary level set image of underwater fish;

[0010] S4 processes the binary level set image obtained in S3 through image morphology, merges regions based on similarity measurement, eliminates noise points on the binary level set image, and obtains a clear boundary contour image of the underwater fish;

[0011] S5 eliminates the interference of the background grid and obtains the binary image of the underwater fish;

[0012] S6 predicts the weight of the underwater fish using a weight prediction model based on deep learning based on the binary image of the underwater fish obtained in S5;

[0013] S7 calculates the body area of ​​the underwater fish based on the binary image of the underwater fish obtained in S5, and uses the body area and background percentage as an image trait, and estimates the genetic parameters of the underwater fish using the image trait and the predicted weight obtained in S6.

[0014] The above-mentioned method for estimating genetic parameters of underwater fish also includes S8: using linear model 1, linear model 2 and nonlinear model to estimate the breeding value of underwater fish through the predicted weight and the image trait; the linear model 1 is rr-GBLUP, the linear model 2 is BayesB, and the nonlinear model is RKHS

[0015] In the above-mentioned method for estimating genetic parameters of underwater fish, the improved binary level set model is:

[0016]

[0017] The body regions of underwater fish are: Ω b ; The background area is: Ω; that is

[0018] The piecewise constant approximation u of the binary level set image u0 is constructed as follows:

[0019]

[0020] For the binary level set image, u=c1 in Ω b Inside, u=c2 in Ω b Outside, and there

[0021] Minimize the following Mumford-Shah function and use the improved binary level set model to find the segmentation boundary of the binary level set image:

[0022]

[0023] in

[0024] Improved Binarized Level Set Model The constraint is:

[0025]

[0026] In the above-mentioned method for estimating genetic parameters of underwater fish, the penalty formula used in the improved binary level set model is:

[0027]

[0028]

[0029]

[0030]

[0031] For a given F η The minimum value meets:

[0032]

[0033] in

[0034] For fixed

[0035]

[0036] pass Get the binary level set image of underwater fish.

[0037] In the above-mentioned method for estimating genetic parameters of underwater fish, the step S4 comprises:

[0038] S41 performs an erosion operation on the binary level set image to eliminate noise points around the outer body of the underwater fish, slides the structure element 1 on the binary level set image, and sets the position of the structure element anchor point O to the position of the image pixel point. The grayscale value of the target pixel point is set to the minimum value of the pixels in the corresponding image area in the area where the structure element value is 1;

[0039] The formula is:

[0040] Where element is the structural element 1, (x, y) is the position of the anchor point O, x' and y' are the position offsets of the pixels with the structural element value of 1 relative to the anchor point, src represents the binary level set image, and dst represents the binary level set image after corrosion;

[0041] S42 performs an expansion operation on the binary level set image after corrosion in S41 to make the outline of the underwater fish body smooth and clear, slides the structure element on the binary level set image after corrosion in S41, and makes the position of the second anchor point O' of the structure element correspond to the position of the image pixel point, and sets the gray value of the target pixel point to the maximum value of the pixel in the corresponding image area in the area where the structure element binary value is 1;

[0042] The formula is:

[0043] Where element' is the structuring element 2, (x', y') is the position of the anchor point O', x' and y' are the position offsets of the pixels with the structuring element value of 1 relative to the anchor point O', src' represents the binary level set image after corrosion, and dst' represents the result image after expansion.

[0044] In the above-mentioned method for estimating genetic parameters of underwater fish, S5 comprises:

[0045] In the curve diffusion stage of the S51 underwater fish boundary contour image, pixels with gradient values ​​lower than the upper bound are identified, and neighborhood connectivity groups these pixels and designates them as the internal growth area of ​​the underwater fish;

[0046] S52 identifies the adjacent growth areas inside underwater fish and designates them as potential areas; for each potential area, the seed shared with its largest neighboring pixel is marked as the parent seed of the area, which ensures a high similarity between the parent seed and the potential area; otherwise, it is judged as a non-underwater fish interior area;

[0047] S53 Merges the segmented regions of underwater fish based on similarity measures; Merges regions and their neighbors by generating mean and covariance measures for each region and its adjacent neighbors;

[0048] S54 fishing net grid pattern recognition;

[0049] Comparing the texture content of the inner area of ​​non-underwater fish to judge the mesh of the fishing net;

[0050] Determine whether a pixel is a bifurcation point based on the binarized texture skeleton and the eight-neighborhood pixel values ​​of each pixel;

[0051] Each bifurcation point will be used as a seed and will continue to spread outward along the direction of the skeleton until its adjacent bifurcation point is found. Each bifurcation point will find its adjacent bifurcation point. If there is a connection between the bifurcation points, it will be judged as a fishing net grid structure. The fishing net grid will be removed to obtain the underwater fish binary image.

[0052] In the above method for estimating genetic parameters of underwater fish, in S54: if three values ​​in the eight neighborhoods are 1, the pixel is a trifurcation point.

[0053] As described above, the present invention has the following beneficial effects:

[0054] (1) The present invention collects images of underwater fish in their natural state and realizes end-to-end processing, thus avoiding the stimulation and damage caused to underwater fish by traditional fishing photography, and realizes high-throughput and automated data acquisition, which fits the actual application scenario and has extremely high application value.

[0055] (2) The boundary contour image of underwater fish in the present invention adopts an improved binary level set model to realize the motion target detection of underwater fish and effectively obtain a closed and complete motion contour of underwater fish.

[0056] (3) The present invention uses image morphological processing, corrosion and dilation, and similarity measurement region merging operations to eliminate the interference of fishing net grid noise outside the underwater fish body in the underwater fish image, further clarifying the boundary contour image of the underwater fish, which is beneficial to the final image segmentation.

[0057] (4) The present invention eliminates the interference of background grids, solves the problem of cage grids interfering with underwater fish image segmentation, and has the advantages of high throughput, automation, easy operation, and no damage to fish.

[0058] (5) The control experiment shows that the conventional threshold segmentation method can obtain the underwater segmentation image of fish, but cannot remove the interference of the fishing net grid. The method of the present invention can effectively eliminate the grid interference when obtaining the underwater segmentation image of fish. BRIEF DESCRIPTION OF THE DRAWINGS

[0059] Figure 1 is a flow chart of a method according to an embodiment of the present invention;

[0060] Figure 2 The original image of underwater fish in the embodiment of the present invention;

[0061] Figure 3 is a boundary outline image of an underwater fish in an embodiment of the present invention;

[0062] Figure 4 This is a binary image of underwater fish in an embodiment of the present invention. DETAILED DESCRIPTION

[0063] The specific embodiments of the present invention are described in further detail below in conjunction with the accompanying drawings and examples. The following uses the aquatic animal, grouper, as an example to illustrate the present invention, but is not intended to limit the scope of the present invention.

[0064] Example 1

[0065] like Figure 1 As shown, a method for estimating genetic parameters of underwater fish includes:

[0066] S1 obtains the original image data of underwater fish;

[0067] S2 uses two-dimensional Gaussian kernel matched filtering to enhance the original image of underwater fish;

[0068] The two-dimensional Gaussian function has rotational symmetry, which means that the degree of smoothing of the filter is the same in all directions. Generally speaking, the direction of the texture of an image is unknown, so it is impossible to determine whether more smoothing is required in one direction than in another before filtering. Rotational symmetry means that the Gaussian smoothing filter will not be biased towards any direction in subsequent edge detection. The two-dimensional Gaussian kernel function can be expressed as:

[0069] k(||x-xc||)=exp{-||x-xc 2 / (2*σ) 2 )}

[0070] Where x is the image pixel, xc is the center of the kernel function, and σ is the width parameter of the function, which controls the radial range of the Gaussian kernel function.

[0071] S3 uses the improved binary level set model to segment the enhanced original image of underwater fish to obtain a binary level set image of underwater fish;

[0072] The moving target detection of fish under the conditions of underwater camera shooting is realized, and the closed and complete fish movement outline is effectively obtained.

[0073] S4 processes the binary level set image obtained in S3 through image morphology, merges regions based on similarity measurement, eliminates noise points on the binary level set image, and obtains a clear boundary contour image of the underwater fish;

[0074] In step S4, image morphology processing and similarity measurement region merging are performed to eliminate noise on the underwater fish image after segmentation in S3, and further clarify the boundary contour image of the underwater fish.

[0075] S5 eliminates the interference of the background grid and obtains the binary image of the underwater fish;

[0076] S6 predicts the weight of the underwater fish using a weight prediction model based on deep learning based on the binary image of the underwater fish obtained in S5;

[0077] An underwater fish binary image dataset is obtained, wherein the underwater fish binary image dataset includes: underwater fish binary images and real weights; underwater fish binary images are obtained, the real weights in the underwater fish binary images are labeled, and the underwater fish binary images with labeled weights are used as datasets for a weight prediction model.

[0078] Build a deep learning weight prediction model and train the weight prediction model;

[0079] Input the binary image of underwater fish into the trained weight prediction model to get the predicted weight;

[0080] S7 calculates the body area of ​​the underwater fish based on the binary image of the underwater fish obtained by S5, and uses the body area and background percentage as an image trait, and estimates the genetic parameters of the underwater fish using the image trait and the predicted weight obtained by S6.

[0081] Among them, it also includes S8: using linear model 1, linear model 2 and nonlinear model to estimate the breeding value of underwater fish through the predicted weight and the image traits; the linear model 1 is rr-GBLUP, the linear model 2 is BayesB, and the nonlinear model is RKHS.

[0082] Among them, the improved binary level set model is:

[0083]

[0084] The body regions of underwater fish are: Ω b ; The background area is: Ω; that is

[0085] The piecewise constant approximation u of the binary level set image u0 is constructed as follows:

[0086]

[0087] For the binary level set image, u=c1 in Ω b Inside, u=c2 in Ω b Outside, and there

[0088] Minimize the following Mumford-Shah function and use the improved binary level set model to find the segmentation boundary of the binary level set image:

[0089]

[0090] in

[0091] Improved Binarized Level Set Model The constraint is:

[0092]

[0093] Among them, the penalty formula used in the improved binary level set model is:

[0094]

[0095]

[0096]

[0097]

[0098] For a given F η The minimum value meets:

[0099]

[0100] in

[0101] For fixed

[0102]

[0103] pass Get the binary level set image of underwater fish.

[0104] Among them, S4 includes:

[0105] S41 performs an erosion operation on the binary level set image to eliminate noise points around the outer body of the underwater fish, slides the structure element 1 on the binary level set image, and sets the position of the structure element anchor point O to the position of the image pixel point. The grayscale value of the target pixel point is set to the minimum value of the pixels in the corresponding image area in the area where the structure element value is 1;

[0106] The formula is:

[0107] Where element is the structural element 1, (x, y) is the position of the anchor point O, x' and y' are the position offsets of the pixels with the structural element value of 1 relative to the anchor point, src represents the binary level set image, and dst represents the binary level set image after corrosion;

[0108] S42 performs an expansion operation on the binary level set image after corrosion in S41 to make the outline of the underwater fish body smooth and clear, slides the structure element on the binary level set image after corrosion in S41, and makes the position of the second anchor point O' of the structure element correspond to the position of the image pixel point, and sets the gray value of the target pixel point to the maximum value of the pixel in the corresponding image area in the area where the structure element binary value is 1;

[0109] The formula is:

[0110] Where element' is the structural element 2, (x', y') is the position of the anchor point O', x' and y' are the position offsets of pixels with structural element value 1 relative to the anchor point O', src' represents the binary level set image after corrosion, and dst' represents the boundary contour image of the underwater fish after expansion.

[0111] Among them, S5 includes:

[0112] In the curve diffusion stage of the S51 underwater fish boundary contour image, pixels with gradient values ​​lower than the upper bound are identified, and neighborhood connectivity groups these pixels and designates them as the internal growth area of ​​the underwater fish;

[0113] S52 identifies the adjacent growth areas inside underwater fish and designates them as potential areas; for each potential area, the seed shared with its largest neighboring pixel is marked as the parent seed of the area, which ensures a high similarity between the parent seed and the potential area; otherwise, it is judged as a non-underwater fish interior area;

[0114] S53 Merges the segmented regions of underwater fish based on similarity measures; Merges regions and their neighbors by generating mean and covariance measures for each region and its adjacent neighbors;

[0115] S54 fishing net grid pattern recognition;

[0116] Comparing the texture content of the inner area of ​​non-underwater fish to judge the mesh of the fishing net;

[0117] Determine whether a pixel is a bifurcation point based on the binarized texture skeleton and the eight-neighborhood pixel values ​​of each pixel;

[0118] Each bifurcation point will be used as a seed and will continue to spread outward along the direction of the skeleton until its adjacent bifurcation point is found. Each bifurcation point will find its adjacent bifurcation point. If there is a connection between the bifurcation points, it will be judged as a fishing net grid structure. The fishing net grid will be removed to obtain the underwater fish binary image.

[0119] S55 obtains the final segmentation image of the eastern starfish individual by constructing a fish body mask;

[0120] The binary image of underwater fish obtained by S54 is used as a mask, and is overlaid on the original image of underwater fish enhanced by S2. The masked part is displayed normally to obtain a color image of underwater fish.

[0121] Among them, in S54: if three values ​​in the eight neighborhoods are 1, the pixel is a trifurcation point.

[0122] The present invention eliminates the interference of background grids, solves the problem of cage grids interfering with underwater fish image segmentation, and has the advantages of high throughput, automation, simple operation and no damage to fish.

[0123] Example 2

[0124] The specific embodiments of the present invention are described in further detail below in conjunction with the accompanying drawings and examples. The following uses the aquatic animal, grouper, as an example to illustrate the present invention, but is not intended to limit the scope of the present invention.

[0125] The eastern grouper (Platyctolagus) is highly sought after for its high nutritional and ornamental value, resulting in a high economic value in the international market. Furthermore, the grouper participates in the regulation of homeostasis and energy flow within marine ecosystems, playing a crucial role in maintaining their stability and health. However, the grouper industry has faced significant challenges in recent years, including devastating diseases and environmental stresses, resulting in substantial economic losses and hindering its sustainable development. To overcome these production challenges, molecular breeding methods are urgently needed. Weight, a key economic trait and a crucial indicator for improving grouper breeding, is a critical factor in large-scale deep-sea cage aquaculture. However, weight measurement is largely unavailable in large-scale deep-sea cage aquaculture. Therefore, a simple trait that can be used as a surrogate for weight is urgently needed for molecular breeding. To overcome these challenges, an underwater grouper image segmentation method based on a binary level set model was used to segment underwater grouper images. Fish body area and background percentage (IBAP) was proposed as an image trait, which was used to estimate the breeding value of grouper using this image trait in place of weight.

[0126] like Figure 1-4 As shown, a method for estimating genetic parameters of Eastern star grouper: the method comprises:

[0127] S1 uses an underwater camera to obtain the original image data of the eastern starfish;

[0128] S2 uses two-dimensional Gaussian kernel matched filtering to enhance the original image of the eastern star spot;

[0129] S3 uses the improved binary level set model to segment the enhanced original image of the eastern star spot and obtains the binary level set image of the eastern star spot;

[0130] The moving target detection of fish under the conditions of underwater camera shooting is realized, and the closed and complete motion outline of the grouper is effectively obtained.

[0131] S4 processes the binary level set image obtained in S3 through image morphology, merges regions based on similarity measurement, eliminates noise points on the binary level set image, and obtains a clear boundary contour image of the eastern star spot;

[0132] In step S4, image morphology processing and similarity measurement region merging are performed to eliminate the noise on the eastern star spot image after segmentation in step S3, and further clarify the boundary contour image of the eastern star spot.

[0133] S5 eliminates the interference of the background grid and obtains the binary image of the eastern star spot;

[0134] S6 predicts the weight of the grouper using a weight prediction model based on deep learning based on the binary image of the grouper obtained in S5;

[0135] Obtaining a binary image dataset of grouper, wherein the binary image dataset of grouper includes: a binary image of grouper and a real weight; obtaining the binary image of grouper, annotating the real weight in the binary image of grouper, and using the binary image of grouper with the annotated weight as a dataset of a weight prediction model;

[0136] Build a deep learning weight prediction model and train the weight prediction model;

[0137] Input the binary image of the eastern star spot into the trained weight prediction model to obtain the predicted weight;

[0138] S7 calculates the body area of ​​the eastern star spot based on the binary image of the eastern star spot obtained in S5, and uses the body area and background percentage as an image trait, and estimates the genetic parameters of the eastern star spot using the predicted weight and image traits obtained in S6.

[0139] Among them, S8 is also included: using linear model 1, linear model 2 and nonlinear model to estimate the breeding value of Oriental star spot by predicting weight and image traits; linear model 1 is rr-GBLUP, linear model 2 is BayesB, and nonlinear model is RKHS.

[0140] Genetic parameters of Oriental star grouper were estimated using image traits and body weight, and the heritability of the two traits was calculated using GCTA (version 1.94). The heritability of the image trait was 0.511 (SE 0.08) and the heritability of body weight was 0.539 (SE 0.05), both of which were moderate heritabilities, indicating that the two phenotypes were significantly affected by genetics and had potential for molecular breeding. The phenotypic correlation coefficient and genetic correlation coefficient of the two traits were 0.811 (SE 0.001) and 0.896 (SE 0.021), respectively. This shows that the image trait has a high phenotypic and genetic correlation with body weight, indicating that the image trait and the growth trait body weight are two relatively dependent traits that can be improved simultaneously. The image trait can be used to replace the body weight trait for breeding value estimation.

[0141] Breeding values ​​for the two traits were estimated using two linear models, rr-GBLUP and Bayes B, and one nonlinear model, RKHS. The average prediction accuracy for these two traits for rr-GBLUP, Bayes B, and RKHS was above 0.6. Based on the optimal GS model, the maximum average prediction accuracy for IBAP (RKHS) and BW (RKHS) was 0.683 and 0.699, respectively. Overall, RKHS outperformed other methods, rr-GBLUP and Bayes B, for both IBAP and BW. Furthermore, using different numbers of SNPs for breeding value estimation showed that prediction accuracy remained stable up to 10,000 SNPs, indicating that using 10,000 SNPs for whole-genome selection breeding is a cost-effective option.

[0142] This embodiment avoids the direct measurement of weight, an economic growth trait of grouper, and uses image traits instead of weight for genetic parameter analysis and breeding value estimation, thereby solving the problem of difficult weight measurement in large-scale breeding processes, simplifying the steps and processes for obtaining target traits, reducing experimental costs, and facilitating the efficient genome selection and breeding of grouper and protection of germplasm resources.

Claims

1. A method for estimating genetic parameters of underwater fish, characterized by: The method includes: S1 obtains the original image data of underwater fish; S2 uses two-dimensional Gaussian kernel matched filtering to enhance the original image of underwater fish; S3 uses the improved binary level set model to segment the enhanced original image of underwater fish to obtain a binary level set image of underwater fish; S4 processes the binary level set image obtained in S3 through image morphology, merges regions based on similarity measurement, eliminates noise points on the binary level set image, and obtains a clear boundary contour image of the underwater fish; S5 eliminates the interference of the background grid and obtains the binary image of the underwater fish; S6 predicts the weight of the underwater fish using a weight prediction model based on deep learning based on the binary image of the underwater fish obtained in S5; S7 calculates the body area of ​​the underwater fish based on the binary image of the underwater fish obtained in S5, and uses the body area and background percentage as an image trait, and estimates genetic parameters of the underwater fish using the image trait and the predicted weight obtained in S6; The improved binary level set model is: The body regions of underwater fish are: Ω b ; The background area is: Ω; that is The piecewise constant approximation u of the binary level set image u0 is constructed as follows: For the binary level set image, u=c1 in Ω b Inside, u=c2 in Ω b Outside, and there Minimize the following Mumford-Shah function and use the improved binary level set model to find the segmentation boundary of the binary level set image: in Improved Binarized Level Set Model The constraint is: The penalty formula used in the improved binary level set model is: For a given F η The minimum value meets: in For fixed pass Get the binary level set image of underwater fish; It also includes S8: using linear model 1, linear model 2 and nonlinear model to estimate the breeding value of underwater fish through the predicted weight and the image traits; the linear model 1 is rr-GBLUP, the linear model 2 is BayesB, and the nonlinear model is RKHS.

2. The method for estimating genetic parameters of underwater fish according to claim 1, wherein: The S4 includes: S41 performs an erosion operation on the binary level set image to eliminate noise points around the outer body of the underwater fish, slides the structure element 1 on the binary level set image, and sets the position of the structure element anchor point O to the position of the image pixel point. The grayscale value of the target pixel point is set to the minimum value of the pixels in the corresponding image area in the area where the structure element value is 1; The formula is: Where element is the structural element 1, (x, y) is the position of the anchor point O, x' and y' are the position offsets of the pixels with the structural element value of 1 relative to the anchor point, src represents the binary level set image, and dst represents the binary level set image after corrosion; S42 performs an expansion operation on the binary level set image after corrosion in S41 to make the outline of the underwater fish body smooth and clear, slides the structure element on the binary level set image after corrosion in S41, and makes the position of the second anchor point O' of the structure element correspond to the position of the image pixel point, and sets the gray value of the target pixel point to the maximum value of the pixel in the corresponding image area in the area where the structure element binary value is 1; The formula is: Where element' is the structural element 2, (x', y') is the position of the anchor point O', x' and y' are the position offsets of pixels with structural element value 1 relative to the anchor point O', src' represents the binary level set image after corrosion, and dst' represents the boundary contour image of the underwater fish after expansion.

3. The method for estimating genetic parameters of underwater fish according to claim 1, wherein: The S5 includes: In the curve diffusion stage of the S51 underwater fish boundary contour image, pixels with gradient values ​​lower than the upper bound are identified, and neighborhood connectivity groups these pixels and designates them as the internal growth area of ​​the underwater fish; S52 identifies the adjacent growth areas inside underwater fish and designates them as potential areas; for each potential area, the seed shared with its largest neighboring pixel is marked as the parent seed of the area, which ensures a high similarity between the parent seed and the potential area; otherwise, it is judged as a non-underwater fish interior area; S53 Merges the segmented regions of underwater fish based on similarity measures; Merges regions and their neighbors by generating mean and covariance measures for each region and its adjacent neighbors; S54 fishing net grid pattern recognition; The texture content of the non-underwater fish interior area is compared to determine the fishing net grid; the fishing net grid is removed to obtain the underwater fish binary image.

4. The method for estimating genetic parameters of underwater fish according to claim 3, wherein: The method for judging the fishnet grid in S54 is: determining whether the pixel is a bifurcation point based on the binary texture skeleton and the eight-neighborhood pixel values ​​of each pixel; Each bifurcation point will serve as a seed and continue to spread outward along the direction of the skeleton until its adjacent bifurcation point is found; each bifurcation point finds its adjacent bifurcation point, and if there is a connection between the bifurcation points, it is judged to be a fishing net grid structure.

5. The method for estimating genetic parameters of underwater fish according to claim 4, wherein: The method to determine whether a pixel is a bifurcation point is that if three values ​​in the eight neighborhoods are 1, the pixel is a three-way bifurcation point.

Images