A system and method for determining the growth of flounder

Abstract

The present invention relates to the field of aquaculture technology. A system and method for distinguishing the growth of flounder and flatfish are disclosed. A high-definition camera is used to shoot farmed fish bodies at multiple angles to obtain unobstructed, non-overlapping, high-resolution farmed fish body images. A YOLOv8 segmentation model is established, and a recognition module is established. The recognition module obtains measurable parameters and framework parameters based on the mask of the generated fish body. After completing data correction, species analysis and recognition are completed. The species information is input into the YOLOv8 model, and the growth of the farmed fish body is distinguished according to the final growth of body thickness, body length, and body width. The present invention completes farmed fish body image segmentation by establishing a YOLOv8 segmentation model. After obtaining measurable parameters and framework parameters, automatic generation and discrimination are completed according to the body thickness derivation results of different species, so as to improve the efficiency and accuracy of farmed management and provide strong data support for scientific feeding, feed optimization, growth regulation, disease warning, etc. in the farmed process.

Description

Technical Field

[0001] The present invention relates to the technical field of aquaculture, and in particular to a system and method for distinguishing the growth of flounder. Background Art

[0002] With the advancement of science and technology, the aquaculture industry has gradually transitioned from traditional aquaculture methods to diversified large-scale and intensive aquaculture models such as ponds, factories, cages, fences, and work boats. These large-scale aquaculture integrates modern technologies such as civil engineering, mechanical electronics, instrumentation, chemistry, bioengineering, automatic control, and socioeconomics. They implement semi-automatic or fully automatic monitoring of water quality, water flow, water temperature, feeding, sewage discharge, disease prevention, water treatment, and recycling throughout the aquaculture production process, enabling them to maintain suitable physiological and ecological conditions from beginning to end under high-density aquaculture conditions without causing internal and external environmental pollution, thereby achieving the health and rapid growth of aquaculture objects and improving the output and quality of unit water bodies. With the gradual development of factory aquaculture technology, it can solve many problems existing in traditional aquaculture.

[0003] During the fish farming process, it is impossible to accurately feed the fish because the survival rate cannot be determined. However, in order to improve the efficiency of farming, farmers often feed excessive amounts of bait, resulting in waste of bait and causing diseases such as enteritis in the fish. At the same time, excess bait settles to the bottom of the water and produces a large amount of harmful substances through fermentation, causing water quality to deteriorate and causing other fish diseases. The determination of growth performance can not only achieve precise feeding and nutritional management, but also determine the health status of fish through growth performance, which is of great significance for disease early warning. In addition, through in-depth analysis of farming data, it can also help farmers optimize production processes, improve product quality, and thus enhance market competitiveness. Summary of the Invention

[0004] In view of the deficiencies in the prior art, the present invention provides a system and method for distinguishing the growth of flatfish, which solves the problems raised in the above-mentioned background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a system and method for determining the growth of flatfish, comprising the following specific steps:

[0006] S1: Use a high-definition camera to shoot farmed fish from multiple angles to obtain multiple sets of unobstructed, non-overlapping, high-resolution farmed fish images. Motion is determined between adjacent frames of farmed fish images. Fish that do not move are photographed for the next cycle. Fish that do not show significant movement for five consecutive shooting cycles are considered dead.

[0007] S2: Establish a YOLOv8 segmentation model, the YOLOv8 segmentation model performs segmentation based on the farmed fish image, generates a mask of the fish body, and calculates the body length d and body width p of the farmed fish body based on the mask;

[0008] S3: Establishing an identification module, which obtains measurable parameters and framework parameters based on the generated fish mask, and completes species analysis and identification through DataRobot software after completing data correction;

[0009] S4: Input the species information into the YOLOv8 model. The YOLOv8 segmentation model derives the body thickness h based on the body width p, body length d, and species, and judges the growth of the farmed fish based on the final body thickness h, body length d, and body width p.

[0010] A further improvement of the technical solution of the present invention is that: the YOLOv8 segmentation model performs mask detection based on contour detection technology, obtains the head and coordinate points of the farmed fish, and calculates the body length of the fish based on the distance formula;

[0011]

[0012] Where x head 、y head and x tail 、y tail These are the coordinates of the head and tail of the farmed fish obtained based on the contour detection technology;

[0013] The YOLOv8 segmentation model performs mask detection based on contour detection technology, identifies the left and right edge coordinates of the widest part of the fish body, and calculates the vertical distance between the left and right edges as the body width p.

[0014] A further improvement of the technical solution of the present invention is that: the recognition module selects recognition feature coordinate points based on the coordinates obtained by the contour detection technology, and completes variety analysis and recognition in combination with the measurable parameters and framework parameters;

[0015] The identification feature coordinate points include the coordinate point of the front end of the snout of the farmed fish, the coordinate point of the first dorsal fin, the coordinate point of the starting point of the pelvic fin, the coordinate point of the second dorsal fin, the coordinate point of the starting point of the anal fin, the coordinate point of the starting point of the upper fifth adipose fin, the coordinate point of the starting point of the lower fifth adipose fin, and the coordinate point of the last tail vertebra.

[0016] A further improvement of the technical solution of the present invention is that the measurable parameters in the identification module include snout length, eye diameter, eye distance, tail peduncle length, tail peduncle height, body width p, and body length d.

[0017] A further improvement of the technical solution of the present invention is that: the framework parameters include the length of the line connecting the front end coordinate point of the snout and the starting point coordinate point of the first dorsal fin, the line connecting the front end coordinate point of the snout and the starting point coordinate point of the pelvic fin, the line connecting the first dorsal fin and the starting point coordinate point of the pelvic fin, the line connecting the first dorsal fin and the starting point coordinate point of the second dorsal fin, the line connecting the starting point coordinate point of the pelvic fin and the starting point coordinate point of the anal fin, the line connecting the starting point coordinate point of the second dorsal fin and the starting point coordinate point of the upper fifth adipose fin, the line connecting the starting point coordinate point of the anal fin and the starting point coordinate point of the lower fifth adipose fin, the line connecting the starting point of the upper fifth adipose fin and the coordinate point of the most rear tail vertebra, and the line connecting the starting point coordinate point of the lower fifth adipose fin and the coordinate point of the most rear tail vertebra.

[0018] In the above, the recognition module obtains the measurable parameters and frame parameters based on the generated fish body mask. After completing the data correction, the DataRobot software is used to complete the species analysis and identification. The data correction specifically includes the correction formula;

[0019] e=exp(lnY-b(lnX-lnX L ));

[0020] Where, e is the correction value, Y is the original measured value of the measurable parameters snout length, eye diameter, eye distance, caudal peduncle length, caudal peduncle height, body width p, and body length d, X is the length of the line connecting the coordinate point of the front end of the snout and the coordinate point of the starting point of the first dorsal fin, and X is the length of the line connecting the coordinate point of the front end of the snout and the coordinate point of the starting point of the first dorsal fin. L is the average value of the lengths of each line, b is the slope of lnY to lnX, and the corrected parameters of snout length, eye diameter, eye distance, caudal peduncle length, caudal peduncle height, body width p, and body length d are compared with the parameters of snout length, eye diameter, eye distance, caudal peduncle length, caudal peduncle height, body width p, and body length d of various species of flatfish through software to achieve species comparison and discrimination.

[0021] A further improvement of the technical solution of the present invention is that the body thickness h in S4 is derived from an empirical formula, wherein the empirical formula is:

[0022] Body thickness h = k1 × body length d + k2 × body width p + k3;

[0023] Wherein, body length d and body width p are calculated based on the distance between coordinate points obtained by contour detection technology, k1 and k2 are variety variable coefficients derived from a large number of manual measurements, and k3 is the error compensation coefficient.

[0024] Taking turbot as an example, the empirical formula is body thickness h = 0.055 × body length d + 0.019 × body width p + 0.686.

[0025] A further improvement of the technical solution of the present invention is that a flatfish growth identification system is also provided, which includes a model training module, an image acquisition module, a species identification module, and an intelligent analysis module;

[0026] The model training module trains the YOLOv8 model based on the prepared training set and adjusts it according to the results of cross-validation;

[0027] The image acquisition module shoots the fish body from multiple angles based on a high-definition camera;

[0028] The variety identification module performs variety identification based on measurable parameters and framework parameters;

[0029] The intelligent analysis module determines the growth status based on body width, body length, and body thickness.

[0030] Compared with the prior art, the beneficial effects of the present invention are as follows: by establishing a YOLOv8 segmentation model, the image segmentation of farmed fish is completed, and a mask of the fish body is generated. The YOLOv8 segmentation model completes high-precision measurement of the length and width of the farmed fish body based on the mask, with small and stable errors. After obtaining measurable parameters and framework parameters, data correction is completed to ensure accuracy while realizing species analysis. Automatic generation and discrimination are completed based on the body thickness deduction results of different species, which can better reflect the actual growth status of the fish body and solve the problem of cumbersome measurement and identification processes in traditional farming. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 The figure is a flow chart of the method for determining the growth of flatfish;

[0032] Figure 2 This is a schematic diagram of the fish body framework parameters in the flatfish growth discrimination method;

[0033] Figure 3 This is a flow chart of the flatfish growth identification system;

[0034] Figure 4 This is a table showing the difference between the automated measurement results and the manual measurement results in Example 3. DETAILED DESCRIPTION

[0035] Various exemplary embodiments, features, and aspects of the present application will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise indicated.

[0036] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0037] In addition, numerous specific details are provided in the following specific examples to better illustrate the present application. Those skilled in the art will appreciate that the present application can be practiced without certain specific details. In some instances, methods, means, and components well known to those skilled in the art are not described in detail in order to highlight the main purpose of the present application.

[0038] In embodiment 1, the present invention provides a system and method for determining the growth of flatfish, comprising the following specific steps:

[0039] S1: Use a high-definition camera to shoot farmed fish from multiple angles to obtain multiple sets of unobstructed, non-overlapping, high-resolution farmed fish images. Motion is determined between adjacent frames of farmed fish images. Fish that do not move are photographed for the next cycle. Fish that do not show significant movement for five consecutive shooting cycles are considered dead.

[0040] S2: Establish a YOLOv8 segmentation model, the YOLOv8 segmentation model performs segmentation based on the farmed fish image, generates a mask of the fish body, and calculates the body length d and body width p of the farmed fish body based on the mask;

[0041] S3: Establishing an identification module, which obtains measurable parameters and framework parameters based on the generated fish mask, and completes species analysis and identification through DataRobot software after completing data correction;

[0042] S4: Input the species information into the YOLOv8 model. The YOLOv8 segmentation model derives the body thickness h based on the body width p, body length d, and species, and judges the growth of the farmed fish based on the final body thickness h, body length d, and body width p.

[0043] The YOLOv8 segmentation model extracts image features to complete the farmed fish image segmentation, and then uses the median filtering algorithm to eliminate the background and interference of the farmed fish image, while retaining the outline of the farmed fish to obtain the fish mask.

[0044] The YOLOv8 segmentation model performs mask detection based on contour detection technology, obtains the head and coordinate points of the farmed fish, and calculates the body length of the fish based on the distance formula;

[0045]

[0046] Where x head 、y head and x tail 、y tail These are the coordinates of the head and tail of the farmed fish obtained based on the contour detection technology;

[0047] The YOLOv8 segmentation model performs mask detection based on contour detection technology, identifies the left and right edge coordinates of the widest part of the fish body, and calculates the vertical distance between the left and right edges as the body width p.

[0048] The recognition module selects the recognition feature coordinate points based on the coordinates obtained by the contour detection technology, and completes the variety analysis and recognition in combination with the measurable parameters and framework parameters;

[0049] The identification feature coordinate points include the coordinate point of the front end of the snout of the farmed fish, the coordinate point of the first dorsal fin, the coordinate point of the starting point of the pelvic fin, the coordinate point of the second dorsal fin, the coordinate point of the starting point of the anal fin, the coordinate point of the starting point of the upper fifth adipose fin, the coordinate point of the starting point of the lower fifth adipose fin, and the coordinate point of the last tail vertebra.

[0050] The measurable parameters in the identification module include snout length, eye diameter, eye distance, tail peduncle length, tail peduncle height, body width p, and body length d.

[0051] The framework parameters include the length of the line between the front end coordinate point of the snout and the starting point coordinate point of the first dorsal fin, the line between the front end coordinate point of the snout and the starting point coordinate point of the pelvic fin, the line between the first dorsal fin and the starting point coordinate point of the pelvic fin, the line between the first dorsal fin and the starting point coordinate point of the second dorsal fin, the line between the starting point coordinate point of the pelvic fin and the starting point coordinate point of the anal fin, the line between the starting point coordinate point of the second dorsal fin and the starting point coordinate point of the fifth adipose fin, the line between the starting point coordinate point of the anal fin and the starting point coordinate point of the fifth adipose fin, the line between the starting point of the fifth adipose fin and the last tail vertebra, and the line between the starting point coordinate point of the fifth adipose fin and the last tail vertebra. Figure 2 As shown;

[0052] In the above, the recognition module obtains the measurable parameters and frame parameters based on the generated fish body mask. After completing the data correction, the DataRobot software is used to complete the species analysis and identification. The data correction specifically includes the correction formula;

[0053] e=exp(lnY-b(lnX-lnX L ));

[0054] Where, e is the correction value, Y is the original measured value of the measurable parameters snout length, eye diameter, eye distance, caudal peduncle length, caudal peduncle height, body width p, and body length d, X is the length of the line connecting the coordinate point of the front end of the snout and the coordinate point of the starting point of the first dorsal fin, and X is the length of the line connecting the coordinate point of the front end of the snout and the coordinate point of the starting point of the first dorsal fin. L is the average value of the lengths of each line, b is the slope of lnY to lnX, and the corrected parameters of snout length, eye diameter, eye distance, caudal peduncle length, caudal peduncle height, body width p, and body length d are compared with the parameters of snout length, eye diameter, eye distance, caudal peduncle length, caudal peduncle height, body width p, and body length d of various species of flatfish through software to achieve species comparison and discrimination.

[0055] The body thickness h in S4 is derived from an empirical formula, wherein the empirical formula is:

[0056] Body thickness h = k1 × body length d + k2 × body width p + k3;

[0057] Wherein, body length d and body width p are calculated based on the distance between coordinate points obtained by contour detection technology, k1 and k2 are variety variable coefficients derived from a large number of manual measurements, and k3 is the error compensation coefficient.

[0058] Taking turbot as an example, the empirical formula is body thickness h = 0.055 × body length d + 0.019 × body width p + 0.686.

[0059] In the second embodiment, the present invention further provides a flatfish growth identification system applicable to the above method, comprising a model training module, an image acquisition module, a species identification module, and an intelligent analysis module;

[0060] The model training module trains the YOLOv8 model based on the prepared training set and adjusts it according to the results of cross-validation;

[0061] The image acquisition module shoots the fish body from multiple angles based on a high-definition camera;

[0062] The variety identification module performs variety identification based on measurable parameters and framework parameters;

[0063] The intelligent analysis module determines the growth status based on body width, body length, and body thickness.

[0064] In Example 3, turbot was generated and identified based on the system and method disclosed above. The subjects of this example were flatfish from an experimental farm, covering samples of different growth stages. Based on actual farming conditions, the fish lengths in this example fluctuated between 10 and 20 centimeters, and the body width data correspondingly varied between 8 and 15 centimeters. This sample can well represent flatfish of different body shapes and growth stages. The results of this example can also more comprehensively reflect the applicability and performance of the model.

[0065] Shooting equipment: Use a high-definition camera with a resolution of 1920x1080 to capture high-resolution images of farmed fish. Shoot from multiple angles to ensure a clear outline of the fish in the image and avoid occlusion and overlap.

[0066] Computing equipment: The computational tasks required for the examples were performed on a high-performance computing workstation equipped with an NVIDIA RTX 3080 GPU accelerator to support real-time segmentation and detection of the YOLOv8 model. The workstation also has an Intel i5 processor and 64GB of memory to ensure computational stability when processing large batches of samples.

[0067] Manual measurement tools: To verify the accuracy of the automated measurements, manual measurements were performed using a traditional measuring tape and electronic scale. The length, width, and thickness of each fish were measured manually and used as a comparison standard for the YOLOv8 model measurements.

[0068] Operating system: Ubuntu 22.04 LTS, a stable operating environment that supports the YOLOv8 model and its related tools; YOLOv8 version: YOLOv8v1.0, this embodiment uses this version of the model for fish segmentation and detection tasks; Python version: Python 3.10 is the main programming language of this embodiment, integrating the YOLOv8 model and related data processing tools; deep learning framework: PyTorch 1.11 is the main framework for model training and inference, which can efficiently support the YOLOv8 model to run on high-performance GPUs; auxiliary tools include OpenCV, NumPy, and Matplotlib, etc. These tools are mainly used for image processing, data calculation, and result visualization.

[0069] In this example, data processing is a key step in model evaluation, directly impacting the accuracy and stability of the YOLOv8 model's measurement results. The processing includes data collation, error calculation, and error source analysis. First, the length, width, and inferred thickness data measured by the YOLOv8 model are collated and compared with manually measured data. The manually measured data serves as the "true value" for evaluating the accuracy of the model's predictions.

[0070] Then, the manual measurement results were compared with the model prediction data, and the errors of body length, body width, and body thickness were calculated one by one. The error is defined as the difference between the model prediction value and the actual measurement value, and the calculation formula is as follows:

[0071] Error = |Model measurement value - manual measurement value |

[0072] By calculating the error for each sample, we can accurately evaluate the model's performance across various sample types. Data processing also analyzes the distribution of errors, identifies samples with large errors, and explores possible sources of error, such as image quality, fish posture, and model generalization capabilities. This analysis helps identify the strengths and weaknesses of the model's performance and provides a basis for subsequent optimization.

[0073] The specific steps are as follows:

[0074] Data Preparation: First, a high-definition camera was used to capture the fish from multiple angles, ensuring that its complete outline and various postures were captured. The fish's length, width, and thickness were also manually measured and recorded as ground truth. This manually annotated data was used to train and test the YOLOv8 model.

[0075] YOLOv8 model training: The YOLOv8 model is trained using the prepared training set. During training, the model learns the edges, contours, and features of the fish, gradually improving its segmentation and measurement accuracy. During training, model hyperparameters (such as the learning rate and number of training rounds) are adjusted based on cross-validation results to ensure optimal model performance.

[0076] Model Prediction and Measurement: After model training is complete, it is validated using a test set. During this process, the YOLOv8 model automatically identifies the fish outline in the test image and calculates the fish's length and width. Based on these measurements, software compares the snout length, eye diameter, eye distance, caudal peduncle length, caudal peduncle height, body width, and body length characteristics of various flatfish species using snout length, eye diameter, eye distance, caudal peduncle length, caudal peduncle height, body width p, and body length d. This enables species comparison and identification. An empirical formula is used to derive the body thickness of each fish species, generating complete prediction data.

[0077] Results Analysis: Finally, the model's predicted length, width, and thickness were compared in detail with the manually measured values. The errors were calculated and the overall model performance was analyzed. The distribution and trend of the errors were used to assess the model's stability and generalization ability. The model's performance on fish of different sizes was summarized to determine whether further optimization or adjustments were needed.

[0078] The results are as attached Figure 4 As shown in this example, the length and width of flatfish fluctuate between 10 and 20 centimeters. The following are the length, width, and thickness measurements of some samples, including those obtained using the YOLOv8 segmentation model and actual manually measured data. The error between the derived thickness and the actual value is also shown, with an average error of 0.52 cm. This indicates that most measurements have a small deviation from the actual values, but the deviation is within an acceptable range.

[0079] Maximum Error: The largest error occurs in fish number 2, with an error of 0.9 cm. This specimen has a relatively small length and width, and the model may have some prediction bias for small fish. Due to the small size of the fish, the image captures relatively less detail, which may amplify errors in segmentation and measurement.

[0080] Minimum error: The minimum error is 0.2 cm, which occurs in fish number 1. This shows that for medium-sized fish, the model's prediction is more accurate and stable, and the derived thickness value is highly consistent with the manual measurement value.

[0081] Error Range: The overall error range is between 0.2cm and 0.9cm, with all errors within 1cm, demonstrating the reliability of the YOLOv8 model's measurements. The error for most samples is around 0.5cm, meeting the accuracy requirements for automated measurement.

[0082] The experimental fish used in this example were all turbot, and the accuracy rate in species identification was 100%. The body thickness derivation formula used in this test was: body thickness = 0.055 × body length + 0.019 × body width + 0.686

[0083] The formula is based on a linear regression model and is obtained by fitting historical data. It aims to deduce the body thickness of fish from the measurements of body length and width.

[0084] Based on the current data analysis, the model's overall performance in the test is acceptable. For medium-sized fish, the prediction error is small, indicating good performance. However, for extremely small and large fish, the error is larger, suggesting that further optimization of the model may be necessary to improve its applicability. The error is kept within 1 cm, indicating that the model has a certain degree of accuracy and stability.

[0085] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0086] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A method for determining the growth of flatfish, characterized in that: The specific steps include: S1: Use a high-definition camera to shoot farmed fish from multiple angles to obtain multiple sets of unobstructed, non-overlapping, high-resolution farmed fish images. Motion is determined between adjacent frames of farmed fish images. Fish that do not move are photographed for the next cycle. Fish that do not show significant movement for five consecutive shooting cycles are considered dead. S2: Establish a YOLOv8 segmentation model, the YOLOv8 segmentation model performs segmentation based on the farmed fish image, generates a mask of the fish body, and calculates the body length d and body width p of the farmed fish body based on the mask; S3: Establishing an identification module, which obtains measurable parameters and framework parameters based on the generated fish mask, and completes species analysis and identification through DataRobot software after completing data correction; S4: Input the species information into the YOLOv8 model. The YOLOv8 segmentation model derives the body thickness h according to the body width p, body length d, and species, and judges the growth of the farmed fish according to the final body thickness h, body length d, and body width p; The measurable parameters in the recognition module include snout length, eye diameter, eye distance, caudal peduncle length, caudal peduncle height, body width p, and body length d; The framework parameters include the length of the line connecting the front end coordinate point of the snout and the starting point coordinate point of the first dorsal fin, the line connecting the front end coordinate point of the snout and the starting point coordinate point of the pelvic fin, the line connecting the first dorsal fin and the starting point coordinate point of the pelvic fin, the line connecting the first dorsal fin and the starting point coordinate point of the second dorsal fin, the line connecting the starting point coordinate point of the pelvic fin and the starting point coordinate point of the anal fin, the line connecting the starting point coordinate point of the second dorsal fin and the starting point coordinate point of the fifth adipose fin, the line connecting the starting point coordinate point of the anal fin and the starting point coordinate point of the fifth adipose fin, the line connecting the starting point of the fifth adipose fin and the last tail vertebra, and the line connecting the starting point coordinate point of the fifth adipose fin and the last tail vertebra; The correction formula used for data correction in S3 is: e=exp(lnY-b(lnX-lnX L )); Where, e is the correction value, Y is the original measured value of the measurable parameters snout length, eye diameter, eye distance, tail peduncle length, tail peduncle height, body width p, and body length d, X is the length of the line between the coordinate points, and X is the length of the line between the coordinate points. L is the average value of the lengths of each line, b is the slope of lnY to lnX, and the corrected parameters of snout length, eye diameter, eye distance, caudal peduncle length, caudal peduncle height, body width p, and body length d are compared with the parameters of snout length, eye diameter, eye distance, caudal peduncle length, caudal peduncle height, body width p, and body length d of various species of flatfish through software to achieve species comparison and discrimination.

2. A method for distinguishing the growth of flatfish according to claim 1, characterized in that: The YOLOv8 segmentation model extracts image features to complete the farmed fish image segmentation, and then uses the median filtering algorithm to eliminate the background and interference of the farmed fish image, while retaining the outline of the farmed fish to obtain the fish mask.

3. The method for determining the growth of flatfish according to claim 1, wherein: The YOLOv8 segmentation model performs mask detection based on contour detection technology, obtains the head and coordinate points of the farmed fish, and calculates the body length of the fish based on the distance formula; Body length Where x head 、y head and x tail 、y tail These are the coordinates of the head and tail of the farmed fish obtained based on the contour detection technology; The YOLOv8 segmentation model performs mask detection based on contour detection technology, identifies the left and right edge coordinates of the widest part of the fish body, and calculates the vertical distance between the left and right edges as the body width p.

4. The method for determining the growth of flatfish according to claim 1, wherein: The recognition module selects the recognition feature coordinate points based on the coordinates obtained by the contour detection technology, and completes the variety analysis and recognition in combination with the measurable parameters and framework parameters; The identification feature coordinate points include the coordinate point of the front end of the snout of the farmed fish, the coordinate point of the first dorsal fin, the coordinate point of the starting point of the pelvic fin, the coordinate point of the second dorsal fin, the coordinate point of the starting point of the anal fin, the coordinate point of the starting point of the upper fifth adipose fin, the coordinate point of the starting point of the lower fifth adipose fin, and the coordinate point of the last tail vertebra.

5. The method for determining the growth of flatfish according to claim 1, wherein: The body thickness h in S4 is derived from an empirical formula, wherein the empirical formula is: Body thickness h = k1 × body length d + k2 × body width p + k3; Wherein, body length d and body width p are calculated based on the distance between coordinate points obtained by contour detection technology, k1 and k2 are variety variable coefficients derived from a large number of manual measurements, and k3 is the error compensation coefficient.

6. A flatfish growth identification system, applied to the flatfish growth identification method according to any one of claims 1 to 5, characterized in that: Including model training module, image acquisition module, variety identification module, and intelligent analysis module; The model training module trains the YOLOv8 model based on the prepared training set and adjusts it according to the results of cross-validation; The image acquisition module shoots the fish body from multiple angles based on a high-definition camera; The variety identification module performs variety identification based on measurable parameters and framework parameters; The intelligent analysis module determines the growth status based on body width, body length, and body thickness.

Images