Freshwater fish phenotype data non-contact measurement device and method based on machine vision

Abstract

The invention provides a freshwater fish phenotype data non-contact measurement device and method based on machine vision. The device comprises a fishing module, a vision module, a power supply module and a control module. The underwater camera and the industrial camera are used for obtaining fish school image data and uploading the fish school image data to the upper computer for image processing, the upper computer introduces non-linear combination and interaction items into the SVR model, in order to enable the SVR model to be adjusted according to different fish feature importance, different weight coefficients are introduced for each morphological feature, and therefore the SVR model can be adjusted according to different fish feature importance. The coefficients are optimized through an SOA algorithm, and finally accurate measurement of fish phenotype data is achieved. According to the invention, non-contact measurement of the phenotype data of the freshwater fish can be realized in a natural environment, the measurement difficulty is reduced while the measurement precision is improved, the damage to the fish body is reduced, the mechanization and intelligence level of fish phenotype measurement is improved, and a culturist is helped to optimize management decisions.

Description

Technical Field

[0001] The invention belongs to the technical field of freshwater fish phenotypic data measurement under natural environment, and in particular relates to a non-contact measurement device and method for freshwater fish phenotypic data based on machine vision. Background Art

[0002] my country is the world's largest aquaculture country and the only major fishing country in the world where the total amount of farmed aquatic products exceeds the total amount of fishing; freshwater fish has high nutritional value, delicious meat, rich in high-quality protein, multiple vitamins and trace elements, and is a good tonic for preventing diseases and improving human immunity. Although freshwater fish farming has developed rapidly in recent years, most of them are still based on experience, with a low level of scientific farming, and there are also problems in the industry that cannot be ignored.

[0003] Although there are some intelligent fish phenotypic data measurement devices to assist scientific breeding, the existing intelligent fish phenotypic data measurement devices mainly rely on use in laboratories or cage breeding environments. These methods usually require the fish to be caught out of the water for measurement, which can easily cause physical damage and stress reactions to the fish. In addition, these devices are more effective in relatively ideal environments, but in natural waters, especially turbid waters, they are often limited by factors such as light and water quality, resulting in reduced measurement accuracy. Many devices rely on manual operation, the operation process is cumbersome, and real-time monitoring cannot be achieved, which greatly limits their application in large-scale aquaculture.

[0004] Prior art 1 (CN111696150A) discloses a method for measuring the phenotypic data of forktail catfish, which takes fish images from multiple angles, annotates key points, and uses a deep learning algorithm based on the Hourglass network in combination with three-dimensional space transformation to obtain the phenotypic data of the fish. This method is easily affected by the posture of the fish and environmental factors during the image processing process, resulting in errors in the positioning of key points and limited measurement accuracy.

[0005] Prior art 2 (CN112669348B) discloses a method and device for estimating fish posture and measuring fish phenotype data, which uses the Resnext101-ssd300 network combined with the feature pyramid FPN to detect the target of the fish image, detects the key points of the fish through the Hourglassnet deep learning network model, and combines the three-dimensional space transformation to estimate the fish posture and calculate the phenotype data. Although target detection and key point positioning improve the measurement accuracy, the detection effect will be limited under different environmental conditions, such as light, water quality, etc.; in addition, the calculation cost of complex network models is relatively high.

[0006] Prior art 3 (CN117876456A) discloses a non-contact fish weight measurement method, device, electronic device and medium, which generates the three-dimensional surface area of ​​the fish body based on the Mask R-CNN deep learning algorithm, combined with the registration of color images and depth images, and uses the weight fitting model to predict the weight of the fish body, thereby realizing non-contact weight measurement. Due to environmental conditions, the quality of the depth image and the generated point cloud data may be inaccurate, affecting the accuracy of weight prediction; at the same time, this method relies on high-quality image acquisition equipment.

[0007] In summary, the above-mentioned prior art has problems such as unstable measurement accuracy, poor environmental adaptability, and high computing resource usage. To this end, the present invention designs a non-contact measurement device and method for freshwater fish phenotypic data based on machine vision, aiming to optimize the collection and analysis process of fish phenotypic data by improving machine vision technology and algorithms, so as to better adapt to the measurement needs in the natural environment, improve accuracy and robustness, and reduce dependence on environmental conditions, so as to achieve intelligent, non-destructive, and efficient phenotypic data measurement. Summary of the invention

[0008] In view of the deficiencies in the prior art, the present invention provides a non-contact measurement device and method for freshwater fish phenotypic data based on machine vision, which can efficiently and non-destructively measure the phenotypic data of freshwater fish in a natural environment.

[0009] The present invention achieves the above technical objectives through the following technical means.

[0010] A non-contact measurement device for freshwater fish phenotypic data based on machine vision, comprising a fishing module, a vision module, a power supply module, and a control module; the fishing module comprises a frame with a float installed, an aluminum alloy plate is installed on the top of the frame, a crane is installed under the aluminum alloy plate, the crane is connected to a steel wire rope, the other end of the steel wire rope is connected to a tension sensor, the tension sensor is connected to a fish cage through the steel wire rope, and a water level sensor is installed under the fish cage; the vision module comprises an underwater camera installed on the inside of the fish cage and an industrial camera installed under the aluminum alloy plate; the control module comprises a control cabinet installed on the fishing module and a weighing transmitter, a main control board, and a battery in the control cabinet, and also comprises an outdoor router and a host computer.

[0011] The non-contact measurement method of freshwater fish phenotypic data based on machine vision using the above-mentioned non-contact measurement device of freshwater fish phenotypic data based on machine vision includes the following process:

[0012] Step 1: Place the fishing device within the feeding radius of freshwater fish. When the freshwater fish swim into the fish cage, the water level sensor senses the water level change and transmits a signal to the main control board, triggering the underwater camera and industrial camera to take pictures, and obtain the top view and side view of the fish school;

[0013] Step 2: After a delay, the fishing module pulls up the fish cage through the crane, wire rope and tension sensor to weigh the freshwater fish; the tension sensor senses the weight change and outputs an analog voltage, and the weighing transmitter reads the voltage signal and converts the weighing data and transmits it to the main control board;

[0014] Step 3: The main control board transmits the collected image data and weight data to the host computer through the outdoor router. The host computer processes the image and obtains the predicted phenotypic data of each fish.

[0015] Furthermore, the specific process of step 3 is as follows:

[0016] Step 3.1: grayscale conversion and filtering are performed on the side view and top view of freshwater fish captured by the underwater camera and the industrial camera;

[0017] Step 3.2: Use Canny edge detection to identify and separate fish body pixels from background pixels;

[0018] Step 3.3: Post-process the segmented results to obtain accurate fish body boundaries;

[0019] Step 3.4: extracting the area, length, width and thickness data from the morphological features of the processed image;

[0020] Step 3.5: Measure phenotypic data through SOA-SVR algorithm: Introduce nonlinear combinations and interaction terms related to morphological features into the SVR regression model, introduce different weight coefficients for each morphological feature, optimize the weight coefficients through SOA, and use the optimized and improved SVR regression model to predict phenotypic data.

[0021] Furthermore, the specific process of step 3.5 is as follows:

[0022] Step 3.5.1: Assign a weight coefficient to each morphological feature in the SVR regression model. The SVR regression function is:

[0023] y i * =α 1 ·A i +α 2 ·L i +α 3 ·W i +α 4 ·T i +b

[0024] Step 3.5.2: Introduce the interaction terms between features, and the improved SVR regression model formula is:

[0025]

[0026] Step 3.5.3: Based on the improved regression model, adjust the objective function of SVR. The improved SVR objective function is:

[0027]

[0028] Step 3.5.4: Use SOA to optimize the morphological feature weighting coefficient α in SVR j , feature interaction weight coefficient β k As well as the penalty factor C and kernel function parameter θ of SVR;

[0029] Step 3.5.5: Initialize the seagull population, each seagull individual represents a parameter combination;

[0030] Step 3.5.6: Calculate the fitness value F of each seagull individual:

[0031]

[0032] Step 3.5.7: Use the linear flight and spiral flight formulas of SOA to gradually optimize the parameter combination;

[0033] Step 3.5.8: Determine whether the number of iterations has been reached. If so, record the optimal solution. Otherwise, return to step 3.5.7.

[0034] Step 3.5.9: Find the optimal parameter combination through SOA;

[0035] Step 3.5.10: Input the optimal parameters into the improved SVR regression model to calculate the predicted phenotypic data for each fish.

[0036] The present invention has the following beneficial effects:

[0037] The present invention can be directly deployed in natural waters, and can realize efficient non-destructive detection of freshwater fish phenotypic data in natural environments. Even in turbid environments, the accuracy and real-time performance of the measurements can be guaranteed. The fully automated operation does not require human intervention, and the mechanization and intelligence level of fish phenotypic measurement is improved. A large amount of manpower and material resources can be saved, and the investment in labor costs can be reduced, helping breeders optimize management decisions and improve production efficiency and economic benefits. The device structure of the present invention adopts a lightweight design, has a simple and reasonable structure, is easy to use and maintain, and is easy to promote. It can play a guiding and improving role in freshwater fish phenotypic measurement and growth monitoring, and has a high social and economic value.

[0038] The present invention introduces the association between morphological features: there is a potential nonlinear relationship between area, body length, body width and body thickness; simple linear relationship may not be able to fully capture the mutual influence of these features, therefore, nonlinear combination and interaction terms are introduced into the SVR model; the present invention optimizes feature weights: in order to enable the SVR model to be adjusted according to the importance of different features, different weight coefficients are introduced for each morphological feature, and these coefficients are optimized by the Seagull optimization algorithm. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a three-dimensional schematic diagram of the non-contact measurement device for freshwater fish phenotypic data based on machine vision according to the present invention;

[0040] Figure 2 It is a front schematic diagram of the non-contact measurement device for freshwater fish phenotypic data based on machine vision according to the present invention;

[0041] Figure 3 is a schematic diagram of the visual module of the present invention;

[0042] Figure 4 is a schematic diagram of the power supply module of the present invention;

[0043] Figure 5 is a schematic diagram of the control module of the present invention;

[0044] Figure 6 It is a flowchart of the image processing process;

[0045] Figure 7 It is the flow chart of Seagull optimized support vector machine regression algorithm.

[0046] In the figure: fishing module 1, vision module 2, power supply module 3, control module 4, float 21, pipe joint 22, aluminum alloy plate 23, aluminum alloy pipe 24, crane 25, wire rope 26, tension sensor 27, fish cage 28, water level sensor 29, underwater camera 31, camera support frame 32, fixer 33, camera bracket 34, industrial camera 35, solar panel 41, support frame 42, weighing transmitter 51, Raspberry Pi 4B main control board 52, outdoor router 53, host computer 54, control cabinet 55, battery 56. DETAILED DESCRIPTION

[0047] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments, but the protection scope of the present invention is not limited thereto.

[0048] like Figure 1 As shown, the non-contact measurement device for freshwater fish phenotypic data based on machine vision of the present invention includes a fishing module 1, a vision module 2, a power supply module 3, and a control module 4.

[0049] like Figure 1 , 2 As shown, the fishing module 1 is a fishing device placed in the aquaculture waters, including a float 21, a pipe joint 22, an aluminum alloy plate 23, an aluminum alloy pipe 24, a crane 25, a steel wire rope 26, a tension sensor 27, a fish cage 28, and a water level sensor 29. A plurality of aluminum alloy pipes 24 are mutually combined to form a frame through the pipe joint 22, and the float 21 is installed at the four corners of the lower part of the frame to achieve buoyancy support and ensure the overall stability of the device; an aluminum alloy plate 23 is also installed on the top of the frame, and a crane 25 is installed at the lower part of the aluminum alloy plate 23. The crane 25 is connected to the steel wire rope 26, and the other end of the steel wire rope 26 is connected to the tension sensor 27, and the tension sensor 27 is connected to the fish cage 28 through the steel wire rope 26. The tension sensor 27 can be used to assist in calculating the average weight, and the body weight can be calculated through the obtained phenotypic data later, and the accuracy of monitoring can be verified from the side, and the growth status of the fish can also be observed through the weight change; the water level sensor 29 is installed below the fish cage 28 to measure the water level change.

[0050] like Figure 1 , 3 As shown, the visual module 2 includes an underwater camera 31, a camera support frame 32, a fixer 33, a camera bracket 34, and an industrial camera 35; the underwater camera 31 is installed on the inner side of the fish cage 28 through the camera support frame 32, and is used to shoot a side view of freshwater fish; the industrial camera 35 is installed on the camera bracket 34, and the camera bracket 34 is installed under the aluminum alloy plate 23 through the fixer 33, and is used to shoot a top view of freshwater fish.

[0051] like Figure 1 , 4 As shown, the power supply module 3 is installed above the fishing module 1, specifically on the upper surface of the aluminum alloy plate 23, and includes a solar panel 41 and a support frame 42; the solar panel 41 is installed on the support frame 42 and fixed at an angle of 30 degrees to maximize the reception of sunlight and energy conversion efficiency throughout the year.

[0052] like Figure 1 , 5 As shown, the control module 4 includes a control cabinet 55 installed on the aluminum alloy tube 24 on the side of the fishing module 1, a weighing transmitter 51, a Raspberry Pi 4B main control board 52, and a battery 56 in the control cabinet 55, and also includes an outdoor router 53 and a host computer 54. The weighing transmitter 51 is used to convert the voltage signal into a 485 signal and transmit it to the Raspberry Pi 4B main control board 52; the Raspberry Pi 4B main control board 52 is used to receive and send signals; the outdoor router 53 and the host computer 54 are installed on the shore to reduce the weight of the device; the battery 56 is used to store electricity.

[0053] The measuring method using the non-contact measuring device for freshwater fish phenotype data based on machine vision of the present invention comprises the following process:

[0054] Step 1: Place the fishing device 1 within the feeding radius of freshwater fish, and use the visual module 2 to capture the image of freshwater fish in the fish cage 28. When the freshwater fish swim into the fish cage 28, the water level sensor 29 senses the water level change and transmits a 485 signal to the Raspberry Pi4B main control board 52, triggering the underwater camera 31 and the industrial camera 35 to take pictures, and obtain the top view and side view of the fish school;

[0055] Step 2: After a delay, the fishing module 1 pulls up the fish cage 28 through the crane 25, the wire rope 26 and the tension sensor 27 to weigh the freshwater fish; the tension sensor 27 senses the weight change and outputs an analog voltage, and the weighing transmitter 51 reads the voltage signal and converts the weighing data into an RS485 signal with high anti-interference performance and transmits it to the Raspberry Pi 4B main control board 52;

[0056] Step 3: The Raspberry Pi 4B main control board 52 transmits the collected image data and weight data to the host computer 54 through the outdoor router 53, and the host computer 54 performs image processing; Figure 6 As shown, the specific process is as follows:

[0057] Step 3.1: grayscale conversion and filtering are performed on the side view and top view of the freshwater fish captured by the underwater camera 31 and the industrial camera 35;

[0058] Step 3.2: Use Canny edge detection to help identify and separate fish pixels from background pixels;

[0059] Step 3.3: Post-process the segmented results, including removing small segmented objects, filling holes, smoothing boundaries, etc., to obtain a more accurate fish body boundary;

[0060] Step 3.4: extracting the area, body length, body width, and body thickness from the morphological features of the processed image;

[0061] Step 3.5: The phenotypic data is measured by the seagull optimized support vector machine regression (SOA-SVR) algorithm; the SOA-SVR algorithm constructs an optimal hyperplane in the feature space, maps the input sample to the high-dimensional feature space, and establishes a linear regression model to more accurately predict the value of the target variable by minimizing the error between the predicted value and the true value and controlling it within a certain range; the Seagull optimization algorithm (SOA) simulates the foraging behavior of seagulls, optimizes the key parameters and in the support vector machine regression (SVR), improves the prediction accuracy of the support vector machine model, and makes the measurement of phenotypic data (such as fish body area, body length, body width, body thickness, etc.) more accurate; Figure 7 As shown, the specific process is as follows:

[0062] Step 3.5.1: Morphological characteristics (area, body length, body width and body thickness) may have different effects on different phenotypic data. In the SVR regression model, a weighted coefficient is assigned to each morphological characteristic to reflect its contribution to the prediction target. The SVR regression function is:

[0063] y i * =α 1 ·A i +α 2 ·L i +α 3 ·W i +α 4 ·T i +b

[0064] In the formula, α 1 , α 1 , α 3 , α 4 is the weighted coefficient of the morphological characteristics, reflecting the influence of each characteristic on the phenotypic data; b is the bias term of the regression model; y i * is the predicted phenotypic data; A i , L i , W i , T i They represent the fish body area, length, width and thickness respectively.

[0065] Step 3.5.2: In practice, there may be nonlinear relationships between morphological features. For example, the ratio between the length and width of a fish may have an important influence on the predicted volume. In order to capture this nonlinear relationship, the interaction terms between features are introduced. The improved SVR regression model formula is:

[0066]

[0067] Among them, β 1 , β 2 , β 3 is the weight coefficient of the feature interaction term, which is used to represent the interaction between morphological features; A i ·L i , L i ·W i and W i ·T i is the feature interaction term, which is used to capture the complex relationship between morphological features; Represents the final predicted phenotypic data results.

[0068] Step 3.5.3: Based on the improved regression model, adjust the objective function of SVR to reflect the influence of these morphological characteristics. The improved SVR objective function is:

[0069]

[0070] In the formula, ‖ω‖ 2 is the norm of the weight vector, ensuring the smoothness of the model; C is the penalty factor, controlling the tolerance of the model to errors; is the sensitive loss function, L ∈ represents the insensitive loss function, y i represents the true value of the measurement, i represents the index of the sample (the i-th sample), n represents the total number of samples, and ∈ is the threshold of the tolerance error; is a regularization term used to control the size of the feature weighting coefficient to avoid overfitting, δ represents the regularization strength, j represents the index of the morphological feature, and k represents the index of the interaction term.

[0071] Step 3.5.4: Use SOA to optimize the morphological feature weighting coefficient α in SVR j , feature interaction weight coefficient β k As well as the penalty factor C and kernel function parameter γ of SVR.

[0072] Step 3.5.5: Initialize the seagull population [α 1 , α 2 , α 3 , α 4 , β 1 , β 2 , β 3 , C, γ], each seagull individual represents a parameter combination.

[0073] Step 3.5.6: Calculate the fitness value F of each seagull individual according to the improved fitness function:

[0074]

[0075] Step 3.5.7: Based on fitness evaluation, the linear flight and spiral flight formulas in SOA are used to update the position of the seagull individual (i.e., parameter combination); linear flight simulates the flight behavior of the seagull in the global search process during hunting, and the position of the seagull individual is adjusted according to the direction of the optimal individual; spiral flight is used for local search, simulating the flight of the seagull around the prey; using the linear flight and spiral flight formulas, the parameter combination is gradually optimized until the optimal solution is found. The update formulas for the two are as follows:

[0076] X n =X c +r 1 ·(X b -X c )

[0077] X n =D·e bl ·cos(2πl)+X c

[0078] Among them, X n is the updated parameter combination; X c is the parameter position of the current seagull individual; X b is the position of the current optimal individual; r 1 is a random number used to control the step size of position update; D is the distance between the individual seagull and the optimal solution; b and l are constants of spiral flight, which are used to adjust the flight amplitude.

[0079] Step 3.5.8: Determine whether the number of iterations has been reached. If so, record the optimal solution; otherwise, return to step 3.5.7.

[0080] Step 3.5.9: The optimal parameter combination found by SOA is:

[0081] [α 1 * , α 2 * , α 3 * , α 4 * , β 1 * , β 2 * , β 3 * , C * , γ * ]

[0082] Among them, α 1 * , α 2* , α 3 * , α 4 * represents the optimal weighting coefficient of morphological features; β 1 * , β 2 * , β 3 * represents the optimal weight coefficient of the interaction term of morphological features; C * represents the optimal penalty factor; γ * represents the optimal kernel function parameters.

[0083] Step 3.5.10: Input the optimal parameters into the improved SVR regression model, and calculate the predicted phenotypic data of each fish using the optimized model parameters based on the input morphological characteristic data.

[0084] The embodiments are preferred implementations of the present invention, but the present invention is not limited to the above-mentioned implementations. Any obvious improvements, substitutions or modifications that can be made by those skilled in the art without departing from the essential content of the present invention belong to the protection scope of the present invention.

Claims

1. A non-contact measurement device for freshwater fish phenotypic data based on machine vision, characterized in that: The invention comprises a fishing module (1), a visual module (2), a power supply module (3), and a control module (4); the fishing module (1) comprises a frame on which a floating ball (21) is installed, an aluminum alloy plate (23) is installed on the top of the frame, a crane (25) is installed under the aluminum alloy plate (23), the crane (25) is connected to a steel wire rope (26), the other end of the steel wire rope (26) is connected to a tension sensor (27), the tension sensor (27) is connected to a fish cage (28) through the steel wire rope (26), and a water level sensor (29) is installed under the fish cage (28); the visual module (2) comprises an underwater camera (31) installed inside the fish cage (28) and an industrial camera (35) installed under the aluminum alloy plate (23); the control module (4) comprises a control cabinet (55) installed on the fishing module (1), a weighing transmitter (51), a main control board, and a battery (56) in the control cabinet (55), and also comprises an outdoor router (53) and a host computer (54).

2. A method for non-contact measurement of freshwater fish phenotypic data based on machine vision using the non-contact measurement device for freshwater fish phenotypic data based on machine vision according to claim 1, characterized in that: The process includes the following: Step 1: Place the fishing device (1) within the feeding radius of freshwater fish. When the freshwater fish swim into the fish cage (28), the water level sensor (29) senses the water level change and transmits a signal to the main control board, triggering the underwater camera (31) and the industrial camera (35) to take pictures, thereby obtaining a top view and a side view of the fish school. Step 2: After a delay, the fishing module (1) pulls up the fish cage (28) through the crane (25), the wire rope (26) and the tension sensor (27) to weigh the freshwater fish; the tension sensor (27) senses the weight change and outputs an analog voltage, and the weighing transmitter (51) reads the voltage signal and converts the weighing data and transmits it to the main control board; Step 3: The main control board transmits the collected image data and weight data to the host computer (54) through the outdoor router (53), and the host computer (54) performs image processing to obtain the predicted phenotypic data of each fish.

3. The non-contact measurement method for freshwater fish phenotypic data based on machine vision according to claim 2, characterized in that: The specific process of step 3 is as follows: Step 3.1: grayscale conversion and filtering are performed on the side view and top view of the freshwater fish captured by the underwater camera (31) and the industrial camera (35); Step 3.2: Use Canny edge detection to identify and separate fish body pixels from background pixels; Step 3.3: Post-process the segmented results to obtain accurate fish body boundaries; Step 3.4: extracting the area, length, width and thickness data from the morphological features of the processed image; Step 3.5: Measure phenotypic data through SOA-SVR algorithm: Introduce nonlinear combinations and interaction terms related to morphological features into the SVR regression model, introduce different weight coefficients for each morphological feature, optimize the weight coefficients through SOA, and use the optimized and improved SVR regression model to predict phenotypic data.

4. The non-contact measurement method for freshwater fish phenotypic data based on machine vision according to claim 3 is characterized in that: The specific process of step 3.5 is as follows: Step 3.5.1: Assign a weight coefficient to each morphological feature in the SVR regression model. The SVR regression function is: y i * =α1·A i +α2·L i +α3·W i +α4·T i +b In the formula, α1, α2, α3, and α4 are weighted coefficients of morphological characteristics, reflecting the influence of each characteristic on phenotypic data; b is the bias term of the regression model; y i * is the predicted phenotypic data; A i , L i , W i 、T i They represent the fish body area, length, width and thickness respectively; Step 3.5.2: Introduce the interaction terms between features, then the improved SVR regression model formula is: Among them, β1, β2, and β3 are the weighted coefficients of feature interaction terms, which are used to represent the interaction between morphological features; A i ·L i , L i ·W i and W i ·T i is the feature interaction term; Indicates the final predicted phenotypic data results; Step 3.5.3: Based on the improved SVR regression model, adjust the objective function of SVR to: In the formula, ‖ω‖ 2 is the norm of the weight vector; C is the penalty factor; is the sensitive loss function, L ∈ represents the insensitive loss function, y i represents the true value of the measurement, i represents the index of the sample, n represents the total number of samples, and ∈ is the threshold of tolerance error; is the regularization term, δ represents the regularization strength, j represents the index of the morphological feature, and k represents the index of the interaction term; Step 3.5.4: Use SOA to optimize the morphological feature weighting coefficient α in SVR j , feature interaction weight coefficient β k As well as the penalty factor C and kernel function parameter γ of SVR; Step 3.5.5: Initialize the seagull population, each seagull individual represents a parameter combination; Step 3.5.6: Calculate the fitness value F of each seagull individual: Step 3.5.7: Use the linear flight and spiral flight formulas of SOA to gradually optimize the parameter combination; Step 3.5.8: Determine whether the number of iterations has been reached. If so, record the optimal solution. Otherwise, return to step 3.5.

7. Step 3.5.9: Find the optimal parameter combination through SOA; Step 3.5.10: Input the optimal parameters into the improved SVR regression model to calculate the predicted phenotypic data for each fish.

Images