A method for identifying sick juvenile fish based on an improved YOLOv10 model

Abstract

This paper proposes a method for identifying sick juvenile fish based on an improved YOLOv10 model. The method includes collecting images of fish in a water area as images to be detected, preprocessing the images to obtain an image dataset, constructing a YOLOv10-UDFishNet model, inputting the image dataset into the YOLOv10-UDFishNet model, and training the YOLOv10-UDFishNet model using the image dataset to obtain a trained YOLOv10-UDFishNet model. Finally, the images of fish in the water area to be detected are input into the trained YOLOv10-UDFishNet model to obtain fish detection results and location information of individual fish targets. This method can accurately and efficiently identify sick juvenile fish in turbid underwater environments.

Description

Technical Field

[0001] The invention belongs to the technical field of intelligent identification, and specifically discloses a method for identifying diseased juvenile fish based on an improved YOLOv10 model.

[0002] Background

[0003] As a highly economically valuable farmed fish, the health and disease control of the redfin pufferfish (Takifugu rubripes) directly impact the sustainable development and economic benefits of the aquaculture industry. Therefore, accurately identifying diseased juvenile redfin pufferfish is not only crucial for ensuring fish health and reducing mortality, but also a key driver of aquaculture revenue growth. However, due to the unique characteristics of aquaculture environments, turbid underwater conditions degrade image quality, hindering the accuracy of diseased juvenile fish detection. Furthermore, the small size of juvenile redfin pufferfish further complicates diseased juvenile fish detection. Therefore, developing a diseased juvenile redfin pufferfish detection method that can adapt to turbid underwater environments and effectively address the challenges of diseased juvenile redfin pufferfish detection is crucial for improving aquaculture profitability and promoting the sustainable development of the aquaculture industry.

[0004] Computer vision, with its powerful image processing and pattern recognition capabilities, can accurately extract feature information from target objects, making it possible to accurately identify diseased juvenile fish underwater. With the integration of computer vision and deep learning technologies, some studies have achieved preliminary identification of diseased juvenile fish by capturing changes in fish appearance. However, diagnosing diseased juvenile fish solely based on appearance is often limited. Since fish behavior is a direct indicator of their health, a comprehensive analysis of both appearance and behavior can achieve more accurate detection of diseased juvenile fish. Li et al. fused fish outline information with the original image and used a modified YOLOv5 algorithm to enhance the fish's position in the image, enabling accurate detection of early abnormal fish behavior. However, this method relies on a relatively clean dataset for training, and its detection performance remains insufficient in complex and changing underwater environments. To address the challenges of abnormal individual detection posed by underwater interference, Cai et al. designed the NAM-YOLOv7 network by integrating the NAM Attention mechanism. This network effectively focuses on key appearance and behavioral features of abnormal fish while discarding irrelevant background information, resulting in high detection accuracy. However, this approach is not suitable for aquaculture environments with strong light. Liu et al. introduced deformable convolution into the YOLOv8 model to enhance its behavioral discrimination capabilities in fluctuating lighting conditions, achieving significant detection results. However, they did not consider the adverse effects of other underwater factors on diseased juvenile fish detection. To accurately identify diseased juvenile fish in complex aquaculture environments, Li et al. proposed a model, DDEYOLOv10, based on the YOLOv10 model by introducing the DRNELAN4 module, enhancing the model's global feature extraction capabilities. Although beneficial for diseased juvenile fish detection in turbid scenes, the DRNELAN4 module increases the model's receptive field, resulting in loss of feature detail and reduced detection accuracy for small individual diseased juvenile fish. Wang et al. proposed a diseased juvenile fish detection model, DFYOLO, which combines multiple convolutional kernels into a convolutional kernel group to enhance the network's ability to extract fish surface features. Furthermore, the CBAM module was introduced to suppress background interference, facilitating the precise localization of small targets in turbid underwater environments and achieving better detection of small diseased juvenile fish. However, due to the high similarity between diseased and healthy fish, this method still suffers from a high rate of missed detection of diseased juvenile fish.

[0005] In summary, computer vision technology still faces challenges in underwater diseased larvae detection. The complex underwater environment significantly degrades image quality, greatly limiting the effective extraction of diseased larvae feature information and reducing the accuracy of diseased larvae detection. The small size of diseased larvae also limits the critical information they carry, exacerbating the difficulty of accurately distinguishing diseased larvae from normal fish, resulting in a low recall rate for diseased larvae detection. To address these issues, it is necessary to design a new diseased larvae identification method based on an improved YOLOv10 model to overcome the problems existing in existing diseased larvae identification methods. Summary of the Invention

[0006] In order to solve the problems existing in existing sick juvenile fish identification methods, such as poor extraction of sick juvenile fish feature information in complex water environments and low accuracy of sick juvenile fish detection, the present invention proposes a sick juvenile fish identification method based on an improved YOLOv10 model.

[0007] The present invention provides a method for identifying diseased juvenile fish based on an improved YOLOv10 model, comprising the following steps:

[0008] S1. Collect images of fish in waters as images to be detected, and preprocess the images to be detected to obtain an image dataset;

[0009] S2. Construct a YOLOv10-UDFishNet model, including replacing the strided convolutions of Stage 1 and Stage 2 in the backbone network of the YOLOv10 model with SPD-Conv modules, adding auxiliary branches to the backbone network of the YOLOv10 model, adding the HFG-based fusion feature fusion method to the neck network of the YOLOv10 model, and replacing the loss function of the YOLOv10 model with the SlideLoss loss function, to obtain the YOLOv10-UDFishNet model.

[0010] S3. Input the image dataset obtained after preprocessing in step S1 into the YOLOv10-UDFishNet model obtained in step S2, train the YOLOv10-UDFishNet model using the image dataset, and obtain a trained YOLOv10-UDFishNet model;

[0011] S4. Input the fish image in the water area to be detected into the trained YOLOv10-UDFishNet model obtained in step S3 to obtain the fish detection results and the location information of individual fish targets.

[0012] According to a method for identifying sick juvenile fish based on an improved YOLOv10 model in some embodiments of the present application, in step S1, the preprocessing includes adjusting the size of the image to be detected, and marking both abnormal appearance and abnormal behavior in the image to be detected as abnormal labels, and marking normal individuals as normal labels.

[0013] According to a method for identifying sick juvenile fish based on an improved YOLOv10 model in some embodiments of the present application, in step S2, the SPD-Conv module includes a space-to-depth layer and a non-step convolution layer, and the space-to-depth layer is used to slice the given feature map X(S, S, C1) to obtain scale 2 The size is The sub-feature maps are spliced ​​in the channel dimension to obtain a new feature map

[0014] The slicing operation formula is shown in formula (1):

[0015]

[0016] Among them, f a,b Represents a sub-feature map, a∈(0,1,2…scale-1), b∈(0,1,2…scale-1), S represents the spatial size of the feature map, C1 represents the number of feature map channels, and scale represents the number of sub-feature maps that are split.

[0017] The non-strided convolution layer is used to add a non-strided convolution layer using C2 convolution kernels to obtain the final feature map X" Among them C2 <C1<scale 2 .

[0018] According to a method for identifying sick juvenile fish based on an improved YOLOv10 model in some embodiments of the present application, in step S2, the auxiliary branch includes a first auxiliary branch and a second auxiliary branch, the first auxiliary branch is connected in parallel with Stage 1, and the second auxiliary branch is connected in parallel with Stage 2;

[0019] The first auxiliary branch includes an SPD-Conv module and an improved ShuffleBlock module connected in sequence, and the second auxiliary branch includes an improved ShuffleBlock module; the improved ShuffleBlock module includes an improved feature extraction unit and an improved downsampling unit;

[0020] The improved feature extraction unit replaces the 1×1 convolution after the depth-separable convolution of the feature extraction unit in the basic module ShuffleBlock of ShuffleNetv2 with a partial convolution. The partial convolution performs a standard convolution operation on some input channels to extract spatial features, while the remaining channels remain unchanged. The partial ratio of the partial convolution is r=1 / 2;

[0021] The improved downsampling unit is to replace the depth-separable convolution of the downsampling unit in the basic module ShuffleBlock of ShuffleNetv2 with the SPD-Conv module.

[0022] According to a method for identifying sick juvenile fish based on an improved YOLOv10 model in some embodiments of the present application, in step S2, the HFG-based fusion feature fusion method includes the following steps:

[0023] Step a. Given a low-level feature map X low ∈R H×W×C and high-level feature maps X high ∈R H×W×C , the high-level feature map X high ∈R H×W×C The channels are evenly split into G groups to obtain G group feature maps. If they cannot be evenly divided, they are rounded down.

[0024] Step b. Perform 1×1 Conv operation on each group of feature maps after splitting, so that the number of channels of each group of feature maps is the same as that of the low-level feature map X low The number of channels is the same, and each set of feature maps after dimensionality increase is recorded as X g ;

[0025] Step c. Each set of feature maps X g Respectively with X low After element-by-element addition, the feature maps of each G group are accumulated to obtain the feature map X′ high , as shown in formula (2):

[0026]

[0027] Among them, g∈(1,2…G);

[0028] Step d. The feature map X′ high Perform global average pooling in the spatial dimension to compress the entire spatial information into a global feature, and obtain the C-dimensional global feature. The compression process is shown in formula (3):

[0029]

[0030] Among them, z c ∈R cRepresents the global feature, H represents the height of the feature map, W represents the width of the feature map, u c Represents the input matrix, i represents the row of the input matrix, j represents the column of the input matrix, and c represents the number of channels of the feature map;

[0031] Step e. Pass the global feature through a fully connected layer consisting of a ReLU activation function and a Sigmoid activation function to obtain a weighted output, as shown in formula (4):

[0032] s=σ(W2δ(W1z c )) (4)

[0033] Where s represents the weight output, δ is the ReLU activation function, σ is the Sigmoid activation function, W1(·) represents the fully connected layer operation of the ReLU activation function, and W2(·) represents the fully connected layer operation of the Sigmoid activation function.

[0034] Step f. Combine the weight output with the low-level feature map X low Perform element-by-element multiplication and concatenate it with the high-level feature map along the channel to obtain the final output, as shown in formula (5):

[0035]

[0036] Among them, Out represents the final output, represents element-by-element multiplication; [·,·] represents concatenation operation.

[0037] According to a method for identifying sick juvenile fish based on an improved YOLOv10 model in some embodiments of the present application, in step S2, the SlideLoss loss function is as shown in formula (6):

[0038]

[0039] Among them, I n represents the loss of the nth category, n∈(1,2…n), f(x) represents the addition of the Slide weighting function on the basis of the baseline model loss function, w n represents the weight factor, y n represents the nth actual value, σ represents the Sigmoid function, x n Represents the nth predicted value, and x represents the function input;

[0040] Add the Slide weighting function f(x) to the baseline model loss function as shown in formula (7):

[0041]

[0042] Where μ represents the average of the intersection-over-union (IoU) between all predicted boxes and the true boxes.

[0043] According to some embodiments of the present application, a method for identifying diseased juvenile fish based on an improved YOLOv10 model, in step S3, the improved YOLOv10-UDFishNet model is trained based on the image dataset until the training is completed when a set learning round is reached, thereby obtaining a trained YOLOv10-UDFishNet model.

[0044] According to a method for identifying sick juvenile fish based on an improved YOLOv10 model in some embodiments of the present application, the fish detection results obtained in step S4 include normal and abnormal.

[0045] The present invention proposes a method for identifying sick juvenile fish based on an improved YOLOv10 model, which can accurately and efficiently identify sick juvenile fish in turbid underwater environments. By designing an auxiliary branch based on the backbone network, the problem of interference of turbid water background on feature extraction is solved, the network's feature extraction capability in low-quality images is enhanced, and the detection effect of underwater sick juvenile fish is improved. The SPD-Conv module is used to reduce the loss of small target juvenile fish information caused by the downsampling process. In the neck network, an adaptive fusion method guided by high-level features, namely the HFG-based fusion feature fusion method, is proposed to achieve efficient reuse of low-level sick juvenile fish features of the network, and improve the resolution ability of small target sick juvenile fish. By introducing the SlideLoss loss function to adaptively adjust the sample weights, the attention paid to sick juvenile fish samples during training is increased, and the detection performance of sick juvenile fish is further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] Figure 1 Schematic diagram of a flow chart of a method for identifying sick juvenile fish based on an improved YOLOv10 model of the present invention;

[0047] Figure 2 Schematic diagram of the structure of the YOLOv10-UDFishNet model in Example 2 of the present invention;

[0048] Figure 3 Schematic diagram of the SPD-Conv module structure in Example 2 of the present invention;

[0049] Figure 4 The following are visualization diagrams of the feature maps before and after Stage 2 of the backbone network in Example 2 of the present invention: (a) is a visualization diagram of the feature map before Stage 2, and (b) is a visualization diagram of the feature map after Stage 2;

[0050] Figure 5(a) is a schematic diagram of the structure of the feature extraction unit of ShuffleBlock in Example 3 of the present invention, (b) is a schematic diagram of the structure of the downsampling unit of ShuffleBlock in Example 3 of the present invention, (c) is a schematic diagram of the structure of the improved feature extraction unit of the improved ShuffleBlock module in Example 3 of the present invention, and (d) is a schematic diagram of the structure of the improved downsampling unit of the improved ShuffleBlock module in Example 3 of the present invention;

[0051] Figure 6 Schematic diagram of the structure of partial convolution in Example 2 of the present invention;

[0052] Figure 7 Schematic diagram of the structure of the HFG-based fusion feature fusion method in Example 2 of the present invention;

[0053] Figure 8 Figure 3 is a heat map of the detection effects of different models on small target diseased juvenile fish in Example 3 of the present invention. (a) is a heat map of the detection effect of the original image, (b) is a heat map of the detection effect of RT-DETR, (c) is a heat map of the detection effect of YOLOv8, (d) is a heat map of the detection effect of YOLOv9, (e) is a heat map of the detection effect of YOLOv10, (f) is a heat map of the detection effect of YOLOv8-Head-ECAM, (g) is a heat map of the detection effect of DDEYOLOv9, (h) is a heat map of the detection effect of DFYOLO, and (i) is a heat map of the detection effect of YOLOv10-UDFishNet. DETAILED DESCRIPTION

[0054] The following embodiments of the present invention are described in further detail with reference to the accompanying drawings and examples. The following examples are used to illustrate the present invention but are not intended to limit the scope of the present invention.

[0055] Example 1: This example provides a method for identifying sick juvenile fish based on an improved YOLOv10 model. Figure 1 As shown, the following steps are included:

[0056] S1. Collect images of fish in waters as images to be detected, and preprocess the images to be detected to obtain an image dataset;

[0057] S2. Build the YOLOv10-UDFishNet model. This involves replacing the strided convolutions in Stage 1 and Stage 2 of the YOLOv10 backbone network with SPD-Conv modules, adding auxiliary branches to the YOLOv10 backbone network, adding the HFG-based fusion feature fusion method to the YOLOv10 neck network, and replacing the YOLOv10 loss function with the SlideLoss loss function. This results in the YOLOv10-UDFishNet model.

[0058] S3. Input the image dataset obtained after preprocessing in step S1 into the YOLOv10-UDFishNet model obtained in step S2, train the YOLOv10-UDFishNet model using the image dataset, and obtain a trained YOLOv10-UDFishNet model;

[0059] S4. Input the fish image in the water area to be detected into the trained YOLOv10-UDFishNet model obtained in step S3 to obtain the fish detection results and the location information of individual fish targets.

[0060] Example 2: This example provides a method for identifying diseased juvenile fish based on an improved YOLOv10 model.

[0061] Due to the turbid underwater environment and the small size of the juveniles, it is difficult to detect sick juvenile redfin pufferfish. In addition, in order to ensure real-time detection of sick juveniles so as to quickly adjust the corresponding breeding strategies, a high recognition speed is essential for the model. Therefore, the selected model must have both powerful feature extraction capabilities and high real-time response. YOLOv10 is the latest version of the YOLO series algorithm, which surpasses previous YOLO versions and other mainstream models of the same period in terms of the trade-off between accuracy and detection speed. In the YOLOv10 series, YOLOv10x has introduced more convolutional layers and residual connections, showing the highest detection accuracy. Therefore, YOLOv10x is used as the baseline model in this embodiment.

[0062] The YOLOv10 model consists of three parts: the input, the backbone, and the head network. The input unifies images to a resolution of 640×640. The backbone network consists of a convolutional neural network for extracting image features. It utilizes the C2f and C2fCIB structures to obtain richer gradient flows, facilitating the extraction and learning of richer image features. The head network consists of two parts: the neck and the detection head, responsible for feature fusion and object prediction, respectively. The neck is the intermediate layer connecting the backbone network and the detection head. Through feature fusion and resolution adjustment strategies, it significantly improves the detection head's prediction performance. The detection head integrates a classifier and regressor to predict the location and category of objects based on the output features of the neck.

[0063] YOLOv10 is an advanced real-time target detection algorithm that performs well in most scenarios. However, the turbid underwater environment severely restricts image quality, directly reducing the detection accuracy of YOLOv10 in aquaculture environments. In addition, the effective information carried by individual juvenile fish in the image is relatively limited, which increases the difficulty of detecting sick juvenile fish. To address the above problems, this embodiment proposes an improved YOLOv10 model to achieve accurate identification of sick juvenile fish. Its network structure is as follows: Figure 2 As shown in the figure. First, the SPD-Conv module is introduced to optimize the downsampling process and reduce the loss of detailed information of larval fish. Secondly, based on the backbone network of YOLOv10, by designing and integrating auxiliary branches, the feature extraction capability of diseased larval fish is enhanced, the diversity and comprehensiveness of feature information are achieved, and the detection effect of diseased larval fish in turbid water conditions is improved. At the same time, a new HFG-based fusion feature fusion method is proposed in the neck network. Through the guidance of high-level semantic information, the channel weights of the low-level feature maps are dynamically adjusted to achieve efficient reuse of low-level features of diseased larval fish and adaptive fusion of high- and low-level features, thereby improving the recall rate of small-target diseased larval fish. Finally, the SlideLoss loss function is introduced to focus on diseased larval fish samples, which further improves the network's detection performance for underwater diseased larval fish.

[0064] This embodiment provides a method for identifying diseased juvenile fish based on an improved YOLOv10 model, comprising the following steps:

[0065] S1. Collect images of fish in waters as images to be detected, and preprocess the images to be detected to obtain an image dataset;

[0066] Specifically, the preprocessing includes adjusting the size of the image to be detected, marking abnormal appearance and abnormal behavior in the image to be detected as abnormal labels, and marking normal individuals as normal labels; the final annotation information can be saved as a .txt format file;

[0067] S2. Build the YOLOv10-UDFishNet model. This involves replacing the strided convolutions in Stage 1 and Stage 2 of the YOLOv10 backbone network with SPD-Conv modules, adding auxiliary branches to the YOLOv10 backbone network, adding the HFG-based fusion feature fusion method to the YOLOv10 neck network, and replacing the YOLOv10 loss function with the SlideLoss loss function. This results in the YOLOv10-UDFishNet model.

[0068] In images captured in aquaculture environments, juvenile fish are relatively small and of a relatively fixed size. Directly applying a convolution operation with a stride of 2 as a downsampling method inevitably results in a significant loss of detailed fish features, making the detection of small underwater fish difficult. Therefore, this embodiment uses the SPD-Conv module to replace the existing strided convolutions in the backbone network Stages 1 and 2, mitigating the loss of juvenile fish features caused by downsampling.

[0069] In this embodiment, the SPD-Conv module includes a space-to-depth layer and a non-strided convolution layer, such as Figure 3 As shown, the space-to-depth layer is used to slice the given feature map X(S,S,C1) to obtain scale 2 The size is The sub-feature maps are spliced ​​in the channel dimension to obtain a new feature map

[0070] The slicing operation formula is shown in formula (1):

[0071]

[0072] Among them, f a,b Represents a sub-feature map, a∈(0,1,2…scale-1), b∈(0,1,2…scale-1), S represents the spatial size of the feature map, C1 represents the number of feature map channels, and scale represents the number of sub-feature maps that are split.

[0073] The non-strided convolution layer is used to add a non-strided convolution layer using C2 convolution kernels to obtain the final feature map Among them C2 <C1<scale 2 .

[0074] To further verify the effect of the optimized SPD-Conv module on small target fish detection, the feature maps before and after the backbone network Stage 2 are visualized. Figure 4As shown, Stages 1 and 2 extract richer details about small targets, and this fine-grained information is crucial for detecting individual small targets. However, Stage 2 extracts more abstract and advanced features, and excessive fine-grained information may no longer be necessary and may even introduce noise and redundancy, interfering with detection performance. Therefore, replacing only the strided convolutions in Stages 1 and 2 within the backbone network preserves this fine-grained information while avoiding interference in subsequent stages, thereby improving the network's detection of small fish targets.

[0075] Specifically, due to the limited feeding activity of young fish, excessive feed often remains in the water, causing turbidity in the aquaculture water and increasing the difficulty of detecting diseases. To meet this challenge, existing research mainly adopts a composite backbone network structure, which performs feature fusion after a parallel dual backbone network to enhance the diversity of model feature extraction. However, this approach will cause some features to be extracted repeatedly, increasing the redundancy of the overall features. In addition, the dual-backbone design will significantly increase the number of model parameters, which is not conducive to underwater real-time tasks. Therefore, this embodiment is based on the backbone network of YOLOv10, and compensates for the shortcomings of feature extraction of a single backbone network by designing and integrating auxiliary branches. The auxiliary branches are mainly stacked by the basic module ShuffleBlock of ShuffleNetv2, which ensures the improvement of model performance while not significantly increasing the computational burden. As Figure 2 As shown on the left, the input feature channel is first split, and the features are extracted through the backbone network and auxiliary branches respectively. After Stage 1 and Stage 2, they are spliced ​​along the channel dimension to achieve diversity and richness of feature extraction. To ensure that the feature map size is consistent when the channels are spliced, a downsampling operation is added before the first ShuffleBlock, thereby ensuring the smooth progress of the subsequent feature fusion process. ShuffleBlock contains two basic units, namely the feature extraction unit with a convolution stride of 1 (stride = 1) and the downsampling unit with a convolution stride of 2 (stride = 2). The unit structures are as follows: Figure 5 (a) Figure 5 However, the traditional ShuffleBlock structure needs to maintain the same number of input and output channels, which limits the improvement of model efficiency to a certain extent. Therefore, the structure of ShuffleBlock is optimized.

[0076] Therefore, in this embodiment, the auxiliary branch includes a first auxiliary branch and a second auxiliary branch, the first auxiliary branch is connected in parallel with Stage 1, and the second auxiliary branch is connected in parallel with Stage 2;

[0077] The first auxiliary branch includes an SPD-Conv module and an improved ShuffleBlock module connected in sequence, and the second auxiliary branch includes an improved ShuffleBlock module; the improved ShuffleBlock module includes an improved feature extraction unit and an improved downsampling unit;

[0078] like Figure 5 As shown in (c), the improved feature extraction unit replaces the 1×1 convolution after the depth-separable convolution of the feature extraction unit in the basic module ShuffleBlock of ShuffleNetv2 with partial convolution. Partial convolution performs standard convolution operations on some input channels to extract spatial features, while the remaining channels remain unchanged. This feature significantly reduces the amount of memory access. Its structure is as follows Figure 6 As shown, for the feature map X∈R h×w×c , take the continuous c p The channels are used as representatives of the entire feature map for convolution calculation. At this time, the FLOPs of the partial convolution PConv is only h×w×k 2 ×c p 2 , while the FLOPs of conventional convolution is h×w×k 2 ×c 2 , in order to enable the model to learn more refined feature representations, this embodiment sets the partial ratio of partial convolution to r = 1 / 2, so that more input channels participate in the calculation while still saving a large amount of computational cost. In addition, although some channels are not directly involved in the convolution calculation, the improved channel shuffling mechanism in ShuffleBlock effectively compensates for this potential information loss. By disrupting the order of channels, the information between different channels can fully interact and fuse, thereby providing comprehensive information from all channels for the partial convolution PConv operation in the next stage. This mechanism ensures that even if some channels are not directly involved in the calculation at a specific stage, their information can still be effectively used in subsequent processing, thereby reducing the risk of information loss and improving the overall performance of the model;

[0079] like Figure 5 As shown in (d), the improved downsampling unit is to replace the depth-separable convolution of the downsampling unit in the basic module ShuffleBlock of ShuffleNetv2 with the SPD-Conv module, achieving dual optimization of the backbone network and auxiliary branches in feature preservation;

[0080] In a complex underwater environment, the feature boundaries between sick juveniles and healthy fish become blurred due to their high similarity. At the same time, the effective feature information carried by the juveniles in the image is relatively limited, which directly leads to a decrease in the recall rate of small target sick juveniles. Therefore, making full use of the rich local semantic information in the low-level features is crucial for accurately identifying small target sick juveniles. However, traditional methods often simply splice high and low-level features along the channel and are unable to dynamically adjust the importance of different channels of the low-level feature map. Therefore, this embodiment designs an HFG-based fusion feature fusion method in the neck network, such as Figure 7 As shown in the figure, high-level semantic information is used to guide the learning of low-level feature map channel weights, and adaptive fusion of high-level and low-level features is performed, thereby efficiently reusing the low-level features of diseased juvenile fish and achieving more accurate individual status differentiation. Specifically, the HFG-based fusion feature fusion method includes the following steps:

[0081] Step a. Given a low-level feature map X low ∈R H×W×C and high-level feature maps X high ∈R H×W×C , the high-level feature map X high ∈R H×W×C The channels are evenly split into G groups. In this embodiment, G=4 to obtain G groups of feature maps. When the channels cannot be evenly divided, they are rounded down.

[0082] Step b. Perform 1×1 Conv operation on each group of feature maps after splitting, so that the number of channels of each group of feature maps is the same as that of the low-level feature map X low The number of channels is the same, and each set of feature maps after dimensionality increase is recorded as X g ;

[0083] Step c. Each set of feature maps X g Respectively with X low After element-by-element addition, the feature maps of each G group are accumulated to obtain the feature map X′ high , as shown in formula (2):

[0084]

[0085] Among them, g∈(1,2…G);

[0086] Step d. Feature map X′ high Perform global average pooling in the spatial dimension to compress the entire spatial information into a global feature, and obtain the C-dimensional global feature. The compression process is shown in formula (3):

[0087]

[0088] Among them, z c ∈R cRepresents the global feature, H represents the height of the feature map, W represents the width of the feature map, u c Represents the input matrix, i represents the row of the input matrix, j represents the column of the input matrix, and c represents the number of channels of the feature map;

[0089] Step e. Pass the global features through a fully connected layer consisting of ReLU activation function and Sigmoid activation function to automatically learn the feature importance of each channel and obtain the weighted output, as shown in formula (4):

[0090] s=σ(W2δ(W1z c )) (4)

[0091] Where s represents the weight output, δ is the ReLU activation function, σ is the Sigmoid activation function, W1(·) represents the fully connected layer operation of the ReLU activation function, and W2(·) represents the fully connected layer operation of the Sigmoid activation function.

[0092] Step f. Combine the weight output with the low-level feature map X low Perform element-by-element multiplication to achieve adaptive weighting of the low-level feature map, and splice it with the high-level feature map along the channel to obtain the final output, as shown in formula (5):

[0093]

[0094] Among them, Out represents the final output, represents element-by-element multiplication; [·,·] represents concatenation operation.

[0095] In real aquaculture, diseased juvenile fish samples are severely scarce. Through statistical analysis of the images of the self-built dataset, it was found that the ratio of the total number of fish to the number of diseased juvenile fish is about 30:1. This significant imbalance in quantity poses a severe challenge to the training of the network, resulting in insufficient attention to the scarce diseased juvenile fish samples. Therefore, the SlideLoss loss function is introduced as a new loss function to enhance the network's learning ability for diseased juvenile fish samples and further improve the network's detection performance for diseased juvenile fish. Slide is a sample balance function with adaptive parameters, and its calculation formula is shown in formula (7). Slide divides positive and negative samples according to the threshold μ. By adaptively learning the threshold parameter, a higher weight is set near μ to increase the relative loss of difficult-to-classify samples, thereby focusing more attention on the learning of difficult-to-classify samples. The Slide weighting function f(x) is added to the baseline model loss function. Specifically, the SlideLoss loss function is shown in formula (6):

[0096]

[0097] Among them, In represents the loss of the nth category, n∈(1,2…n), f(x) represents the addition of the Slide weighting function on the basis of the baseline model loss function, w n represents the weight factor, y n represents the nth actual value, σ represents the Sigmoid function, x n Represents the nth predicted value, and x represents the function input;

[0098] Add the Slide weighting function f(x) to the baseline model loss function as shown in formula (7):

[0099]

[0100] Among them, μ represents the average value of the intersection-over-union ratio between all predicted boxes and the true boxes;

[0101] S3. Input the image dataset obtained after preprocessing in step S1 into the YOLOv10-UDFishNet model obtained in step S2, train the YOLOv10-UDFishNet model using the image dataset, and obtain a trained YOLOv10-UDFishNet model;

[0102] Specifically, the improved YOLOv10-UDFishNet model is trained based on the image dataset until the training is completed when the set learning rounds are reached, and the trained YOLOv10-UDFishNet model is obtained;

[0103] S4. Input the fish image in the water area to be detected into the trained YOLOv10-UDFishNet model obtained in step S3 to obtain the fish detection results and the location information of the individual fish targets;

[0104] Specifically, the obtained fish test results include normal and abnormal.

[0105] Example 3. This example selects redfin pufferfish juveniles as experimental subjects and adopts the diseased juvenile fish identification method based on the improved YOLOv10 model in Example 2. The diseased juvenile fish data set is indispensable for studying diseased juvenile fish detection. In view of the current lack of publicly available diseased juvenile fish data sets, it is necessary to carry out the diseased juvenile fish data set collection work. To ensure the representativeness and authenticity of the data set, field research was conducted at the farm several times before data collection, and based on the research results, a data collection plan was planned. Afterwards, a farm environment simulation experiment was carried out in the laboratory by controlling relevant parameters. Finally, after arranging the data set obtained from the simulated environment and the data set obtained from the real breeding pond, a diseased juvenile fish data set was constructed.

[0106] In this example, 90 juvenile redfin pufferfish (Takifugu rubripes) were selected as experimental subjects. These fish were 7-10 cm long and weighed 15-18 grams. To simulate real-world aquaculture conditions, an appropriate amount of water was added to the tank and allowed to stand for 10 minutes. The experimental parameters were then measured: the water temperature was maintained at 13-18°C, the pH was 6.5-6.9, and the dissolved oxygen level was maintained at no less than 5 mg / L. The experimental setup consisted of an LED light source, mounted 1 meter above the inside of the tank to ensure adequate and even lighting. A camera was mounted next to the aquaculture tank, 20 cm above the water surface, and connected to a computer for automatic storage in MP4 video format. To prevent stress reactions in the new environment that could affect the accuracy of the experimental data, the juveniles were placed in separate containers for acclimatization for 2-3 days before subsequent experiments. Data collection lasted approximately 40 days, with data collected daily at two different time periods: 9-10 AM and 4-5 PM. During data collection, the camera remained on and connected to the computer in real time to ensure that all video data was stored on the computer's hard drive. Subsequently, the collected video files were edited and frame-processed to construct a dataset of diseased juvenile redfin pufferfish.

[0107] Pathological analysis revealed that the primary disease affecting redfin pufferfish was Vibrio harveyi infection, which resulted in abnormalities in both appearance and behavior. The affected fish exhibited obvious ulcers and round spots on the body, exposed muscles, and varying degrees of damage to the fins and tail. Abnormal behavior included swimming sideways and holding the fish perpendicular to the water surface.

[0108] After obtaining the image of sick juvenile fish, it is necessary to draw bounding boxes and assign labels to the targets in the image, but the process of selecting the target is extremely cumbersome, and the selection results will be subjective due to different understanding and judgment of the subject. In this embodiment, data annotation consists of two parts: semi-automatic annotation and manual fine-tuning to optimize the data annotation process and improve the annotation quality and efficiency. First, the YOLOv10 model is trained using 1,000 manually annotated images of sick juvenile fish, and the labels are uniformly "fish". Afterwards, the remaining 2,000 unlabeled images in this experiment are input into the trained YOLOv10 model for drawing the bounding boxes. At this point, it is necessary to manually retrieve the sick juvenile fish individuals in 3,000 images. The labels of all detected fish are reassigned, and the abnormal appearance and abnormal behavior are uniformly labeled as "diseased". The remaining normal individuals are changed to the "normal" label, and the final annotation information is saved as a .txt format file.

[0109] To achieve better network training results, the labeled diseased juvenile fish dataset was randomly divided in a 7:2:1 ratio, resulting in 2,100 training, 600 validation, and 300 test sets. To ensure more reliable evaluation results on the test set, the dataset was partitioned so that the test and training sets came from different video segments, effectively preventing data leakage.

[0110] YOLOv10 is an advanced real-time target detection algorithm that performs well in most scenarios. However, the turbid underwater environment severely restricts image quality, directly reducing the detection accuracy of YOLOv10 in aquaculture environments. In addition, the effective information carried by individual juvenile fish in the image is relatively limited, which increases the difficulty of detecting sick juvenile fish. To address the above problems, this embodiment proposes an improved YOLOv10 model to achieve accurate identification of sick juvenile fish. Its network structure is as follows: Figure 2 As shown in the figure. First, the SPD-Conv module is introduced to optimize the downsampling process and reduce the loss of detailed information of larval fish. Secondly, based on the backbone network of YOLOv10, by designing and integrating auxiliary branches, the feature extraction capability of diseased larval fish is enhanced, the diversity and comprehensiveness of feature information are achieved, and the detection effect of diseased larval fish in turbid water conditions is improved. At the same time, a new HFG-based fusion feature fusion method is proposed in the neck network. Through the guidance of high-level semantic information, the channel weights of the low-level feature maps are dynamically adjusted to achieve efficient reuse of low-level features of diseased larval fish and adaptive fusion of high- and low-level features, thereby improving the recall rate of small-target diseased larval fish. Finally, the SlideLoss loss function is introduced to focus on diseased larval fish samples, which further improves the network's detection performance for underwater diseased larval fish.

[0111] To ensure the accuracy and reliability of the experimental data, this study trained and tested the underwater diseased larvae detection network in the same environment. The experimental environment is shown in Table 1. Experimental hyperparameter settings included a batch size of 8, 300 epochs, and an initial learning rate of 0.0001.

[0112] Table 1 Experimental environment

[0113]

[0114] Recognition accuracy and high inference speed are crucial for real-time monitoring of diseased juvenile fish underwater. To comprehensively evaluate the performance of the proposed method, a confusion matrix was used as an evaluation tool. This matrix includes the following key metrics: True Positives (TP), representing samples correctly identified as exhibiting abnormal behavior; False Positives (FP), representing samples incorrectly identified as exhibiting abnormal behavior; and False Negatives (FN), representing samples mistakenly identified as background.

[0115] In terms of network detection accuracy, we selected precision (P), recall (R), and mean average precision (mAP@0.5) as evaluation metrics. mAP combines the accuracy performance across different categories. Generally speaking, a higher mAP value indicates better overall model performance. Furthermore, diseased larvae detection is a task with specific requirements. Therefore, the recall rate of diseased larvae should be the primary metric for evaluating model detection accuracy, compared to the total recall rate. In terms of detection speed, we selected model parameter count (Params) and inference speed (Inference speed) as evaluation metrics for model size and real-time performance, respectively.

[0116] The calculation of accuracy P is shown in formula (8):

[0117]

[0118] The calculation of recall rate R is shown in formula (9):

[0119]

[0120] The mean average precision mAP is shown in formula (10):

[0121]

[0122] Where N represents the number of categories, n∈(1,2,3,…N), AP represents the average precision of a category,

[0123] AP is shown in formula (11):

[0124]

[0125] To verify the effectiveness of the improved YOLOv10-UDFishNet model, we designed an ablation experiment. The experimental scheme is as follows: the auxiliary branch, SPD-Conv module, HFG-based fusion feature fusion method, and SlideLoss loss function are sequentially added to the baseline model. The results are YOLOv10-A with the auxiliary branch added, YOLOv10-AS with the auxiliary branch and SPD-Conv module added, and YOLOv10-ASH with the auxiliary branch, SPD-Conv module, and HFG-based fusion feature fusion method added. The experimental results are shown in Table 2.

[0126] Table 2 Ablation experiment results

[0127]

[0128] As shown in Table 2, the results demonstrate that these improvements significantly improve the performance of the baseline model. The final YOLOv10-UDFishNet model achieved 94.0% precision and 90.1% recall of diseased juvenile fish, respectively, representing significant improvements of 2.2% and 4.8% over the baseline model.

[0129] The complex underwater environment degrades image quality, significantly hindering model feature extraction and directly limiting the accuracy and recall of diseased larvae detection. By integrating auxiliary branches into the backbone network, the model not only enhances the comprehensiveness of diseased larvae feature extraction but also achieves information diversity complementarity, which is particularly critical for diseased larvae detection in low-quality images. This improves accuracy and recall by 1.3% and 2.3%, respectively, while reducing the number of parameters and accelerating detection. This is achieved by splitting the input and directing a portion of it to the auxiliary branches, effectively lightweighting the network and reducing the number of network parameters. Furthermore, the improved partial convolution (PConv) in the ShuffleBlock module performs convolution operations on only a subset of the input feature maps, while maintaining identity mapping for the remaining channels. This reduces computational redundancy and significantly improves computational efficiency. The introduction of the SPD-Conv module mitigates the loss of detailed information about small target fish during network deepening, enabling more accurate localization of individual underwater larvae and improving detection accuracy to a certain extent. The HFG-based fusion method then adaptively enhances the importance of low-level feature map channels, enabling efficient reuse of detailed features of diseased larvae at lower levels. This promotes the flow of information from shallow to deep layers, facilitating the differentiation of diseased and healthy individuals, significantly improving the recall rate of diseased larvae by 1.8%. The SlideLoss loss function, which weights samples, further enhances the model's detection performance for diseased larvae. Compared to the baseline model, the resulting network demonstrates significant advantages in both accuracy and speed in diseased larvae detection.

[0130] To further verify the effectiveness and advancement of the YOLOv10-UDFishNet model used in this example for underwater diseased juvenile fish detection, we selected leading models in this field for comparative experiments. These included mainstream object detection models RT-DETR, YOLOv8, YOLOv9, YOLOv10, and YOLOv8-Head-ECAM, as well as DDEYOLOv9 and DFYOLO, which are known for their excellent performance in underwater diseased juvenile fish detection. The experimental results are shown in Table 3.

[0131] Table 3 Comparison of detection performance of different models

[0132]

[0133] Table 3 shows that compared to other models, the YOLOv10-UDFishNet model proposed in this example achieves the best overall performance. Compared to RT-DETR, the YOLOv10-UDFishNet model significantly improves detection accuracy by 8.3% and recall by 13.7%. This is because in turbid underwater scenes, the RT-DETR model's global receptive field cannot accurately focus on individual diseased juveniles, resulting in severe under-extraction of diseased juvenile features and poor detection results. Compared to the mainstream YOLO series of algorithms, the YOLOv10-UDFishNet model achieves significant improvements in recall of diseased juveniles by 5.6%, 6.0%, and 4.8%, respectively. This is primarily attributed to the image quality degradation caused by the turbid underwater environment, which blurs the feature boundary between diseased and healthy juveniles. Furthermore, the traditional downsampling process inevitably results in spatial information loss, which weakens the detection ability of small targets and affects the model's performance in detecting diseased juveniles. YOLOv8-Head-ECAM effectively alleviates the challenge of small target positioning by adding a large detection head. Although it demonstrates higher accuracy than the YOLO algorithm, turbid environments inevitably interfere with the detection of diseased juvenile fish, resulting in a lower recall rate than the YOLOv10-UDFishNet model in this example.

[0134] Compared with the underwater sick juvenile fish detection models DDEYOLOv9 and DFYOLO, the YOLOv10-UDFishNet model also shows its advancedness. Although DDEYOLOv9 has improved the feature extraction capability of sick juvenile fish to a certain extent by introducing the DRNELAN4 module, its enlarged receptive field ignores the detailed features of small target fish, limiting the improvement of the recall rate; while DFYOLO enhances the model's focus on key image areas and the extraction of fish surface features through the combination of convolution kernel groups and CBAM modules. However, the redfin pufferfish juveniles are small in size, and their sick juvenile individuals are easily confused with background information, resulting in insufficient improvement in the recall rate of sick juveniles. Figure 8The heat map intuitively shows the detection effect of each model on small target diseased juvenile fish. In the figure, (a) is the heat map of the detection effect of the original image, (b) is the heat map of the detection effect of RT-DETR, (c) is the heat map of the detection effect of YOLOv8, (d) is the heat map of the detection effect of YOLOv9, (e) is the heat map of the detection effect of YOLOv10, (f) is the heat map of the detection effect of YOLOv8-Head-ECAM, (g) is the heat map of the detection effect of DDEYOLOv9, (h) is the heat map of the detection effect of DFYOLO, and (i) is the heat map of the detection effect of YOLOv10-UDFishNet. The color depth in the heat map directly reflects the degree of attention paid to the area by the model. Through the color contrast of the heat map, it can be seen that the YOLOv10-UDFishNet model of this embodiment has a high degree of focus and accurate detection ability for small target diseased juvenile fish in underwater turbid environments.

[0135] To evaluate the applicability of the proposed YOLOv10-UDFishNet model and its robustness in small-target fish detection, this example further extends the method to the field of target detection, focusing on the challenging task of small-target fish detection. Specifically, 2,420 images were randomly selected from the fish dataset of the China Agricultural Artificial Intelligence Innovation and Entrepreneurship Competition. The dataset was divided into a ratio of 7:2:1, and target detection experiments were conducted simultaneously with RT-DETR, YOLOv8-Head-ECAM, YOLOv8, YOLOv9, and the baseline model YOLOv10 to compare model performance. The experimental results are shown in Table 4:

[0136] Table 4 Comparison of model detection performance on public datasets

[0137]

[0138] As shown in Table 4, the data shows that the YOLOv10-UDFishNet model of this embodiment demonstrates a significant advantage in recall, exceeding RT-DETR and the baseline model by 8.6% and 1.3% respectively. Furthermore, compared to the YOLOv8-Head-ECAM, which integrates the efficient attention mechanism ECAM, the YOLOv10-UDFishNet model of this embodiment achieves the same level of accuracy in individual fish detection, with an accuracy rate of up to 99.1%. The YOLOv10-UDFishNet model of this embodiment maintains excellent performance in both recall and accuracy for fish detection, demonstrating its robustness in the task of detecting small, diseased juvenile fish underwater.

[0139] In the task of detecting sick juvenile redfin pufferfish in complex underwater environments, this example proposes a new sick juvenile fish identification method based on an improved YOLOv10 model, aiming to accurately and efficiently detect sick juvenile redfin pufferfish in turbid underwater environments. By designing auxiliary branches based on the backbone network, the problem of interference from the turbid water background on feature extraction is resolved, significantly enhancing the model's adaptability in complex environments. To more accurately distinguish sick juveniles from normal individuals, the SPD-Conv module is used to replace the traditional strided convolution to reduce the loss of small target information caused by the downsampling process. In the neck network, an adaptive fusion method guided by high-level features, HFG-based fusion, is proposed to achieve efficient reuse of low-level sick juvenile features and improve the network's ability to distinguish small targets of sick juveniles. Finally, the SlideLoss loss function is introduced to adaptively adjust sample weights, increase attention to difficult samples, and further improve the network's detection performance for sick juveniles. Experimental results show that compared with the baseline model, the improved YOLOv10-UDFishNet model improves the accuracy, total recall, diseased fry recall and average precision by 2.2%, 2.9%, 4.8% and 1.5% respectively.

[0140] The embodiments of the present invention are presented for purposes of illustration and description and are not intended to be exhaustive or to limit the invention to the disclosed forms. Many modifications and variations will be apparent to those skilled in the art. The embodiments are chosen and described in order to better illustrate the principles of the invention and its practical application and to enable those skilled in the art to understand the invention and design various embodiments with various modifications as suited for specific applications.

Claims

1. A method for identifying sick juvenile fish based on an improved YOLOv10 model, characterized in that: The steps include: S1. Collecting fish images in waters as images to be detected, and preprocessing the images to be detected to obtain an image dataset; S2. Build the YOLOv10-UDFishNet model. This involves replacing the strided convolutions in Stage 1 and Stage 2 of the YOLOv10 backbone network with SPD-Conv modules, adding auxiliary branches to the YOLOv10 backbone network, adding the HFG-based fusion feature fusion method to the YOLOv10 neck network, and replacing the YOLOv10 loss function with the SlideLoss loss function. This results in the YOLOv10-UDFishNet model. S3. Inputting the image dataset obtained after preprocessing in step S1 into the YOLOv10-UDFishNet model obtained in step S2, training the YOLOv10-UDFishNet model using the image dataset to obtain a trained YOLOv10-UDFishNet model; S4. The fish image in the waters to be detected is input into the trained YOLOv10-UDFishNet model obtained in step S3 to obtain the fish detection results and the location information of individual fish targets; In step S2, the auxiliary branch includes a first auxiliary branch and a second auxiliary branch, the first auxiliary branch is connected in parallel with Stage 1, and the second auxiliary branch is connected in parallel with Stage 2; The first auxiliary branch includes an SPD-Conv module and an improved ShuffleBlock module connected in sequence, and the second auxiliary branch includes an improved ShuffleBlock module; the improved ShuffleBlock module includes an improved feature extraction unit and an improved downsampling unit; The improved feature extraction unit replaces the 1×1 convolution after the depth-separable convolution of the feature extraction unit in the basic module ShuffleBlock of ShuffleNetv2 with a partial convolution. The partial convolution performs a standard convolution operation on some input channels to extract spatial features, while the remaining channels remain unchanged. The partial ratio of the partial convolution is r=1 / 2; The improved downsampling unit is to replace the depth-separable convolution of the downsampling unit in the basic module ShuffleBlock of ShuffleNetv2 with the SPD-Conv module.

2. The method for identifying sick juvenile fish based on the improved YOLOv10 model according to claim 1, characterized in that: In the step S1, the preprocessing includes adjusting the size of the image to be detected, marking abnormal appearance and abnormal behavior in the image to be detected as abnormal labels, and marking normal individuals as normal labels.

3. The method for identifying sick juvenile fish based on the improved YOLOv10 model according to claim 1, characterized in that: In step S2, the SPD-Conv module includes a space-to-depth layer and a non-step convolution layer. The space-to-depth layer is used to slice the given feature map X (S, S, C1) to obtain scale 2 The size is ( , ,C1) of the sub-feature map, the sub-feature map is spliced ​​in the channel dimension to obtain a new feature map X'( , ,C1×scale 2 ); The slicing operation formula is shown in formula (1): （1） in, represents the sub-feature map, , , S represents the size of the feature map space, C1 represents the number of channels of the sub-feature map, Represents the number of sub-feature maps that are split out. The non-strided convolution layer is used to add a non-strided convolution layer using C2 convolution kernels to obtain the final feature map X'' ( , , C2), where C2 <C1<scale 2 .

4. The method for identifying sick juvenile fish based on the improved YOLOv10 model according to claim 1, characterized in that: In step S2, the HFG-based fusion feature fusion method includes the following steps: Step a. Given a low-level feature map X low ∈R H×W×C and high-level feature map X high ∈R H×W×C , the high-level feature map X high ∈R H×W×C The channels are evenly split into G groups to obtain G group feature maps. If they cannot be evenly divided, they are rounded down. Step b. Perform 1×1 Conv operation on each group of split feature maps to make the number of channels of each group of feature maps the same as the low-level feature map X low The number of channels is the same, and each set of feature maps after dimensionality increase is recorded as X g ; Step c. Each set of feature maps X g Respectively with X low After adding each element, the feature maps of each G group are accumulated to obtain the feature map , as shown in formula (2): （2） in, ; Step d. The feature map Perform global average pooling in the spatial dimension to compress the entire spatial information into a global feature, and obtain the C-dimensional global feature. The compression process is shown in formula (3): （3） in, Represents the global feature, H represents the height of the feature map, W represents the width of the feature map, Represents the input matrix, i represents the row of the input matrix, j represents the column of the input matrix, and c represents the number of channels of the feature map; Step e. Pass the global feature through a fully connected layer consisting of a ReLU activation function and a Sigmoid activation function to obtain the weighted output, as shown in formula (4): （4） Among them, s represents the weight output, δ is the ReLU activation function, σ is the Sigmoid activation function, W1 (·) represents the fully connected layer operation of the ReLU activation function, and W2 (·) represents the fully connected layer operation of the Sigmoid activation function. Step f. Combine the weight output with the low-level feature map X low Perform element-by-element multiplication and concatenate it with the high-level feature map along the channel to obtain the final output, as shown in formula (5): （5） Among them, Out represents the final output, represents element-by-element multiplication; [·,·] represents a concatenation operation.

5. The method for identifying sick juvenile fish based on the improved YOLOv10 model according to claim 1, characterized in that: In step S2, the SlideLoss loss function is shown in formula (6): （6） in, represents the loss of the nth category, , Indicates adding the Slide weighting function based on the baseline model loss function. represents the weight factor, represents the nth actual value, represents the Sigmoid function, represents the nth predicted value, Represents function input; Add Slide weighting function based on the baseline model loss function As shown in formula (7): （7） Where µ represents the average of the intersection-over-union (IoU) between all predicted boxes and the true boxes.

6. The method for identifying sick juvenile fish based on the improved YOLOv10 model according to any one of claims 1 to 5, characterized in that: In step S3, the improved YOLOv10-UDFishNet model is trained based on the image dataset until the training is completed when a set learning round is reached, thereby obtaining a trained YOLOv10-UDFishNet model.

7. The method for identifying sick juvenile fish based on the improved YOLOv10 model according to claim 1, characterized in that: The fish detection results obtained in step S4 include normal and abnormal.

Images