A machine vision-based image processing method and system for jackfish

Abstract

The application relates to an image processing method and system for Trachurus japonicus based on machine vision, and belongs to the technical field of aquatic disease detection. The method comprises the following steps: constructing an image dataset of Trachurus japonicus; constructing an image processing model for Trachurus japonicus based on an improved YOLOv11, wherein the model comprises a backbone network, a hierarchical pyramid spatial attention module, a neck network and a detection head; inputting an original Trachurus japonicus image into the model to obtain a preliminary detection result; performing HSV color space conversion and disease hue filtering to generate a corresponding binary mask; based on the preliminary detection result, verifying the hue similarity scores of the detection boxes through detection box hue consistency, and through confidence fusion and filtering, removing the detection boxes below a threshold value to obtain a calibrated detection result. The application can realize rapid, accurate and automatic detection of ulcer disease of Trachurus japonicus.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of aquaculture disease detection, and particularly relates to an image processing method and system for Trachinotus ovatus based on machine vision. BACKGROUND

[0002] Trachinotus ovatus, also known as golden pompano, is an important high-economic-value marine fish in the coastal areas of southern China. In recent years, with the expansion of intensive aquaculture, disease problems have become increasingly prominent, among which the ulcer disease caused by Vibrio and other pathogens is particularly common. The typical symptoms of the disease are round or nearly round ulcer lesions on the body surface, about 2-5 mm in diameter, with obvious congestion or bleeding at the edge. In addition, there are eye diseases, fin damage, and gill diseases, which can cause mass mortality and significant economic losses.

[0003] Currently, the detection of the above-mentioned Trachinotus ovatus diseases mainly relies on two methods: one is manual visual observation, which is carried out by experienced aquaculture personnel who regularly patrol the fish population and find suspicious lesions to make preliminary judgments; the other is laboratory molecular detection technology, such as polymerase chain reaction (PCR), which confirms the presence of pathogens by extracting and amplifying DNA from fish tissue samples. However, both methods have obvious limitations: the manual detection method is greatly affected by subjective factors, has low efficiency, and has a high rate of missed detection of early and small lesions (less than 5 mm in diameter); although the PCR detection has high accuracy, the process is complicated and time-consuming, and cannot realize real-time and online disease early warning and prevention and control, which is difficult to meet the management needs of modern aquaculture intelligence and high density. In addition, existing research attempts to introduce image processing and machine learning methods for fish disease identification, but still faces many challenges in practical application: the underwater shooting environment is complex, with uneven lighting, color cast, and large changes in turbidity, resulting in unstable image quality; traditional visual algorithms rely heavily on color and texture features, and have limited recognition accuracy for small target lesions in complex backgrounds, especially for diseases with similar colors and backgrounds (such as ulcers) with high false detection rates. Therefore, developing an intelligent detection method that can adapt to underwater environments, integrate multi-modal information, and have high accuracy and strong robustness has become an urgent need in the field of disease prevention and control in modern aquaculture. SUMMARY

[0004] The present application provides an image processing method and system for Trachinotus ovatus based on machine vision to solve the above problems.

[0005] To achieve the above purpose, the following technical solutions are adopted:

[0006] The present application provides an image processing method for Trachinotus ovatus based on machine vision, comprising the following steps:

[0007] Step one, obtain the image of Trachinotus ovatus, and construct the image dataset of Trachinotus ovatus;

[0008] Step two, construct the Trachinotus ovatus image processing model based on the improved YOLOv11, the model includes a backbone network, a hierarchical pyramid spatial attention module, a neck network and a detection head; the traditional backbone network such as VGG and Darknet has two defects: large amount of calculation, difficult to adapt to real-time detection demand in breeding field; imbalance in extraction of detail features and semantic features of underwater images, leading to features of small targets being submerged in background; the backbone network of the application adopts CSPDarknet53 network as the basic architecture, and inputs the original Trachinotus ovatus image into the backbone network to obtain three levels of feature maps; the CSPDarknet53 network can retain more shallow detail features (such as hyperemic edges of ulcers and tear textures of fin damage) through the cross-stage partial connection structure while reducing the amount of calculation, and adapt to real-time detection in breeding field; the three levels of feature maps provide full-dimensional feature basis for multi-scale enhancement of the subsequent hierarchical pyramid spatial attention module, avoiding missed detection of small lesions caused by single scale features; the three levels of feature maps obtain channel enhanced features through the hierarchical pyramid spatial attention module; the neck network adopts a dynamic weighted feature pyramid network DWFPN, and the channel enhanced features obtain second fusion features through the neck network; the second fusion features obtain preliminary detection results through the detection head; in the process of training the Trachinotus ovatus image processing model based on the improved YOLOv11, the loss function includes boundary box loss, target confidence loss and classification loss;

[0009] Step three, HSV color space conversion and disease hue filtering are performed to generate corresponding binary masks; based on the preliminary detection results, the hue similarity score of each detection box is obtained through detection box hue consistency verification; through confidence fusion and filtering, detection boxes below the threshold are removed to obtain calibrated detection results; color modal information is used to correct the detection results of visual modal, which significantly reduces the false detection caused by color deviation of underwater images.

[0010] Further, the hierarchical pyramid spatial attention module includes a hierarchical feature aggregation layer, a multi-scale spatial attention layer and a channel reweighting layer; the three levels of feature maps are spliced in the channel dimension through the hierarchical feature aggregation layer to obtain an initial feature map, three levels of features are spliced in the channel dimension to fuse detail texture, local semantics and global class information, so that the model has small lesion positioning accuracy and large lesion classification ability at the same time; the initial feature map obtains spatial enhanced features through the multi-scale spatial attention layer; and the spatial enhanced features obtain channel enhanced features through the channel reweighting layer.

[0011] Further, in order to enhance the lesion area response and suppress the background noise, the application adopts a multi-scale spatial attention layer to process the initial feature map; the multi-scale spatial attention layer includes a multi-scale feature extraction unit, a feature fusion unit and a spatial feature enhancement unit; the initial feature map is input into the multi-scale feature extraction unit, and three parallel hollow convolution branches are used to obtain three hollow convolution features; the three hollow convolution features are input into the feature fusion unit, and element-by-element addition is performed to obtain a first fusion feature; the first fusion feature is subjected to convolution operation and Sigmoid activation function to generate a spatial weight map; the convolution operation is used to compress the channel; the initial feature map and the spatial weight map are input into the spatial feature enhancement unit, and the initial feature map and the spatial weight map are multiplied element by element to obtain a spatial enhancement feature.

[0012] Further, the channel re-weighting layer can enhance the lesion-related feature channel (such as ulcer texture / rotten gill color) and suppress irrelevant / redundant channels; the spatial enhancement feature is input into the channel re-weighting layer, and is processed by sequentially performing global average pooling, a first full connection layer, a ReLU activation function, a second full connection layer and a Sigmoid activation function to obtain a channel weight; the global average pooling is used to compress the spatial information to generate a channel description vector; the first full connection layer is used to reduce the dimension of the channel number; and the second full connection layer is used to restore the feature map to the original channel number. The spatial enhancement feature and the channel weight are multiplied element by element by the channel feature enhancement unit to obtain a channel enhancement feature.

[0013] Further, the original Trachurus japonicus image is converted from the RGB color space to the HSV color space; a disease-specific threshold filter is set according to the color characteristics of a specific disease in the H hue channel; the specific setting of the disease-specific threshold filter is as follows: for the ulcer disease, the H value range is [0°, 15°], and the mask color is red; for the rotten gill disease, the H value range is [90°, 120°], and the mask color is green; for the fish fin injury, the H value range is [30°, 50°], and the mask color is blue; and for the fish fin injury, the H value range is [200°, 220°], and the mask color is yellow; the H value range is selected according to the disease category predicted by the detection frame, and a binary mask is generated.

[0014] Further, each candidate detection frame in the preliminary detection result output by the model is mapped back to the HSV space of the original image, and the H value histogram of the pixels in the detection frame region is calculated; the proportion of the pixels falling within the range of the corresponding disease filter in the region is counted to obtain the hue similarity score of each detection frame.

[0015] Further, the confidence score of each detection box output by the model is multiplied by the hue similarity score of each detection box to obtain a final confidence score of each detection box; a fusion score threshold is set; when the final confidence score is lower than the fusion score threshold, the detection box is removed, and a calibrated detection result is finally obtained.

[0016] The application also provides an image processing system for Trachinotus ovatus based on machine vision, which executes the image processing method for Trachinotus ovatus based on machine vision.

[0017] The data acquisition module is configured to acquire the image of Trachinotus ovatus and construct an image dataset of Trachinotus ovatus.

[0018] The model construction module is configured to construct an image processing model for Trachinotus ovatus based on improved YOLOv11, wherein the model comprises a backbone network, a hierarchical pyramid spatial attention module, a neck network and a detection head; and the original image of Trachinotus ovatus is input into the image processing model for Trachinotus ovatus based on improved YOLOv11 to obtain a preliminary detection result.

[0019] The detection box calibration module is configured to perform HSV color space conversion and disease hue filtering on the original image of Trachinotus ovatus to generate a corresponding binary mask; based on the preliminary detection result, the hue similarity score of each detection box is obtained through detection box hue consistency verification; through confidence fusion and filtering, the detection boxes lower than the threshold are removed to obtain a calibrated detection result.

[0020] The application has the following advantages:

[0021] The application significantly enhances the feature extraction capability for small-scale lesions by introducing the improved YOLOv11 model structure and the hierarchical pyramid spatial attention mechanism, and improves the perception performance of the model in complex underwater environments; by fusing the disease-specific hue information in the HSV color space, a visual-color dual-modal verification mechanism is constructed, which effectively overcomes the false detection problem caused by underwater color deviation and light changes of a single visual model, and has strong correction ability for diseases such as gill rot disease with significant color features. The application realizes rapid, accurate and automatic detection of the ulcerative disease of Trachinotus ovatus, and is suitable for real-time monitoring and early warning in aquaculture sites, and provides reliable technical support for intelligent prevention and control of aquatic diseases. BRIEF DESCRIPTION OF DRAWINGS

[0022] The accompanying drawings are included to provide a further understanding of the application, and constitute a part of the specification, which together with the embodiments of the application, serve to explain the application, and do not constitute a limitation of the application.

[0023] Figure 1 The application provides a step flowchart of the method;

[0024] The application provides a step flowchart of the method;Figure 2 Precision-recall curves for the YOLOv11 base model;

[0025] Figure 3 This is a precision-recall curve of the method of the present invention. Detailed Implementation

[0026] 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 embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Example 1

[0028] In this embodiment, as Figure 1 As shown, this invention provides a machine vision-based image processing method for oval pomfret, the specific steps of which include:

[0029] S1. Obtain images of oval pomfret and construct an oval pomfret image dataset.

[0030] S2. Construct an image processing model for oval pomfret based on an improved YOLOv11, the model including a backbone network, a hierarchical pyramid spatial attention (HPSA) module, a neck network, and a detection head;

[0031] Specifically, the backbone network uses the CSPDarknet53 network as its basic architecture; the original oval pomfret image is input into the backbone network to obtain three levels of feature maps;

[0032] Specifically, the hierarchical pyramid spatial attention HPSA module includes a hierarchical feature aggregation layer, a multi-scale spatial attention layer, and a channel reweighting layer; in order to enhance the response of the lesion area and suppress background noise, the multi-scale spatial attention layer includes a multi-scale feature extraction unit (including a first branch, a second branch, and a third branch, which respectively adopt a hollow convolution with a hole rate r=1, a receptive field of 3x3, a hole rate r=3, a receptive field of 7x7, and a hole rate r=5, a receptive field of 11x11), a feature fusion unit, and a spatial feature enhancement unit; the channel reweighting layer can enhance the lesion-related feature channel (such as ulcer texture / rotten gill color) and suppress irrelevant / redundant channels. The channel reweighting layer includes a global average pooling, a first full connection layer, a ReLU activation function, a second full connection layer, a Sigmoid activation function, and a channel feature enhancement unit; the global average pooling is used to compress spatial information and generate a channel description vector; the first full connection layer is used to reduce the dimension of the channel number; and the second full connection layer is used to restore the feature map to the original channel number.

[0033] Specifically, the three-level feature maps are spliced in the channel dimension through the hierarchical feature aggregation layer to obtain an initial feature map; the initial feature map is input into the multi-scale feature extraction unit to obtain three hollow convolution features through three parallel hollow convolution branches; the three hollow convolution features are input into the feature fusion unit to obtain a first fusion feature through element-by-element addition; the first fusion feature is processed through a convolution operation and a Sigmoid activation function to generate a spatial weight map; the convolution operation adopts a 1x1 convolution kernel to compress the channel; and the initial feature map and the spatial weight map are input into the spatial feature enhancement unit to multiply the initial feature map and the spatial weight map element by element to obtain a spatial enhancement feature.

[0034] The spatial enhancement feature is input into the channel reweighting layer and processed through global average pooling, a first full connection layer, a ReLU activation function, a second full connection layer, and a Sigmoid activation function in sequence to obtain a channel weight; the spatial enhancement feature and the channel weight are multiplied element by element through the channel feature enhancement unit to obtain a channel enhancement feature.

[0035] Specifically, the neck network adopts a dynamic weighted feature pyramid network DWFPN; the channel enhancement feature is input into the neck network to obtain a second fusion feature; and the second fusion feature is input into the detection head to obtain a preliminary detection result.

[0036] Specifically, in the process of training the image processing model of the oval pompano based on the improved YOLOv11, the loss function includes a boundary box loss, a target confidence loss and a classification loss.

[0037] S3. The RGB color space is greatly affected by underwater light (such as strong light reflection and weak light darkening), and water color deviation (such as blue water leading to disease color deviation), and the traditional model relies on RGB features, which is easy to misjudge the normal area with color distortion as a disease (such as misjudging the normal fish scale in the blue water body as a diseased gill); the present application performs HSV color space conversion and disease hue filtering to generate a corresponding binary mask; the hue similarity score of each detection box is obtained through detection box hue consistency verification; through confidence fusion and filtering, the detection boxes below the threshold are removed to obtain the calibrated detection result; the color modal information is used to correct the detection result of the visual modal, which significantly reduces the false detection caused by the color deviation of the underwater image.

[0038] Specifically, S31. HSV color space conversion and disease hue filtering: convert the original oval pompano image from the RGB color space to the HSV color space; set a disease-specific threshold filter according to the color characteristics of a specific disease in the H hue channel; the specific setting of the disease-specific threshold filter is: for ulcer disease, the H value range is [0°, 15°], and the mask color is red; for diseased gill disease, the H value range is [90°, 120°], and the mask color is green; for fish fin injury, the H value range is [30°, 50°], and the mask color is blue; for fish fin injury, the H value range is [200°, 220°], and the mask color is yellow; select the H value range according to the disease category predicted by the detection box, and generate a binary mask .

[0039] Specifically, S32. Detection box hue consistency verification: map each candidate detection box in the preliminary detection result output by the model back to the HSV space of the original image, and calculate the H value histogram of the pixels in the detection box region; count the proportion of pixels falling within the range of the corresponding disease filter in the region to obtain the hue similarity score of each detection box, which is expressed as follows:

[0040] ,

[0041] Among them, Hue similarity score is represented by H; represents the number of pixels in the detection box that meet the H range; represents a binary mask, represents a disease area, represents a background area; represents the total number of pixels in the detection box.

[0042] Specifically, S33. Confidence fusion and filtering: multiply the confidence score of each detection box output by the model with the hue similarity score of each detection box in step S32 to obtain the final confidence score of each detection box, which is expressed by the following formula:

[0043] ,

[0044] wherein, represents the confidence score of the detection box, represents the final confidence score; the fusion score threshold is set to 0.7; when the final confidence score is lower than the fusion score threshold, the detection box is removed, and the calibrated detection result is finally obtained.

[0045] Example 2

[0046] In this embodiment, an image dataset of Trachurus japonicus is constructed, the dataset has a total of 8475 images, and the test set contains a total of 1695 Trachurus japonicus images; Yolo11 basic model, Yolo11+HPSA model, Yolo11+HPSA+HSV model experiments are compared on the Trachurus japonicus image dataset constructed in the application. The experimental results are shown in Table 1.

[0047] Table 1: Experimental comparison of the method of the application

[0048]

[0049] The average precision (mAP) is a core index for measuring the detection accuracy of a model in target detection, and the higher the value, the more accurate the model. The F1 score is the harmonic mean of precision and recall, which comprehensively reflects the classification performance of the model. As can be seen from the table, after adding HPSA and HSV color space enhancement in turn, the model performance is significantly improved, and the mAP and F1 score are increased by 21% and 19% respectively, indicating that the method of the application has obvious advantages in improving detection accuracy and classification stability.

[0050] As shown in Figure 2 , Figure 3 From the figure, it can be seen that the PR curve of the method of the application is located in the upper right of the YOLOv11 curve, and at the same recall rate point, the precision of the method of the application is higher; at the same precision point, the recall rate of the method of the application is higher; proving that the method of the application can better balance the accuracy and recall rate. By introducing HPSA and HSV color space enhancement, the method of the application can significantly improve the performance of the YOLOv11 model in the target detection task.

[0051] Example 3

[0052] The embodiment provides an image processing system for Trachurus japonicus based on machine vision, executes an image processing method for Trachurus japonicus based on machine vision, and comprises the following steps:

[0053] A data acquisition module is configured to acquire the image of the Trachurus japonicus and construct an image dataset of the Trachurus japonicus.

[0054] A model construction module is configured to construct an image processing model for the Trachurus japonicus based on an improved YOLOv11, wherein the model comprises a backbone network, a hierarchical pyramid spatial attention module, a neck network and a detection head; and the original image of the Trachurus japonicus is input into the image processing model for the Trachurus japonicus based on the improved YOLOv11 to obtain a preliminary detection result.

[0055] A detection frame calibration module is configured to perform HSV color space conversion and disease hue filtering on the original image of the Trachurus japonicus to generate a corresponding binary mask; based on the preliminary detection result, the hue similarity score of each detection frame is obtained through detection frame hue consistency verification; through confidence fusion and filtering, the detection frames below a threshold value are removed to obtain a calibrated detection result.

[0056] Finally, it should be noted that: the above only describes the preferred embodiments of the present application, and is not used to limit the present application, although the present application has been described in detail with reference to the foregoing embodiments, for those skilled in the art, the technical solutions recorded in the foregoing embodiments can still be modified, or some technical features can be replaced. Any modification, equivalent replacement, improvement, etc. within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A machine vision-based processing method for images of Trachurus japonicus, characterized by, The method comprises the following steps: An image of Trachurus japonicus is acquired, and a Trachurus japonicus image dataset is constructed; An improved YOLOv11-based Trachurus japonicus image processing model is constructed, the model comprising a backbone network, a hierarchical pyramid spatial attention module, a neck network, and a detection head; an original Trachurus japonicus image is input into the backbone network to obtain three levels of feature maps; the three levels of feature maps pass through the hierarchical pyramid spatial attention module to obtain channel enhanced features; The channel enhanced features pass through the neck network to obtain second fusion features; the second fusion features pass through the detection head to obtain preliminary detection results; the hierarchical pyramid spatial attention module comprises a hierarchical feature aggregation layer, a multi-scale spatial attention layer, and a channel reweighting layer; the three levels of feature maps pass through the hierarchical feature aggregation layer to be spliced in the channel dimension to obtain initial feature maps; The initial feature maps pass through the multi-scale spatial attention layer to obtain spatial enhanced features; Specifically, The multi-scale spatial attention layer comprises a multi-scale feature extraction unit, a feature fusion unit, and a spatial feature enhancement unit; The initial feature maps are input into the multi-scale feature extraction unit, pass through three parallel hollow convolution branches, and three hollow convolution features are obtained; The three hollow convolution features are input into the feature fusion unit, are element-wise added to obtain first fusion features, the first fusion features pass through a convolution operation and a Sigmoid activation function to generate a spatial weight map; the initial feature maps and the spatial weight map are input into the spatial feature enhancement unit, the initial feature maps and the spatial weight map are element-wise multiplied to obtain spatial enhanced features; The spatial enhanced features pass through the channel reweighting layer to obtain channel enhanced features; HSV color space conversion and disease hue filtering are performed to generate a corresponding binary mask; Based on the preliminary detection results, the hue similarity scores of the detection boxes are obtained through detection box hue consistency verification; Specifically, Each candidate detection box in the preliminary detection results output by the model is mapped back to the HSV space of the original image, the H value histogram of the pixels in the detection box region is calculated, the proportion of the pixels falling within the corresponding disease filter range in the region is counted, and the hue similarity scores of the detection boxes are obtained; Through confidence fusion and filtering, detection boxes below a threshold are removed, and calibrated detection results are obtained; specifically, The confidence scores of the detection boxes output by the model are multiplied by the hue similarity scores of the detection boxes to obtain final confidence scores of the detection boxes; A fusion score threshold is set; When the final confidence score is below the fusion score threshold, the detection box is removed, and the calibrated detection results are finally obtained.

2. The method according to claim 1, wherein the method is characterized by, The spatial enhanced features pass through the channel reweighting layer to obtain channel enhanced features: The spatial enhanced features are input into the channel reweighting layer, are processed through successive global average pooling, a first fully connected layer, a ReLU activation function, a second fully connected layer, and a Sigmoid activation function, and channel weights are obtained; the spatial enhanced features and the channel weights are element-wise multiplied through a channel feature enhancement unit to obtain channel enhanced features.

3. The method according to claim 2, wherein the image processing method is based on machine vision. HSV color space conversion and disease hue filtering are performed to generate a corresponding binary mask: The original Trachurus japonicus image is converted from an RGB color space to an HSV color space; a disease-specific threshold filter is set according to the color characteristics of the disease in the H hue channel; the specific setting of the disease-specific threshold filter is as follows: for ulcer disease, the H value range is [0°, 15°], and the mask color is red; for gill rot disease, the H value range is [90°, 120°], and the mask color is green; for fish fin injury, the H value range is [30°, 50°], and the mask color is blue; for fish fin injury, the H value range is [200°, 220°], and the mask color is yellow; the H value range is selected according to the disease category predicted by the detection box, and a binary mask is generated.

4. The method according to claim 3, wherein the method further comprises the steps of: determining the number of the Trachurus japonicus in the image; and determining the size of the Trachurus japonicus in the image. The backbone network adopts a CSPDarknet53 network as a basic architecture; the neck network adopts a dynamic weighted feature pyramid network DWFPN.

5. The machine vision-based Trachurus japonicus image processing method according to claim 4, characterized in that, In the process of training the Trachurus japonicus image processing model based on the improved YOLOv11, the loss function includes a boundary box loss, a target confidence loss and a classification loss.

6. A machine vision-based Trachurus japonicus image processing system, which performs a machine vision-based Trachurus japonicus image processing method according to claim 1, characterized in that, It comprises: a data acquisition module for acquiring Trachurus japonicus images and constructing a Trachurus japonicus image dataset; a model construction module for constructing a Trachurus japonicus image processing model based on the improved YOLOv11, the model comprising a backbone network, a hierarchical pyramid spatial attention module, a neck network and a detection head; the original Trachurus japonicus image is input into the Trachurus japonicus image processing model based on the improved YOLOv11 to obtain a preliminary detection result; a detection box calibration module for performing HSV color space conversion and disease hue filtering on the original Trachurus japonicus image to generate a corresponding binary mask; based on the preliminary detection result, the hue similarity score of each detection box is obtained through detection box hue consistency verification; through confidence fusion and filtering, the detection boxes below the threshold are eliminated, and the calibrated detection result is obtained.

Images