Gonad development evaluation method and system for gymnocypris przewalskii in qinghai lake based on image recognition

Abstract

The application provides an image recognition-based evaluation method and system for gonad development of Qinghai Lake Gymnocypris, and relates to the technical field of computers, comprising: acquiring an internal image of gonads, and basic biological parameters of corresponding individuals and water temperature data at a sampling time; performing gonad region segmentation and development stage classification according to the internal image of the gonads, taking a stage corresponding to the highest probability as a gonad development stage determination result; performing feature extraction according to the gonad development stage determination result, to obtain key feature parameters of the gonads; performing individual development level evaluation according to the key feature parameters of the gonads, the gonad development stage determination result, the basic biological parameters and the water temperature data, to obtain a quantitative evaluation index; performing group development state analysis according to the quantitative evaluation index, to obtain an overall state evaluation result; and performing reproduction timing prediction according to the overall state evaluation result and the water temperature data, to obtain a prediction result. The application realizes non-invasive monitoring and evaluation of the Qinghai Lake Gymnocypris.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and more specifically, to a method and system for assessing gonadal development in Qinghai Lake naked carp based on image recognition. Background Technology

[0002] The Qinghai Lake naked carp (Gymnocypris przewalskii) belongs to the genus Gymnocypris in the family Cyprinidae of the class Actinopterygii in the phylum Chordata. It is a cold-water fish endemic to Qinghai Lake and its water system, possessing significant ecological conservation and scientific research value. Currently, its survival is threatened by overfishing and environmental stress, making germplasm resource protection an urgent priority. Gonadal development status is a core indicator for assessing the reproductive capacity, population replenishment potential, and germplasm health of the Qinghai Lake naked carp. Accurate monitoring of its gonadal development stages is crucial for artificial breeding, population recovery, and resource conservation.

[0003] Currently, the assessment of gonadal development in Qinghai Lake naked carp mainly employs traditional methods, including anatomical sampling observation, endoscopic examination, and physiological and biochemical index detection. Among these, anatomical sampling observation requires the slaughter of samples, leading to individual carp mortality, which contradicts the concept of germplasm resource conservation. Furthermore, it cannot achieve continuous live monitoring, has a limited sample size, and fails to reflect the overall gonadal development level of the population. While endoscopic examination allows for live testing, the procedure is cumbersome, requires highly skilled operators, has low efficiency, and relies on subjective judgment of gonadal development stages, resulting in significant errors and hindering large-scale, standardized monitoring. Additionally, this method is difficult to control the stress response of Qinghai Lake naked carp, indirectly affecting the gonadal development process. Physiological and biochemical index detection requires the collection of blood and tissue samples, is complex, has a long testing cycle, and only reflects indirect characteristics of gonadal development, failing to directly and accurately correspond to specific stages of gonadal development, thus failing to meet the needs of rapid monitoring.

[0004] With the gradual application of artificial intelligence technology in aquaculture, using machine learning or deep learning models to assist in judging the gonadal development status of fish has become a research hotspot. For example, Chinese invention patent CN119539209A discloses a method for regulating the aquaculture environment to promote rapid gonadal maturation in perch. This method constructs a deep learning network prediction model based on multilayer perceptrons through aquaculture experiments under different combinations of aquaculture environmental parameters, and combines genetic algorithms to optimize the aquaculture environmental parameters to promote rapid gonadal maturation. Although this method introduces a deep learning model, its goal is to optimize aquaculture environmental parameters. The model inputs environmental variables such as salinity, temperature, photoperiod, and nutrient content, and the outputs predicted values ​​of growth trait parameters and gonadal development parameters. Essentially, it is an indirect prediction method based on the mapping relationship between environmental parameters and developmental parameters, without involving the direct acquisition and analysis of gonadal images, and cannot achieve intuitive and real-time monitoring of individual gonadal development stages.

[0005] Based on the shortcomings of the existing technologies, there is an urgent need for a method and system for assessing gonadal development in Qinghai Lake naked carp based on image recognition. Summary of the Invention

[0006] The purpose of this invention is to provide a method and system for assessing gonadal development in *Gymnocypris qinghai Lake* based on image recognition, in order to improve the aforementioned problems. To achieve the above objective, the technical solution adopted by this invention is as follows: Firstly, this application provides an image recognition-based method for assessing gonadal development in *Gymnocypris qinghai Lake*, including: Images of the gonads inside the naked carp of Qinghai Lake were obtained from the genital opening of the carp after electroanesthesia using a minimally invasive image acquisition device, along with the basic biological parameters of the corresponding individuals and water temperature data at the time of sampling. Gonadal region segmentation and developmental stage classification are performed based on the internal gonadal image. The internal gonadal image is input into a pre-trained multi-task feature extraction network. The segmentation branch of the multi-task feature extraction network outputs a gonadal region mask, and the classification branch outputs the classification probability of each developmental stage. The stage corresponding to the highest probability is taken as the gonadal developmental stage determination result. Based on the gonadal development stage determination results, feature extraction is performed. By cropping the corresponding gonadal region image block from the gonadal region mask, oocyte morphological texture features and testis texture features are extracted from the image block respectively to obtain key gonadal feature parameters. Based on the key gonadal characteristic parameters, the gonadal development stage determination results, the basic biological parameters, and the water temperature data, the individual development level was assessed to obtain a quantitative assessment index of the individual development level of the Qinghai Lake naked carp. The population development status was analyzed based on the quantitative evaluation indicators of multiple individuals within the same sampling batch, and the overall status evaluation results of the population gonadal development were obtained. Based on the overall assessment results of the gonadal development of the population and the water temperature data, the timing of reproduction is predicted, and the prediction results of the optimal timing for artificial reproduction are obtained.

[0007] Secondly, this application also provides an image recognition-based gonadal development assessment system for *Gymnocypris qinghai Lake*, comprising: The acquisition module is used to acquire images of the gonads inside the naked carp of Qinghai Lake through a minimally invasive image acquisition device after electroanesthesia, as well as the basic biological parameters of the corresponding individual and the water temperature data at the time of sampling. The determination module is used to segment the gonadal region and classify the developmental stage based on the internal gonadal image. The internal gonadal image is input into a pre-trained multi-task feature extraction network. The segmentation branch of the multi-task feature extraction network outputs a gonadal region mask, and the classification branch outputs the classification probability of each developmental stage. The stage with the highest probability is taken as the gonadal developmental stage determination result. The extraction module is used to extract features based on the gonadal development stage determination results. It obtains key gonadal feature parameters by cropping the corresponding gonadal region image block from the gonadal region mask and extracting oocyte morphological texture features and testis texture features from the image block. The assessment module is used to assess the individual development level of the Qinghai Lake naked carp based on the key gonadal characteristic parameters, the gonadal development stage determination results, the basic biological parameters and the water temperature data, and to obtain a quantitative assessment index of the individual development level of the Qinghai Lake naked carp. The analysis module is used to analyze the population development status based on the quantitative evaluation indicators of multiple individuals within the same sampling batch, and to obtain the overall status evaluation results of the population gonadal development. The prediction module is used to predict the timing of artificial breeding based on the overall assessment results of the gonadal development of the population and the water temperature data.

[0008] The beneficial effects of this invention are as follows: I. This invention utilizes electroanesthesia combined with genital pore endoscopy for minimally invasive image acquisition of the gonads, replacing traditional anatomical sampling methods. This enables non-invasive monitoring and evaluation of the Qinghai Lake naked carp. Individual samples do not require slaughter during collection and can survive after recovery for subsequent follow-up monitoring. This fundamentally solves the problems of individual injury and mortality caused by traditional anatomical methods, aligning with the concept of Qinghai Lake naked carp germplasm resource conservation and laying the foundation for continuous live monitoring and expanded sample size.

[0009] II. This invention employs a multi-task feature extraction network to segment and classify images of the gonadal interior. The segmentation branch outputs pixel-by-pixel gonadal region masks, while the classification branch outputs classification probability vectors corresponding to each developmental stage. This allows for direct identification of gonadal development into multiple stages, from I to V. Compared to existing technologies that rely on indirect inference based on surface morphological parameters and only distinguish between "mature" and "immature" binary classifications, this invention directly analyzes the internal tissue structure of the gonads, resulting in finer classification granularity, more intuitive judgment criteria, and more accurate results, overcoming errors caused by subjective human judgment.

[0010] Third, after the initial gonadal region mask is output by the segmentation branch, the present invention uses an adversarial generation module to perform adversarial discrimination correction on the boundary smoothness and regional connectivity of the mask, which effectively suppresses segmentation interference from non-gonadal tissues, improves the morphological accuracy and boundary precision of gonadal region segmentation, and provides a high-quality regional foundation for subsequent key feature extraction. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of the present invention and should not be regarded as a limitation on the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart illustrating an image recognition-based method for assessing gonadal development in *Gymnocypris qinghai Lake*, as described in an embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of an image recognition-based gonadal development assessment system for Qinghai Lake naked carp, as described in an embodiment of the present invention. Figure 3 The purpose of this document is to find the reference for fertilization levels based on water temperature grading. Figure 4 A comparison chart of characteristic parameters of oocytes at different developmental stages; Figure 5 This is a mapping diagram of the deviation degree and the developmental consistency score.

[0013] The diagram is labeled as follows: 901, Acquisition Module; 902, Judgment Module; 903, Extraction Module; 904, Evaluation Module; 905, Analysis Module; 906, Prediction Module. Detailed Implementation

[0014] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. The components of the embodiments of the present invention described and shown in the accompanying drawings can generally be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the claimed invention, but merely to illustrate selected embodiments of the invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0015] It should be noted that similar reference numerals and letters in the following figures indicate similar items; therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. Furthermore, in the description of this invention, terms such as "first," "second," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0016] Example 1

[0017] This embodiment provides a method for assessing gonadal development in Qinghai Lake naked carp based on image recognition.

[0018] See Figure 1 The figure shows that the method includes steps S100 to S600.

[0019] Step S100: Obtain images of the gonads inside the Qinghai Lake naked carp from the genital opening using a minimally invasive image acquisition device after electroanesthesia, as well as the basic biological parameters of the corresponding individual and the water temperature data at the sampling time. Understandably, this step forms the data collection foundation for the entire evaluation method. The data collection subjects were Qinghai Lake naked carp samples from different water areas, selecting healthy Qinghai Lake naked carp of different ages, sexes, and gonadal development stages as the collection targets. Before image acquisition, the individuals to be examined were anesthetized using a low-stress electroanesthesia method (25V~30V low-voltage DC). Mesh electrodes were installed at both ends of the electroanesthesia tank, with an electric field strength of 0.4V / cm~0.5V / cm and a current maintained at 0.1A~0.2A to avoid affecting gonadal development. After anesthesia took effect, a fish endoscope (90° viewing angle, 80° field of view) fitted with a sealed glass tube was used as a minimally invasive image acquisition device to collect images of the internal gonads from the genital opening of the Qinghai Lake naked carp. During acquisition, the image resolution was controlled to be ≥1080P, and the acquisition angles were 90° viewing angle and 80° field of view to ensure clear, unobstructed, and unblurred images. Simultaneously, basic biological parameters (including age, sex, weight, body length, etc.) were recorded for each individual under examination, and the daily average water temperature data of the water area where the sampling was conducted was collected as an environmental correction parameter for subsequent developmental assessment. This simultaneous acquisition of data laid a complete data foundation for the subsequent steps of integrating individual morphological indicators, environmental temperature factors, and image visual features for comprehensive analysis.

[0020] Step S200: Perform gonadal region segmentation and developmental stage classification based on the gonadal internal image. By inputting the gonadal internal image into a pre-trained multi-task feature extraction network, the segmentation branch of the multi-task feature extraction network outputs the gonadal region mask, and the classification branch outputs the classification probability of each developmental stage. The stage corresponding to the highest probability is taken as the gonadal development stage determination result. Understandably, this step constructs a multi-task feature extraction network based on an improved U-Net deep learning network. This network includes an input layer, a feature extraction module, a feature fusion module, a segmentation module, and a classification module. The input layer receives the preprocessed gonad images. The feature extraction module uses a multi-layer convolutional neural network (CNN) to extract key features such as texture, color, and morphology of the gonad images through convolution and pooling operations. These features include the size, density, and morphology of oocytes, and the texture clarity and color of testes. These features are directly related to the gonadal development stages of the Qinghai Lake naked carp. The feature fusion module uses skip connections to fuse shallow and deep features, improving the completeness and accuracy of feature extraction. The segmentation module uses transposed convolution to accurately segment the gonadal region and eliminate interference from non-gonadal tissues. The classification module uses fully connected layers combined with the softmax activation function to classify the gonad images into stages I to V. The segmentation and classification branches share the underlying feature extraction module, and the two tasks form complementary constraints at the feature level: the segmentation task focuses on the precise boundary localization of pixel-level regions in the image, while the classification task focuses on the global semantic discrimination of the image. This multi-task joint learning mechanism effectively improves the network's generalization ability and robustness.

[0021] Further, step S200 includes steps S210 to S240.

[0022] Step S210: Perform multi-scale semantic feature extraction based on the internal image of the gonads. Based on the shared coding module of the multi-task feature extraction network, extract the shallow texture detail features and deep semantic abstract features of the image layer by layer. The shallow texture detail features include the size and density distribution features of oocytes and the texture clarity and color features of testes. The deep semantic abstract features include the spatial morphological structure features of gonadal tissue, resulting in multi-level feature mapping. Specifically, in step S210, the acquired images of the gonadal interior are input into the shared encoding module of a pre-trained multi-task feature extraction network. This encoding module employs a multi-layer convolutional network structure to extract features from the input image layer by layer, from shallow to deep. In the shallow layers of the network, each convolutional layer captures low-level visual information in the image through convolution and pooling operations, mainly extracting key features such as texture, color, and morphology of the gonadal image, including the size, density, and morphology of oocytes, and the texture clarity and color of testes. These shallow features play a crucial role in subsequent pixel-level segmentation, directly encoding the visual differences between gonadal tissue and surrounding non-gonadal tissues (such as intestines, adipose tissue, and body cavity walls). In the deeper layers of the network, each convolutional layer gradually increases the receptive field and combines non-linear activation operations to abstract the shallow texture details into higher-level tissue semantic representations, extracting the overall spatial morphological structural features of the gonadal tissue. Shallow texture details and deep semantic abstractions form a complete multi-level feature map between different layers of the network through skip connections, providing consistent feature input for the two subsequent parallel branches.

[0023] Step S220: Perform gonad region segmentation based on multi-level feature mapping. The deep semantic features are upsampled and restored to the original resolution space through the segmentation and decoding branch of the multi-task feature extraction network. The features are combined with the edge details in the shallow features to perform feature concatenation and output a pixel-by-pixel gonad region mask. Understandably, in step S220, the segmentation decoding branch uses the aforementioned multi-level feature maps as input. The segmentation module employs transposed convolution operations to progressively restore the deep semantic features to a resolution space consistent with the original input image. In each upsampling level, the branch introduces the fine edge details contained in the shallow features of the corresponding level in the encoding stage into the decoding process through a skip connection mechanism, achieving feature cascade fusion. This ensures that the decoded features retain both the overall recognition ability of the deep semantics for the gonadal region and the accurate characterization ability of the shallow features for the region boundaries. Finally, the branch outputs a pixel-by-pixel gonadal region mask through a convolutional layer, accurately segmenting the gonadal region and eliminating interference from non-gonadal tissues (such as fish tissue, water, etc.).

[0024] Step S230: Correct the segmentation morphology deviation based on the gonadal region mask. By performing adversarial discrimination correction on the boundary smoothness and regional connectivity of the gonadal region mask, the segmentation interference of non-gonadal tissues is suppressed, and the gonadal region mask after morphology deviation correction is obtained. It should be noted that in step S230, due to factors such as wrinkles on the surface of the gonadal tissue of the Qinghai Lake naked carp, reflection of body fluids, or deviations in the endoscopic viewpoint, the gonadal region mask may exhibit jagged, irregular contours at the boundary, or non-gonadal tissue may be mistakenly included in the segmented region. Therefore, adversarial discrimination correction is performed on the boundary smoothness and regional connectivity of the gonadal region mask: the mask quality is evaluated from multiple dimensions, including boundary smoothness (continuity and curvature change of boundary pixel gradients), regional connectivity (whether the segmented region forms a complete single connected domain), and consistency with anatomical priors. The mask boundary is optimized and adjusted based on the evaluation feedback signal. Through adversarial iterative training, segmentation interference from non-gonadal tissues is gradually suppressed, and boundary jaggedness and island noise are eliminated, ultimately outputting the gonadal region mask after morphological deviation correction.

[0025] Step S240: Classify developmental stages based on multi-level feature mapping. After spatial compression of deep semantic features by the classification branch of the multi-task feature extraction network in the global average pooling layer, the classification probability vector corresponding to each developmental stage is output through fully connected mapping. The stage corresponding to the highest probability is taken as the result of gonadal development stage determination.

[0026] Specifically, in step S240, the classification branch takes multi-level feature maps as input and focuses on using deep semantic abstract features for stage discrimination. The classification module uses multi-level fully connected layers combined with the softmax activation function for classification. The branch first performs spatial dimension compression on the deep feature map in the global average pooling layer, converging the feature map into a global feature vector in the spatial dimension to eliminate the interference of positional information; then, this global feature vector is transformed by the nonlinear transformation of the fully connected mapping layer and mapped into a classification probability vector corresponding to each developmental stage. The gonadal development of the Qinghai Lake naked carp is divided into five stages: stage I (gonadal undevelopment stage, gonads are thread-like), stage II (gonadal development stage, oocytes begin to accumulate yolk), stage III (gonadal rapid development stage, oocytes are full of yolk granules), stage IV (pre-gonadal maturation stage, eggs are full and easily separated), and stage V (gonadal maturation stage, eggs are free and can be expelled). Each dimension in the classification probability vector corresponds to a probability estimate of a developmental stage. The stage corresponding to the dimension with the highest probability value in the vector is determined as the current gonadal development stage of the individual.

[0027] Furthermore, during the model training phase, the cross-entropy loss function was used to calculate the error between the model's predicted values ​​and the labeled values. The model parameters were adjusted using the backpropagation algorithm, and the model was validated using a validation set. The learning rate was set to 0.001, the number of iterations was 100-200, and the batch size was 16. Training was stopped when the model's accuracy on the validation set was ≥95% and the loss value was ≤0.05, resulting in a fully trained model. In the validation experiment, after 120 iterations, the model achieved an accuracy of 96.5% and a loss value of 0.03 on the validation set. The training set included gonad images of *Gymnocypris qinghai Lake* at all developmental stages, from stage I to stage V. All images were labeled using annotation tools, including gonad region, developmental stage (stages I-V), average oocyte diameter (female), and testis texture clarity (male). The annotation quality was verified for consistency before being used for model training.

[0028] Step S300: Based on the results of the gonadal development stage determination, feature extraction is performed. By cropping the corresponding gonadal region image block using a gonadal region mask, oocyte morphological texture features and testis texture features are extracted from the image block to obtain key gonadal feature parameters. Understandably, this step, based on the preliminary assessment of developmental stage completed in step S200, extracts biologically significant quantitative feature parameters from the gonadal region image block. First, using the gonadal region mask output from step S200, a local image block containing only gonadal tissue is cropped from the image of the primitive gonad's interior, effectively removing interference from background areas (such as fish body tissue, water, and other non-gonadal tissues) for subsequent feature extraction. Then, within this image block, oocyte morphological texture features and testicular texture features are extracted based on different morphological characteristics of female and male gonadal tissues, respectively. These are ultimately summarized into key gonadal feature parameters, providing a quantitative basis for subsequent individual developmental level assessment. It should be noted that the key feature parameters include the average diameter, density, and morphology of oocytes in female Qinghai Lake naked carp, and the clarity and color of testicular texture in male Qinghai Lake naked carp.

[0029] Further, step S300 includes steps S310 to S330.

[0030] Step S310: Cut out the corresponding gonadal region image block according to the gonadal region mask and perform quality verification. By using the highest probability value in the classification probability vector as the confidence level and comparing it with the preset effective threshold, qualified image blocks with confidence levels reaching the effective threshold are selected to obtain qualified gonadal region image blocks. Specifically, step S310 first uses a gonadal region mask to crop the image of the original gonadal interior, extracting local image blocks containing only the gonadal tissue region. Then, the highest probability value from the classification probability vector output in step S240 is extracted as a confidence index and compared with a preset effective threshold. A confidence score ≥ 90% is considered a valid identification result; that is, when the model's probability of determining a developmental stage reaches or exceeds the preset effective threshold, the developmental stage determination result of the image is considered sufficiently reliable and included in the set of qualified image blocks. Conversely, if the confidence score is lower than the preset effective threshold, it is marked as a low-quality image and removed. This confidence-based quality screening mechanism effectively reduces the impact of abnormal samples on subsequent individual evaluation results while ensuring the accuracy of feature extraction.

[0031] Step S320: Based on the qualified gonadal region image blocks and the gonadal development stage determination results, female features are extracted. By detecting the oocyte outline in the image block and calculating the average diameter of the oocyte, the density ratio of the oocyte area to the total area of ​​the gonadal region, and the texture index of the regularity of oocyte arrangement based on the gray-level co-occurrence matrix, the key gonadal feature parameters of the female individual are obtained. Specifically, step S320 targets the gonadal region image block of a female individual. Oocyte contours are detected within the image block, and the following three types of feature parameters are calculated: First, the average diameter of the oocytes, which is the arithmetic mean of the equivalent circle diameters of all detected oocyte contours in the image block. This indicator directly reflects the growth and development status of the oocytes, showing a significant increasing trend at different developmental stages. Second, the density ratio of oocyte area to the total area of ​​the gonadal region, which is the ratio of the sum of the cross-sectional areas of all oocytes to the total area of ​​the gonadal region. This indicator characterizes the filling density of oocytes in the ovary, gradually increasing with the progress of development. Third, a texture index based on the regularity of oocyte arrangement calculated using a gray-level co-occurrence matrix. By analyzing the co-occurrence probability distribution of pixel gray values ​​in the spatial direction within the image block, the degree of orderliness of the oocyte arrangement is quantified; the more regular the arrangement, the higher the texture index value. Figure 4 As shown, Figure 4 The figure shows a comparison of oocyte characteristic parameters at different developmental stages. The three grouping indicators in the figure are: average oocyte diameter (blue), oocyte density ratio (green), and oocyte arrangement regularity texture index (orange), with values ​​normalized (0~100). As the developmental stage progresses from stage I to stage V, all three characteristic parameters show a significant increasing trend, with the average oocyte diameter showing the most significant increase. These three types of characteristic parameters together constitute the key gonadal characteristic parameters of female individuals.

[0032] Step S330: Based on the qualified gonadal region image blocks and the gonadal development stage determination results, male features are extracted. By calculating the second-order statistics of energy, contrast and inverse difference moment of the gray-level co-occurrence matrix of the image block in multiple directions, and combining the morphological edge sharpness index of the testis lobule structure, the key gonadal feature parameters of male individuals are obtained.

[0033] Specifically, step S330 targets the gonadal region image block of male individuals. Since testicular tissue does not appear as discrete oocyte granules in the image but rather as a uniform glandular tissue texture, different feature extraction strategies are employed. First, the second-order statistics of the image block's gray-level co-occurrence matrix (GLCM) are calculated in four directions: horizontal (0°), vertical (90°), and diagonal (45° and 135°). These statistics include energy (reflecting the uniformity and coarseness of the texture), contrast (reflecting the sharpness and depth of the texture), and inverse difference moment (reflecting the local uniformity of the texture). Furthermore, the morphological edge sharpness index of the testicular lobule structure is calculated. By analyzing the gradient amplitude distribution of the testicular lobule boundary region in the image block, the clarity of the lobule structure's outline, i.e., the sharpness of the testicular texture, is quantified. The aforementioned GLCM multi-directional statistics and the testicular texture sharpness index together constitute the key feature parameters of the male individual's gonads. The above-mentioned female and male feature extraction is performed based on qualified image blocks after quality screening in step S310, and the corresponding feature extraction path (female path or male path) is automatically selected according to the gonadal development stage determination result, realizing differentiated feature quantification for different sexes and different development stages.

[0034] Step S400: Based on the key characteristic parameters of the gonads, the results of the gonadal development stage determination, the basic biological parameters and water temperature data, the individual development level is assessed to obtain the quantitative assessment index of the individual development level of Qinghai Lake naked carp. It should be noted that this step is a crucial bridge connecting the image-level determination of gonadal development with the comprehensive developmental assessment at the individual level. Previous steps have obtained four types of information: gonadal development stage determination results, key gonadal characteristic parameters, basic biological parameters, and water temperature data. This step integrates these heterogeneous information into a unified quantitative assessment index. Based on the developmental stage and key characteristic parameters output by the model, combined with the basic biological parameters and water temperature data of the Qinghai Lake naked carp, a comprehensive assessment of the reproductive capacity of the Qinghai Lake naked carp is conducted, resulting in a quantitative assessment index of the individual developmental level of the Qinghai Lake naked carp, clarifying whether the sample has reached the reproductive conditions, the optimal breeding time, and the reproductive health status.

[0035] Further, step S400 includes steps S410 to S440.

[0036] Step S410: Calculate the individual body fat index based on the body length and weight data in the basic biological parameters. The individual body fat index is obtained by calculating the ratio of the individual's actual weight to the standard weight reference value under the same body length conditions. Specifically, step S410 extracts the measured values ​​of body length and weight of individuals from basic biological parameters, and combines them with information such as age and sex of the Qinghai Lake naked carp to calculate the condition factor using a variant of the formula: ; In the formula, The body condition index is used to measure an individual's body fat percentage. This refers to actual body weight (in grams). Body length (unit: cm) and These are species-specific parameters obtained by fitting the body length-weight regression relationship of *Gymnocypris qinghai Lake*. The condition index is an important indicator for assessing the nutritional status and energy reserves of fish; a higher value generally indicates better nutritional status.

[0037] Step S420: Determine the water temperature range to which the current sampling time belongs based on the water temperature data. By matching the developmental stage determination result and the water temperature range in the pre-built water temperature range fertility benchmark lookup table, obtain the expected value of fertility benchmark under the corresponding conditions. Understandably, in step S420, since the plumpness of the Qinghai Lake naked carp fluctuates regularly with seasonal water temperature changes, directly using the measured plumpness for cross-seasonal comparisons would result in significant bias. Therefore, this step first determines the water temperature range based on the daily average water temperature data at the sampling time. The water temperature range is divided into four levels: low temperature (<5℃), slightly low temperature (5~10℃), medium temperature (10~15℃), and high temperature (>15℃). Figure 3 As shown, Figure 3 A lookup table for body condition benchmarks based on water temperature is shown. The vertical axis represents the gonadal development stage (stages I to V), and the horizontal axis represents the water temperature range (low temperature, slightly low temperature, medium temperature, high temperature). The color intensity indicates the level of body condition index. At the same developmental stage, the higher the water temperature, the greater the expected body condition benchmark value; under the same water temperature conditions, the later the developmental stage, the greater the expected body condition benchmark value. After determining the range, a pre-built lookup table for body condition benchmarks based on water temperature is consulted. This lookup table uses developmental stage and water temperature range as a joint index to store the expected body condition benchmark values ​​for each combination of conditions. These benchmark values ​​are derived from statistical analysis of multiple batches of Qinghai Lake naked carp samples over many years, representing the statistically expected level of individual body condition at a specific developmental stage and under specific water temperature conditions.

[0038] Step S430: Calculate and map the degree of deviation based on the individual body fullness index and the expected value of body fullness benchmark. By calculating the relative percentage of deviation between the two and performing segmented scoring mapping according to the preset deviation tolerance, the developmental consistency score is obtained. Specifically, step S430 first calculates the percentage deviation of the individual body fat index from the baseline expected value: ; In the formula, The percentage is the relative deviation. The body condition index is used to measure an individual's body fat percentage. This is the baseline expected value for body fatness. Subsequently, the deviation values ​​are segmented and scored according to a preset deviation tolerance. The segmentation rules are as follows: when... When the developmental consistency score is full (100 points), it indicates that the individual's developmental status is consistent with the baseline height under the same stage and water temperature conditions; when When the developmental consistency score decreases linearly (from 100 to 60 points), it indicates that there is some deviation in individual development but it is still within a reasonable range; when When the developmental consistency score drops to the low range (<60 points), it indicates a significant deviation from the baseline developmental status, potentially suggesting malnutrition, disease, or developmental abnormalities. Figure 5 As shown in the figure, the horizontal axis represents the percentage of relative deviation. (0%~30%), with the vertical axis representing developmental consistency score (0~100 points). Green area ( A score of 100 indicates highly consistent development; the yellow area ( A score that decreases linearly from 60 to 100 indicates a certain deviation; the red area ( A score <60 indicates a significant deviation. The blue solid line represents the segmented score mapping curve, and the scatter plots represent the individual sample distribution. The developmental consistency score, as a core intermediate variable for assessing individual developmental level, provides a foundation for subsequent fusion correction.

[0039] Step S440: The developmental consistency score is fused and corrected based on the key gonadal characteristic parameters. The deviation between the quantitative value of the key gonadal characteristic parameters and the standard value of the same developmental stage is used as a correction factor and superimposed on the developmental consistency score. The developmental stage determination result is mapped to the corresponding standard value of gonadal maturity coefficient. The developmental consistency score is then weighted and corrected to obtain a quantitative assessment index of the individual developmental level of Qinghai Lake naked carp.

[0040] It should be noted that in step S440, the developmental consistency score relying solely on plumpness does not fully consider the histological characteristics of the gonads themselves. Therefore, the key gonadal characteristic parameters extracted in step S300 need to be incorporated into the correction. Specifically, firstly, the deviation between the quantitative values ​​of the key gonadal characteristic parameters and the standard values ​​for the same developmental stage is calculated as a correction factor and added to the developmental consistency score, so that the final score simultaneously reflects the degree of consistency between morphological indicators and gonadal tissue characteristics. Subsequently, the gonadal developmental stage determination results are mapped to standard values ​​of gonadal maturity coefficients: Stage I (undeveloped gonads) 0.1, Stage II (gonadal development begins) 0.3, Stage III (rapid gonadal development) 0.5, Stage IV (pre-mature gonadal development) 0.8, Stage V (mature gonadal development) 1.0. Using this maturity coefficient as a weight, the corrected developmental consistency score is weighted and corrected to obtain the final quantitative assessment index of the individual developmental level of the Qinghai Lake naked carp. This indicator integrates multidimensional information such as individual morphology (fattening), environmental correction (water temperature grading), gonadal characteristics (key characteristic parameters), and developmental stage (maturity coefficient), and can comprehensively and objectively characterize an individual's current developmental level.

[0041] The aforementioned fusion correction process ensures the comparability of assessment results for different individuals within the same developmental stage. At the same time, water temperature correction eliminates the interference of seasonal environmental factors on the assessment results, enabling assessment indicators obtained at different sampling times to be tracked longitudinally and compared laterally within a unified framework.

[0042] Step S500: Analyze the population development status based on the quantitative evaluation indicators of multiple individuals within the same sampling batch to obtain the overall evaluation results of the population gonadal development status. Understandably, this step elevates the analytical dimension from the individual level to the population level. Its purpose is to assess the reproductive readiness of the current batch of Qinghai Lake naked carp from the perspective of the entire population, providing a macro-level reference for artificial breeding decisions. Simply statistically analyzing individual developmental stages cannot reflect the differences in developmental quality between individuals. For example, two individuals may both be classified as stage IV (pre-gonadal maturation), but their quantitative assessment indicators may differ significantly; one may be in excellent developmental condition, while the other is in poor condition. Merely statistically analyzing the distribution ratio of stages will not distinguish this quality difference. Therefore, this step uses weighted aggregation based on individual quantitative assessment indicators, allowing individuals with better developmental status to contribute more weight in the population assessment, thus more accurately reflecting the overall reproductive readiness of the population.

[0043] Further, step S500 includes steps S510 to S520.

[0044] Step S510: Based on the quantitative evaluation indicators of each individual in the same sampling batch, obtain the stage distribution ratio vector by statistically analyzing the distribution ratio of the glandular development stage of each individual in each stage. Specifically, step S510 calculates the distribution ratio of each individual's gonadal development stage across the five stages (stages I to V) for all individuals within the same sampling batch. For example, if some individuals in a batch have reached reproductive conditions (stage IV and above), some are in the rapid gonadal development stage (stage III), and some are in the undeveloped gonadal stage (stages I to II), then the stage distribution ratio vector intuitively reflects the numerical composition of individuals at each developmental stage in the population.

[0045] Step S520: Based on the stage distribution ratio vector and the quantitative evaluation indicators of each individual, perform weighted aggregation. By using the gonadal maturity coefficient contained in the quantitative evaluation indicators of each individual as the weight, the stage distribution ratio vector is weighted and summed to obtain the overall status evaluation result of the gonadal development of the group.

[0046] It should be noted that in step S520, the weighted aggregation is implemented by using the gonadal maturity coefficient included in the quantitative evaluation indicators of each individual as a weight to perform weighted statistics on the distribution of the number of individuals within each stage. The core idea is that individuals with higher maturity coefficients (i.e., those that are more mature and have better developmental quality) receive a greater contribution in the group evaluation. The formula for weighted aggregation is: ; In the formula, The overall state assessment results of the group For the first The weight of the gonadal maturity coefficient for each individual. For the first The quantitative assessment index values ​​for each individual. Through this weighted aggregation method, the final overall status assessment result is a scalar value that comprehensively reflects the developmental progress and quality of the group, facilitating horizontal comparisons and vertical tracking between different batches.

[0047] Step S600: Based on the overall assessment results of the gonadal development of the population and the water temperature data, predict the timing of reproduction and obtain the prediction results of the best timing for artificial reproduction.

[0048] Understandably, this step is the final output of the entire method, aiming to scientifically predict the optimal timing for artificial breeding of Qinghai Lake naked carp based on population development status assessment and environmental temperature data. As a plateau cold-water fish, the gonadal development and reproductive behavior of Qinghai Lake naked carp are highly dependent on the cumulative effect of environmental water temperature, rather than simply calendar time. Therefore, this step uses an effective accumulated temperature model to quantify the driving effect of temperature on the developmental process. The greater the effective accumulated temperature, the more heat has been accumulated for gonadal development, and the closer to the optimal breeding time. Simultaneously, considering that Qinghai Lake naked carp are distributed across multiple different water areas in the upper reaches of the Yellow River (such as the upper Yellow River, Qinghai Lake, and Eling Lake), the water temperature accumulation process may differ significantly at the same calendar time due to differences in altitude, latitude, and local climate conditions. Therefore, a cross-water area correction mechanism is introduced to eliminate the influence of geographical location factors on prediction accuracy.

[0049] Further, step S600 includes steps S610 to S630.

[0050] Step S610: Calculate the effective accumulated temperature based on the sampling date and daily water temperature data. The effective accumulated temperature at the current sampling time is obtained by adding the difference between the daily average water temperature and the baseline temperature daily, with the effective accumulated temperature starting temperature for gonadal development of Qinghai Lake naked carp as the baseline. Specifically, in step S610, the calculation of the effective accumulated temperature adopts the standard accumulated temperature model: ; In the formula, where Effective accumulated temperature (unit: daily degrees Celsius). The daily average water temperature The effective accumulated temperature is the starting temperature. The effective accumulated temperature for gonadal development in the Qinghai Lake naked carp is approximately 5℃, meaning that effective accumulation only begins when the daily average water temperature exceeds 5℃; temperatures below 5℃ are not included in the accumulated temperature calculation. This starting temperature is based on physiological studies of the Qinghai Lake naked carp and reflects the minimum temperature threshold for gonadal development in this species. The effective accumulated temperature is calculated by adding it daily from the date the water temperature rises to the starting temperature each spring until the sampling date, obtaining the accumulated effective temperature at the current sampling time.

[0051] Step S620: Based on the effective accumulated temperature and overall status assessment results, conduct a historical comparison analysis. By comparing the effective accumulated temperature with the effective accumulated temperature threshold when the population reaches the optimal reproductive maturity in the same period of each historical year, the deviation of the current accumulated temperature progress from the historical optimal reproductive time can be obtained. It should be noted that in step S620, the effective accumulated temperature at the current sampling time only reflects the temperature-driven developmental progress and has not yet been compared with historical reproductive data. Therefore, this step compares the current effective accumulated temperature with the effective accumulated temperature thresholds for the same period in previous years when the population reached optimal reproductive maturity. The historical data accumulation period is 3 to 5 years. By tracking and monitoring the breeding season of each year, the effective accumulated temperature corresponding to the population reaching optimal reproductive maturity is recorded, and this accumulated temperature is used as the effective accumulated temperature threshold for each year. The deviation of the current accumulated temperature progress from the historical threshold can be positive or negative: a positive value indicates that the current accumulated temperature progress is ahead of the historical level, and the breeding time may be earlier; a negative value indicates that the current accumulated temperature progress is behind the historical level, and the breeding time may be delayed. The deviation is calculated as follows: ; In the formula, This represents the percentage of accumulated temperature deviation. This represents the effective accumulated temperature at the current sampling time. This is the effective accumulated temperature threshold for the same historical period.

[0052] Step S630: Based on the degree of deviation and the sampling water area identifier, perform cross-water area correction and breeding timing prediction. By querying the pre-built cross-water area development progress difference matrix, obtain the water area correction coefficient corresponding to the sampling water area. After correcting the degree of deviation with the water area correction coefficient, map it to the predicted number of days away from the optimal breeding window to obtain the prediction result of the optimal timing for artificial breeding.

[0053] Specifically, in step S630, since the Qinghai Lake naked carp is distributed across multiple waters in the upper reaches of the Yellow River (such as the upper Yellow River, Qinghai Lake, and Eling Lake), the effective accumulated temperature accumulation process under the same calendar date exhibits systematic deviations due to differences in altitude, latitude, and local topography affecting the climate. For example, the gonadal development characteristics of Qinghai Lake naked carp differ across different waters. To eliminate these geographical differences, a cross-water area development progress difference matrix is ​​pre-built. This matrix uses water area identifiers as indexes and stores the development progress difference coefficients of each water area relative to a reference water area. After querying this matrix to obtain the water area correction coefficient corresponding to the currently sampled water area, the deviation calculated in step S620 is corrected using this coefficient to obtain the geographically corrected effective deviation. Subsequently, based on the daily average accumulation rate of effective accumulated temperature (derived from historical data statistics), the corrected effective deviation is converted into the predicted number of days until the optimal breeding window, and an evaluation report is output to clarify the optimal spawning time and reproductive health status, providing data support for artificial breeding.

[0054] The entire processing flow is designed in a progressive manner, including in vivo non-destructive data acquisition, multi-task network segmentation and classification, key feature extraction with confidence screening, individual quantitative assessment with water temperature correction, weighted aggregation population analysis, and effective accumulated temperature-driven prediction of breeding timing. This forms a complete technical chain from individual image acquisition to population development assessment and breeding timing prediction. Employing low-stress electroanesthesia combined with minimally invasive image acquisition, it eliminates the need for sample slaughter and tissue or blood sample collection, enabling continuous monitoring and assessment of live Qinghai Lake naked carp. This avoids the damage to samples caused by traditional methods, effectively protecting the Qinghai Lake naked carp population and resolving the technical pain point of the conflict between traditional dissection methods and germplasm resource conservation. The entire detection process is automated, with a single sample detection time of ≤30s. No professional operators are required, enabling rapid monitoring and assessment of large numbers of Qinghai Lake naked carp. This solves the problems of low efficiency and inability to be applied on a large scale by traditional methods, providing systematic technical support for the conservation and artificial breeding of Qinghai Lake naked carp germplasm resources.

[0055] Example 2: like Figure 2 As shown, this embodiment provides an image recognition-based gonadal development assessment system for *Gymnocypris qinghai Lake*. The system includes: The acquisition module 901 is used to acquire images of the gonads inside the Qinghai Lake naked carp collected from the genital opening of the carp after electroanesthesia using a minimally invasive image acquisition device, as well as the basic biological parameters of the corresponding individual and the water temperature data at the time of sampling. The determination module 902 is used to segment the gonadal region and classify the developmental stage based on the internal image of the gonad. The internal image of the gonad is input into a pre-trained multi-task feature extraction network. The segmentation branch of the multi-task feature extraction network outputs the gonadal region mask, and the classification branch outputs the classification probability of each developmental stage. The stage corresponding to the highest probability is taken as the gonadal developmental stage determination result. The extraction module 903 is used to extract features based on the gonadal development stage determination results. By cropping the corresponding gonadal region image block using a gonadal region mask, and extracting oocyte morphological texture features and testis texture features within the image block, key gonadal feature parameters are obtained. The assessment module 904 is used to assess the individual development level based on key gonadal characteristic parameters, gonadal development stage determination results, basic biological parameters and water temperature data, and obtain quantitative assessment indicators of the individual development level of Qinghai Lake naked carp. Analysis module 905 is used to analyze the developmental status of a population based on quantitative evaluation indicators of multiple individuals within the same sampling batch, and to obtain the overall assessment results of the gonadal development status of the population. The prediction module 906 is used to predict the timing of artificial breeding based on the overall assessment results of the gonadal development of the population and water temperature data.

[0056] In one specific embodiment of this application, the determination module 902 includes: The first determination unit is used to extract multi-scale semantic features based on the internal image of the gonads. The shared coding module of the multi-task feature extraction network extracts the shallow texture details and deep semantic abstract features of the image layer by layer. The shallow texture details include the size and density distribution features of oocytes and the texture clarity and color features of testes. The deep semantic abstract features include the spatial morphological structure features of gonadal tissue, resulting in multi-level feature mapping. The second determination unit is used to segment the gonad region based on the multi-level feature mapping. It upsamples the deep semantic features to restore them to the original resolution space through the segmentation and decoding branch of the multi-task feature extraction network and combines them with the edge details in the shallow features to perform feature concatenation, and outputs a pixel-by-pixel gonad region mask. The third determination unit is used to correct the segmentation morphology deviation based on the gonadal region mask. By performing adversarial discrimination correction on the boundary smoothness and regional connectivity of the gonadal region mask, the segmentation interference of non-gonadal tissues is suppressed, and the gonadal region mask after morphology deviation correction is obtained. The fourth determination unit is used to classify developmental stages based on multi-level feature mapping. After spatial compression of deep semantic features by the classification branch of the multi-task feature extraction network in the global average pooling layer, the fully connected mapping outputs the classification probability vector corresponding to each developmental stage. The stage corresponding to the highest probability is taken as the determination result of gonadal development stage.

[0057] In one specific embodiment of this application, the extraction module 903 includes: The first extraction unit is used to cut out the corresponding gonadal region image block according to the gonadal region mask and perform quality verification. By using the highest probability value in the classification probability vector as the confidence level and comparing it with the preset effective threshold, qualified image blocks with confidence levels reaching the effective threshold are selected to obtain qualified gonadal region image blocks. The second extraction unit is used to extract female features based on qualified gonadal region image blocks and gonadal development stage determination results. By detecting oocyte outlines in the image blocks and calculating the average diameter of oocytes, the density ratio of oocyte area to the total area of ​​gonadal region, and the texture index of oocyte arrangement regularity calculated based on gray-level co-occurrence matrix, the key gonadal feature parameters of female individuals are obtained. The third extraction unit is used to extract male characteristics based on qualified gonadal region image blocks and gonadal development stage determination results. By calculating the second-order statistics of energy, contrast and inverse difference moment of the gray-level co-occurrence matrix of the image block in multiple directions, and combining the morphological edge sharpness index of the testis lobule structure, the key gonadal characteristic parameters of male individuals are obtained.

[0058] In one specific embodiment of this application, the evaluation module 904 includes: The first assessment unit is used to calculate the individual body fat index based on the body length and weight data in the basic biological parameters. The individual body fat index is obtained by calculating the ratio of the individual's actual weight to the standard weight reference value under the same body length conditions. The second evaluation unit is used to determine the water temperature range to which the current sampling time belongs based on the water temperature data. By matching the developmental stage judgment result and the water temperature range in the pre-built water temperature range fertility benchmark lookup table, the expected value of fertility benchmark under the corresponding conditions is obtained. The third assessment unit is used to calculate and map the degree of deviation based on the individual body condition index and the expected value of body condition benchmark. By calculating the relative percentage deviation between the two and performing segmented scoring mapping according to the preset deviation tolerance, a developmental consistency score is obtained. The fourth assessment unit is used to integrate and correct the developmental consistency score based on the key gonadal characteristic parameters. By using the deviation between the quantitative value of the key gonadal characteristic parameters and the standard value of the same developmental stage as a correction factor, it is superimposed on the developmental consistency score. The developmental stage determination result is mapped to the corresponding standard value of gonadal maturity coefficient. The weighted correction of the developmental consistency score is then used to obtain a quantitative assessment index of the individual developmental level of Qinghai Lake naked carp.

[0059] In one specific embodiment of this application, the analysis module 905 includes: The first analysis unit is used to obtain the stage distribution ratio vector by statistically analyzing the distribution ratio of the glandular development stage of each individual in each stage based on the quantitative evaluation index of each individual in the same sampling batch. The second analysis unit is used to perform weighted aggregation based on the stage distribution ratio vector and the quantitative evaluation indicators of each individual. By using the gonadal maturity coefficient contained in the quantitative evaluation indicators of each individual as the weight, the stage distribution ratio vector is weighted and summed to obtain the overall status evaluation result of the gonadal development of the group.

[0060] In one specific embodiment of this application, the prediction module 906 includes: The first prediction unit is used to calculate the effective accumulated temperature based on the sampling date and daily water temperature data. It calculates the effective accumulated temperature at the current sampling time by accumulating the difference between the daily average water temperature and the baseline temperature, using the effective accumulated temperature starting temperature for gonadal development of Qinghai Lake naked carp as the baseline. The second prediction unit is used to conduct historical comparison analysis based on the effective accumulated temperature and the overall status assessment results. By comparing the effective accumulated temperature with the effective accumulated temperature threshold when the population reaches the best reproductive maturity in the same period of each historical year, the deviation of the current accumulated temperature progress from the historical best reproductive time is obtained. The third prediction unit is used to perform cross-water area correction and breeding timing prediction based on the degree of deviation and the sampling water area identifier. It obtains the water area correction coefficient corresponding to the sampling water area by querying the pre-built cross-water area development progress difference matrix, and maps the degree of deviation to the predicted number of days away from the optimal breeding window after being corrected by the water area correction coefficient, thus obtaining the prediction result of the optimal timing for artificial breeding.

[0061] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for assessing gonadal development in *Gymnocypris qinghai Lake* based on image recognition, characterized in that... include: Images of the gonads inside the naked carp of Qinghai Lake were obtained from the genital opening of the carp after electroanesthesia using a minimally invasive image acquisition device, along with the basic biological parameters of the corresponding individuals and water temperature data at the time of sampling. Gonadal region segmentation and developmental stage classification are performed based on the internal gonadal image. The internal gonadal image is input into a pre-trained multi-task feature extraction network. The segmentation branch of the multi-task feature extraction network outputs a gonadal region mask, and the classification branch outputs the classification probability of each developmental stage. The stage corresponding to the highest probability is taken as the gonadal developmental stage determination result. Based on the gonadal development stage determination results, feature extraction is performed. By cropping the corresponding gonadal region image block from the gonadal region mask, oocyte morphological texture features and testis texture features are extracted from the image block respectively to obtain key gonadal feature parameters. Based on the key gonadal characteristic parameters, the gonadal development stage determination results, the basic biological parameters, and the water temperature data, the individual development level was assessed to obtain a quantitative assessment index of the individual development level of the Qinghai Lake naked carp. The population development status was analyzed based on the quantitative evaluation indicators of multiple individuals within the same sampling batch, and the overall status evaluation results of the population gonadal development were obtained. Based on the overall assessment results of the gonadal development of the population and the water temperature data, the timing of reproduction is predicted, and the prediction results of the optimal timing for artificial reproduction are obtained. Specifically, feature extraction is performed based on the gonadal development stage determination results. This involves cropping corresponding gonadal region image blocks from the gonadal region mask and extracting oocyte morphological texture features and testis texture features from these image blocks to obtain key gonadal feature parameters, including: Based on the gonadal region mask, the corresponding gonadal region image block is cropped and quality is checked. By using the highest probability value in the classification probability as the confidence level and comparing it with a preset effective threshold, qualified image blocks with a confidence level reaching the effective threshold are selected to obtain qualified gonadal region image blocks. Based on the qualified gonadal region image blocks and the gonadal development stage determination results, female features are extracted. By detecting the oocyte outline in the image blocks and calculating the average diameter of the oocytes, the density ratio of the oocyte area to the total area of ​​the gonadal region, and the texture index of the regularity of oocyte arrangement based on the gray-level co-occurrence matrix, the key gonadal feature parameters of the female individual are obtained. Male features are extracted based on the qualified gonadal region image blocks and the gonadal development stage determination results. By calculating the second-order statistics of energy, contrast and inverse difference moment of the gray-level co-occurrence matrix of the image blocks in multiple directions, and combining them with the morphological edge sharpness index of the testis lobule structure, the key gonadal feature parameters of male individuals are obtained.

2. The method for assessing gonadal development of *Gymnocypris qinghai Lake* based on image recognition according to claim 1, characterized in that, Based on the internal gonadal images, gonadal region segmentation and developmental stage classification are performed. This involves inputting the internal gonadal images into a pre-trained multi-task feature extraction network. The segmentation branch of the multi-task feature extraction network outputs a gonadal region mask, and the classification branch outputs the classification probability of each developmental stage. The stage corresponding to the highest probability is taken as the gonadal developmental stage determination result, including: Multi-scale semantic feature extraction is performed based on the internal image of the gonads. The shallow texture detail features and deep semantic abstract features of the image are extracted layer by layer based on the shared coding module of the multi-task feature extraction network. The shallow texture detail features include the size and density distribution features of oocytes and the texture clarity and color features of testes. The deep semantic abstract features include the spatial morphological structure features of gonadal tissue, resulting in multi-level feature mapping. Gonadal region segmentation is performed based on the multi-level feature mapping. The deep semantic features are upsampled and restored to the original resolution space through the segmentation and decoding branch of the multi-task feature extraction network. The features are then combined with the edge details in the shallow features for feature concatenation, and a pixel-by-pixel gonadal region mask is output. Based on the gonadal region mask, segmentation morphology deviation correction is performed. By performing adversarial discrimination correction on the boundary smoothness and regional connectivity of the gonadal region mask, segmentation interference from non-gonadal tissues is suppressed, and a gonadal region mask with morphology deviation correction is obtained. Developmental stages are classified according to the multi-level feature mapping. The classification branch of the multi-task feature extraction network performs spatial compression of deep semantic features in the global average pooling layer and then outputs the classification probability vector corresponding to each developmental stage through fully connected mapping. The stage corresponding to the highest probability is taken as the result of the gonadal development stage determination.

3. The method for assessing gonadal development of *Gymnocypris qinghai Lake* based on image recognition according to claim 1, characterized in that, Based on the key gonadal characteristic parameters, the gonadal development stage determination results, the basic biological parameters, and the water temperature data, an individual developmental level assessment was conducted to obtain quantitative assessment indicators for the individual developmental level of the Qinghai Lake naked carp, including: The body length and weight data in the basic biological parameters are used to calculate the body fat index. The body fat index is obtained by calculating the ratio of the individual's actual weight to the standard weight reference value under the same body length. Based on the water temperature data, the water temperature range to which the current sampling time belongs is determined. By matching the developmental stage determination result and the water temperature range in the pre-built water temperature range fertility benchmark lookup table, the expected value of fertility benchmark under the corresponding conditions is obtained. The degree of deviation is calculated and scored based on the individual body condition index and the expected body condition baseline. The developmental consistency score is obtained by calculating the relative percentage deviation between the two and performing segmented scoring based on the preset deviation tolerance. The developmental consistency score is fused and corrected based on the key gonadal characteristic parameters. The deviation between the quantitative value of the key gonadal characteristic parameters and the standard value of the same developmental stage is used as a correction factor and superimposed on the developmental consistency score. The developmental stage determination result is mapped to the corresponding standard value of gonadal maturity coefficient. The developmental consistency score is then weighted and corrected to obtain a quantitative assessment index of the individual developmental level of the Qinghai Lake naked carp.

4. The method for assessing gonadal development of *Gymnocypris qinghai Lake* based on image recognition according to claim 1, characterized in that, Based on the quantitative assessment indicators of multiple individuals within the same sampling batch, a population developmental status analysis was conducted to obtain the overall assessment results of the population gonadal development status, including: Based on the quantitative evaluation indicators of each individual in the same sampling batch, the distribution ratio vector of each stage of glandular development is obtained by statistically analyzing the distribution ratio of each individual's glandular development stage in each stage. The overall status assessment result of the gonadal development of the group is obtained by weighting and aggregating the stage distribution ratio vector and the quantitative evaluation index of each individual, using the gonadal maturity coefficient included in the quantitative evaluation index of each individual as the weight.

5. A gonadal development assessment system for *Gymnocypris qinghai Lake* based on image recognition, characterized in that, include: The acquisition module is used to acquire images of the gonads inside the naked carp of Qinghai Lake through a minimally invasive image acquisition device after electroanesthesia, as well as the basic biological parameters of the corresponding individual and the water temperature data at the time of sampling. The determination module is used to segment the gonadal region and classify the developmental stage based on the internal gonadal image. The internal gonadal image is input into a pre-trained multi-task feature extraction network. The segmentation branch of the multi-task feature extraction network outputs a gonadal region mask, and the classification branch outputs the classification probability of each developmental stage. The stage with the highest probability is taken as the gonadal developmental stage determination result. The extraction module is used to extract features based on the gonadal development stage determination results. It obtains key gonadal feature parameters by cropping the corresponding gonadal region image block from the gonadal region mask and extracting oocyte morphological texture features and testis texture features from the image block. The assessment module is used to assess the individual development level of the Qinghai Lake naked carp based on the key gonadal characteristic parameters, the gonadal development stage determination results, the basic biological parameters and the water temperature data, and to obtain a quantitative assessment index of the individual development level of the Qinghai Lake naked carp. The analysis module is used to analyze the population development status based on the quantitative evaluation indicators of multiple individuals within the same sampling batch, and to obtain the overall status evaluation results of the population gonadal development. The prediction module is used to predict the timing of reproduction based on the overall assessment results of the gonadal development of the population and the water temperature data, so as to obtain the prediction result of the optimal timing for artificial reproduction. The extraction module includes: The first extraction unit is used to cut out the corresponding gonadal region image block according to the gonadal region mask and perform quality verification. By using the highest probability value in the classification probability as the confidence level and comparing it with a preset effective threshold, qualified image blocks with a confidence level reaching the effective threshold are selected to obtain qualified gonadal region image blocks. The second extraction unit is used to extract female features based on the qualified gonadal region image block and the gonadal development stage determination result. By detecting the oocyte outline in the image block and calculating the average diameter of the oocyte, the density ratio of the oocyte area to the total area of ​​the gonadal region, and the texture index of the regularity of oocyte arrangement based on the gray-level co-occurrence matrix, the key gonadal feature parameters of the female individual are obtained. The third extraction unit is used to extract male features based on the qualified gonadal region image block and the gonadal development stage determination result. By calculating the second-order statistics of energy, contrast and inverse difference moment of the gray-level co-occurrence matrix of the image block in multiple directions, and combining the morphological edge sharpness index of the testis lobule structure, the key gonadal feature parameters of the male individual are obtained.

6. The image recognition-based gonadal development assessment system for *Gymnocypris qinghai Lake* according to claim 5, characterized in that, The determination module includes: The first determination unit is used to extract multi-scale semantic features based on the internal image of the gonads. Based on the shared coding module of the multi-task feature extraction network, it extracts the shallow texture detail features and deep semantic abstract features of the image layer by layer. The shallow texture detail features include the size and density distribution features of oocytes and the texture clarity and color features of testes. The deep semantic abstract features include the spatial morphological structure features of gonadal tissue, thus obtaining multi-level feature mapping. The second determination unit is used to segment the gonad region according to the multi-level feature mapping. The deep semantic features are upsampled and restored to the original resolution space through the segmentation and decoding branch of the multi-task feature extraction network, and the edge details in the shallow features are combined to perform feature concatenation, and the gonad region mask is output pixel by pixel. The third determination unit is used to perform segmentation morphology deviation correction based on the gonadal region mask. By performing adversarial discrimination correction on the boundary smoothness and regional connectivity of the gonadal region mask, segmentation interference from non-gonadal tissues is suppressed, and a gonadal region mask after morphology deviation correction is obtained. The fourth determination unit is used to classify developmental stages based on the multi-level feature mapping. After spatial compression of deep semantic features by the classification branch of the multi-task feature extraction network in the global average pooling layer, the classification probability vector corresponding to each developmental stage is output through fully connected mapping. The stage corresponding to the highest probability is taken as the determination result of the gonadal development stage.

7. The image recognition-based gonadal development assessment system for *Gymnocypris qinghai Lake* according to claim 5, characterized in that, The evaluation module includes: The first assessment unit is used to calculate the individual body fat index based on the body length and weight data in the basic biological parameters. The individual body fat index is obtained by calculating the ratio of the individual's actual weight to the standard weight reference value under the same body length conditions. The second evaluation unit is used to determine the water temperature range to which the current sampling time belongs based on the water temperature data, and to obtain the expected value of the fertility benchmark under the corresponding conditions by matching the developmental stage determination result and the water temperature range in the pre-built water temperature range fertility benchmark lookup table. The third assessment unit is used to calculate and score the degree of deviation based on the individual fullness index and the fullness baseline expected value. By calculating the relative percentage deviation between the two and performing segmented scoring based on the preset deviation tolerance, a developmental consistency score is obtained. The fourth evaluation unit is used to fuse and correct the developmental consistency score based on the gonadal key characteristic parameters. By superimposing the deviation between the quantitative value of the gonadal key characteristic parameters and the standard value of the same developmental stage as a correction factor into the developmental consistency score, and mapping the developmental stage determination result to the corresponding gonadal maturity coefficient standard value, the developmental consistency score is weighted and corrected to obtain a quantitative evaluation index of the individual developmental level of the Qinghai Lake naked carp.

8. The image recognition-based gonadal development assessment system for *Gymnocypris qinghai Lake* according to claim 5, characterized in that, The analysis module includes: The first analysis unit is used to obtain a stage distribution ratio vector by statistically analyzing the distribution ratio of the glandular development stage of each individual in each stage according to the quantitative evaluation index of each individual in the same sampling batch. The second analysis unit is used to perform weighted aggregation based on the stage distribution ratio vector and the quantitative evaluation index of each individual. By using the gonadal maturity coefficient included in the quantitative evaluation index of each individual as a weight, the stage distribution ratio vector is weighted and summed to obtain the overall status evaluation result of the gonadal development of the group.

Images