Fish disease detection method, device, equipment, medium and computer program product

Abstract

The invention provides a fish disease detection method, device and equipment, a medium and a computer program product, and the method comprises the steps: marking a historical underwater image of a to-be-detected fish, and obtaining a fish disease detection data set; training a target detection model based on the fish disease detection data set; the trunk structure of the target detection model is a convolutional neural network architecture; inputting the to-be-detected underwater image of the to-be-detected fish into a trained target detection model to obtain fish disease type information and fish disease position information; and determining a fish disease detection result of the to-be-detected fish based on the fish disease type information and the fish disease position information. Fish diseases are automatically detected through the target detection model obtained through training, detection operation such as manual sampling is not needed, and the fish disease detection efficiency and accuracy are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of fish disease detection, and in particular, to a fish disease detection method, device, equipment, medium and computer program product. Background Art

[0002] The existing detection of fish diseases mainly relies on manual observation and reagent detection, which is time-consuming and laborious, and it is difficult to diagnose and prevent minor diseases that cannot be observed manually. In addition, for the disease detection method based on a rapid detection kit for pathogens, although the detection speed is improved, this method requires sampling the tissues of the fish to be detected, and there are problems of complex operation and high professional requirements. Summary of the Invention

[0003] The present invention provides a fish disease detection method, device, equipment, medium and computer program product to solve the defects of low detection efficiency and accuracy in the existing fish disease detection solutions, and to achieve rapid and accurate detection of fish diseases.

[0004] The present invention provides a fish disease detection method, including the following steps: Annotate the historical underwater images of the fish to be detected to obtain a fish disease detection data set; Train an object detection model based on the fish disease detection data set; the backbone structure of the object detection model is a convolutional neural network architecture; Input the underwater image to be detected of the fish to be detected into the trained object detection model to obtain fish disease type information and fish disease location information; Determine the fish disease detection result of the fish to be detected based on the fish disease type information and the fish disease location information.

[0005] According to the fish disease detection method provided by the present invention, the annotating the historical underwater images of the fish to be detected to obtain a fish disease detection data set includes: When the number of historical underwater images of the fish to be detected meets the preset requirements, annotate the historical underwater images of the fish to be detected to obtain a fish disease detection data set; When the number of historical underwater images of the fish to be detected does not meet the preset requirements, obtain the underwater image information of the fish to be detected and the crawled images; Convert the crawled images based on the underwater image information, and use the converted images as the historical underwater images of the fish to be detected.

[0006] According to the fish disease detection method provided by the present invention, the crawled images include crawled underwater images and other crawled images; the converting the crawled images based on the underwater image information includes: When the quantity of the historical underwater images and the crawled underwater images meets the preset requirements, perform at least one of geometric transformation, scale transformation, and inter-domain transformation on the crawled underwater images based on the underwater image information; When the quantity of the historical underwater images and the crawled underwater images does not meet the preset requirements, perform at least one of geometric transformation, scale transformation, and inter-domain transformation on the crawled underwater images and the other crawled images based on the underwater image information.

[0007] According to a fish disease detection method provided by the present invention, the convolutional neural network architecture includes a fully convolutional mask autoencoder framework FCMAE; the loss function of the target detection model is a bounding box regression loss function Wise-IoU; the training of the target detection model based on the fish disease detection dataset includes: Perform regional random masking on the input images in the fish disease detection dataset based on the FCMAE; Extract the global features and local features of the input images based on the masking result to obtain a first feature map; Train the target detection model based on the first feature map and the Wise-IoU.

[0008] According to a fish disease detection method provided by the present invention, the target detection model further includes a large-scale separable convolutional attention module LSKA and a feature extraction module BiFPN; the training of the target detection model based on the first feature map and the Wise-IoU includes: Perform feature fusion on the first feature map based on the BiFPN to obtain a high-low layer fusion feature map; Perform feature extraction on the high-low layer fusion feature map based on the LSKA to obtain an attention feature map; During the process of training the target detection model based on the FCMAE, the BiFPN, and the LSKA, adjust the parameters of the target detection model based on the Wise-IoU.

[0009] According to a fish disease detection method provided by the present invention, after determining the fish disease detection result of the fish to be detected based on the fish disease type information and the fish disease location information, it includes: Obtain the feedback results of the fish disease type information and the fish disease location information; the feedback results include feedback information and feedback images; Annotate the feedback images based on the feedback information; Add the annotated feedback images to the fish disease detection dataset, and return to the step of training the target detection model based on the fish disease detection dataset.

[0010] The present invention also provides a fish disease detection device, including the following modules: An image annotation module, configured to annotate historical underwater images of fish to be detected, so as to obtain a fish disease detection data set; A model training module, configured to train an object detection model based on the fish disease detection data set; the backbone structure of the object detection model is a convolutional neural network architecture; A fish disease detection module, configured to input the underwater image to be detected of the fish to be detected into the trained object detection model, so as to obtain fish disease type information and fish disease location information; A fish disease detection result determination module, configured to determine the fish disease detection result of the fish to be detected based on the fish disease type information and the fish disease location information.

[0011] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored on the memory and running on the processor, where when the processor executes the computer program, the fish disease detection method described in any one of the above is implemented.

[0012] The present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the fish disease detection method described in any one of the above is implemented.

[0013] The present invention also provides a computer program product, including a computer program, and when the computer program is executed by a processor, the fish disease detection method described in any one of the above is implemented.

[0014] The fish disease detection method, device, equipment, medium, and computer program product provided by the present invention annotate historical underwater images of fish to be detected to obtain a fish disease detection data set for training an object detection model, and train an object detection model for detecting fish diseases based on the fish disease detection data set; after the object detection model is trained, the underwater image to be detected of the fish to be detected is inspected to obtain a detection result including the fish disease type and the fish disease location. The present application automatically detects fish diseases through the trained object detection model, without manual sampling and other detection operations, improving the efficiency and accuracy of fish disease detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention, and those of ordinary skill in the art can also obtain other drawings based on these drawings without creative efforts.

[0016] Figure 1It is one of the schematic flowcharts of the fish disease detection method provided by the present invention.

[0017] Figure 2 It is the second of the schematic flowcharts of the fish disease detection method provided by the present invention.

[0018] Figure 3 It is the schematic diagram of the network structure of the target detection model provided by the present invention.

[0019] Figure 4 It is the schematic diagram of the structural design of various large kernel attention modules provided by the present invention.

[0020] Figure 5 It is the schematic diagram of the structure of the fish disease detection device provided by the present invention.

[0021] Figure 6 It is the schematic diagram of the structure of the electronic device provided by the present invention. Detailed implementation manners

[0022] To make the objectives, technical solutions and advantages of the present invention clearer, the technical solutions in the present invention will be clearly and completely described below with reference to the accompanying drawings in the present invention. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in the present invention without making creative efforts shall fall within the protection scope of the present invention.

[0023] The following will be combined with Figures 1-6 Describe the fish disease detection method, device, equipment, medium and computer program product of the present invention.

[0024] Figure 1 It is one of the schematic flowcharts of the fish disease detection method provided by the present invention. As Figure 1 shown, the method includes the following: Step 100: Annotate the historical underwater images of the fish to be detected to obtain a fish disease detection data set; Specifically, in this embodiment, the fish to be detected is illustrated by taking Oplegnathus punctatus as an example. Based on the underwater fish image data acquisition platform built in this application, the acquisition of Oplegnathus punctatus image data is realized. Using an image annotation software (such as LabelImg, an image annotation tool), the types of fish diseases (such as skin ulcer disease and tail rot disease) and the positions of fish diseases in the Oplegnathus punctatus images are annotated to construct an Oplegnathus punctatus disease detection data set. The reason for collecting underwater fish images is that it is closer to the fish farming environment. When obtaining the fish body surface images, there is no need to fish out the fish to obtain the captured images, and the images can be automatically obtained when acquiring images, which is more suitable for the actual farming scenario.

[0025] Step 200: Train a target detection model based on the fish disease detection dataset; the backbone structure of the target detection model is a convolutional neural network architecture; When the number of underwater images of the fish to be detected in the fish disease detection dataset meets certain requirements, train the target detection model provided by this application through the underwater images of the fish to be detected. One of the differences between the target detection model provided by this application and the existing target detection models is that the backbone structure is replaced with a convolutional neural network architecture. The main differences between the target detection model provided by this application and the existing target detection models are described below.

[0026] Step 300: Input the underwater image to be detected of the fish to be detected into the trained target detection model to obtain fish disease type information and fish disease location information; After the target detection model provided by this application is trained, input the underwater image to be detected of the fish to be detected into the trained target detection model provided by this application to obtain the fish disease type information and fish disease location information output by the trained target detection model, mark the detected fish disease location in the underwater image to be detected in the form of a detection frame, and output the detected fish disease type at the fish disease location detection, that is, detect what kind of disease the fish to be detected has.

[0027] Step 400: Determine the fish disease detection result of the fish to be detected based on the fish disease type information and the fish disease location information.

[0028] Based on the fish disease type information and fish disease location information output by the trained target detection model provided by this application, check the model output result, and give feedback on inaccurate detection results or detection results whose accuracy is uncertain. After accurately modifying the inaccurate detection results, add them back to the fish disease detection dataset and re-train the model to improve the accuracy of model detection.

[0029] In this embodiment, the historical underwater images of the fish to be detected are labeled to obtain a fish disease detection dataset for training the target detection model, and a target detection model for detecting fish diseases is trained based on the fish disease detection dataset; after the target detection model is trained, the underwater image to be detected of the fish to be detected is checked to obtain a detection result including the fish disease type and the fish disease location. This application automatically detects fish diseases through the trained target detection model, eliminating the need for manual sampling and other detection operations, and improving the efficiency and accuracy of fish disease detection.

[0030] Figure 2 It is the second flow diagram of the fish disease detection method provided by the present invention. As Figure 2 shown, the method may further include: Step 110: When the number of historical underwater images of the fish to be detected meets the preset requirements, label the historical underwater images of the fish to be detected to obtain a fish disease detection dataset; Step 120: When the number of historical underwater images of the fish to be detected does not meet the preset requirements, obtain the underwater image information and crawled images of the fish to be detected; Step 130: Based on the underwater image information, convert the crawled images, and use the converted images as the historical underwater images of the fish to be detected.

[0031] Specifically, when the number of historical underwater images of the fish to be detected meets the preset requirements, the fish disease detection dataset can be amplified by an image crawling method.

[0032] Image crawling (also known as web crawling or web scraping) refers to the process of extracting images from the Internet through an automated program. The data crawling program sends Hypertext Transfer Protocol (HTTP) requests to the target web page by simulating the user's web browsing behavior, then parses the web page content, extracts the required image data, and stores the image data. When performing image crawling, some basic principles and precautions need to be followed, such as obeying the website's robots.txt file (a text file located in the root directory of the website for managing and controlling the behavior of web crawlers) and terms of use to ensure the legality of crawling; the image crawling method provided in this application respects user privacy and does not perform substantial substitution; and steps such as handling possible exceptions and data cleaning. In addition, data crawling can also combine methods such as Application Programming Interface (API) to obtain image data from social media, news platforms, or open databases to further enrich data sources and analysis means.

[0033] The underwater image information in this embodiment refers to the special information of underwater fish images (such as color and light, etc.). The crawled images also include underwater fish images and non-underwater fish images on the Internet. Perform a certain conversion on the crawled images to obtain underwater images that meet the requirements of the fish disease detection dataset.

[0034] This embodiment amplifies the fish disease detection dataset that does not meet the model training quantity requirements through image crawling and image conversion, and improves the accuracy of the target detection model of this application by increasing the model training samples.

[0035] In one embodiment, the fish disease detection method provided in the embodiments of this application may further include: Step 131. When the quantity sum of the historical underwater images and the crawled underwater images meets the preset requirements, perform at least one of geometric transformation, scale transformation, and inter-domain transformation on the crawled underwater images based on the underwater image information; Step 132. When the quantity sum of the historical underwater images and the crawled underwater images does not meet the preset requirements, perform at least one of geometric transformation, scale transformation, and inter-domain transformation on the crawled underwater images and the other crawled images based on the underwater image information.

[0036] Specifically, the algorithms for the above image transformation mainly include geometric transformation, scale transformation, and inter-domain transformation (between the spatial domain and the frequency domain).

[0037] Among them, geometric transformation mainly involves operations such as image translation, rotation, mirroring, and transposition. Geometric transformation does not change the content of the image, but only changes the position or orientation of the objects in the image. Scale transformation is mainly used to adjust the size and clarity of the image. Scale transformation mainly includes image scaling and interpolation algorithms (such as nearest neighbor interpolation, bilinear interpolation, and bicubic interpolation, etc.); Scale transformation can change the size of the image while trying to maintain the details and clarity of the image. Inter-domain transformation is mainly to reduce the computational amount or improve the processing efficiency. Sometimes, it is necessary to convert the image from the spatial domain to the frequency domain for processing. Common frequency domain transformations include Fourier transform, Walsh transform, and discrete cosine transform, etc. Inter-domain transformation can convert the spatial features of the image into frequency domain features, thus facilitating operations such as image filtering and feature extraction.

[0038] In this embodiment, the crawled images that do not meet the model training are transformed through the image transformation algorithm into images that meet the model training, and the fish disease detection data set is amplified through the image transformation, improving the accuracy of the target detection model of the present application.

[0039] In one embodiment, the fish disease detection method provided by the embodiment of the present application may further include: Step 210. Perform regional random masking on the input images in the fish disease detection data set based on the FCMAE; Step 220. Extract the global features and local features of the input images based on the masking result to obtain the first feature map; Step 230. Train the target detection model based on the first feature map and the Wise-IoU.

[0040] Specifically, the object detection model provided by this application is based on the YOLOv8 object detection algorithm, and uses the ConvNeXt V2 network (a new type of convolutional neural network architecture that integrates self-supervised learning techniques and architectural innovations) to replace the backbone structure. The ConvNeXt V2 network combines a fully convolutional masked autoencoder framework (FCMAE) and a global response normalization (GRN) layer.

[0041] Among them, the fully convolutional masked autoencoder framework is a self-supervised learning method based on convolutional neural networks. It randomly masks some regions on the input image, and then allows the model to try to recover the masked image regions. This enables the model to learn the global and local features of the input image, thereby improving the model's generalization ability. The GRN layer is introduced into the ConvNeXt architecture to enhance feature competition between channels. It solves the problem of feature collapse that may occur during training on masked inputs through three steps: global feature aggregation, feature normalization, and feature calibration.

[0042] The object detection model provided by this application uses a bounding boxes regression (BBR) loss function for object detection (Wise-Intersection Over Union, Wise-IoU) to improve the loss function and balance the image samples. The IoU is replaced with an outlier degree to evaluate the quality of the anchor boxes through a dynamic fractional monotonic focusing mechanism, so as to avoid excessive penalties on the model by geometric factors (such as distance and aspect ratio), as shown in Formulas 1 to 3.

[0043] ; (1) ; (2) ; (3) Among them, is the IoU loss, which will weaken the penalty term for high-quality anchor boxes. In the case where the overlap between the anchor box and the prediction box is relatively high, it strengthens the attention to the distance of the center point; is the penalty term of Wise-IoU, which strengthens the loss of ordinary-quality anchor boxes. The superscript * indicates not participating in backpropagation, effectively preventing the network model from generating gradients that cannot converge. is the normalization factor, representing the sliding average of the increment. is the outlier degree, A smaller value means higher anchor box quality, and a small gradient gain is assigned to it. At the same time, a smaller gradient gain can also be assigned to the prediction boxes with larger outliers, which can effectively reduce the harmful gradients of low-quality training samples, so that the bounding box regression loss focuses on the anchor boxes of ordinary quality, thereby improving the overall performance of the network.

[0044] In this embodiment, the Wise-IoU improved loss function is used to reduce the impact of sample imbalance and small targets on the detection model.

[0045] In one embodiment, the fish disease detection method provided by the embodiments of the present application may further include: Step 231: Perform feature fusion on the first feature map based on the BiFPN to obtain a high-low layer fused feature map; Step 232: Perform feature extraction on the high-low layer fused feature map based on the LSKA to obtain an attention feature map; Step 233: During the process of training the object detection model based on the FCMAE, the BiFPN, and the LSKA, adjust the parameters of the object detection model based on the Wise-IoU.

[0046] Specifically, the present application uses ConvNeXt v2 to replace the backbone structure, uses a Large Separable Kernel Attention (LSKA) and a Bi-directional Feature Pyramid Network (BiFPN), a neural network architecture for object detection and semantic segmentation tasks, to improve feature extraction, and uses Wise-IoU as the model loss function. The network structure is as Figure 3 shown. Among them, Conv represents the Convolutional Layer; Unsample represents downsampling; Fusion represents feature fusion; Block represents the basic unit for constructing the network; Detect is a key step or module in the object detection task, and Detect is responsible for identifying and locating the target objects of interest from the image; F5 is a deeper feature layer.

[0047] LSKA decomposes the two-dimensional convolution kernel of the depth convolution layer into cascaded horizontal and vertical one-dimensional convolution kernels, enabling the direct use of the depth convolution layer with a large convolution kernel in the attention module without additional modules, reducing the number of parameters and computational complexity. Compared with the existing Large Kernel Attention (LKA), as Figure 4As shown, in the LKA-trivial module, 2D large kernel depth convolution is combined with 1×1 convolution for feature extraction. In the LSKA-trivial module, feature extraction is performed by combining cascaded horizontal and vertical 1D large kernel depth convolution with 1×1 convolution. In the LKA module, in the Visual Attention Network (VAN), standard depth convolution and dilated depth convolution kernels are combined with 1×1 convolution for feature extraction. In the LSKA module provided in this application, the first two layers of LKA are decomposed into four layers, each layer consisting of two 1D convolutional layers. Among them, k represents the maximum receptive field, and d represents the dilation rate. Figure 4 In Figure 4 , DW.Conv represents Depthwise Separable Convolution; DW.D.Conv is a complete depthwise separable convolution operation, and DW.Conv is only a part of it.

[0048] ; (4) ; (5) The BiFPN module includes two stages: the top-down stage and the bottom-up stage. In the top-down stage, the high-level feature maps are refined through upsampling and fused with the low-level feature maps, so that the high-level feature maps obtain richer context information. In the bottom-up stage, the low-level feature maps are coarsened through downsampling and fused with the high-level feature maps, so that the low-level feature maps obtain more detailed information. Each layer in the BiFPN structure is connected to the two adjacent layers above and below. This two-way connection design can effectively fuse and interact the feature information of different levels, improving the feature expression ability. The BiFPN network realizes feature fusion through a weighted feature fusion mechanism, convolution, and downsampling. Taking as an example, is obtained by weighted fusion of the input of the 6th layer , and the of the 5th layer. is the intermediate feature of the 6th layer in the top-down direction. The expressions of and are shown in formulas 4 and 5 respectively. Among them, , , , and are weight coefficients; is the downsampling operation; is a parameter used to avoid numerical instability caused by too small weight parameters.

[0049] In this embodiment, the LSKA is used to reduce the number of model parameters and the computational complexity; the BiFPN module is used to improve the image feature expression ability.

[0050] In one embodiment, the fish disease detection method provided by the embodiments of the present application may further include: Step 500, obtaining a feedback result of the fish disease type information and the fish disease location information; the feedback result includes feedback information and a feedback image; Step 600, annotating the feedback image based on the feedback information; Step 700, adding the annotated feedback image to the fish disease detection data set, and returning to the step of training the target detection model based on the fish disease detection data set.

[0051] Specifically, the present application further provides a method for feedback of model output results. By feeding back the fish disease type information and the fish disease location information output by the target detection model provided by the present application, the feedback information includes whether the fish disease type information and the fish disease location information are accurate, and then determining the images with inaccurate fish disease detection results, that is, the feedback images in this embodiment. Then, annotate the feedback images with accurate fish disease type information and fish disease location information, add the annotated feedback images to the fish disease detection data set, and retrain the target detection model provided by the present application.

[0052] In this embodiment, through the feedback method of model output, the accuracy of model detection is further improved.

[0053] Next, the fish disease detection device provided by the present invention will be described. The fish disease detection device described below can be mutually corresponded and referred to the fish disease detection method described above.

[0054] Please refer to Figure 5 , the present invention further provides a fish disease detection device, including: An image annotation module 501, configured to annotate historical underwater images of fish to be detected to obtain a fish disease detection data set; A model training module 502, configured to train a target detection model based on the fish disease detection data set; the backbone structure of the target detection model is a convolutional neural network architecture; A fish disease detection module 503, configured to input the underwater image to be detected of the fish to be detected into the trained target detection model to obtain fish disease type information and fish disease location information; A fish disease detection result determination module 504, configured to determine the fish disease detection result of the fish to be detected based on the fish disease type information and the fish disease location information.

[0055] Optionally, the image annotation module includes: An image annotation unit, configured to annotate historical underwater images of the fish to be detected to obtain a fish disease detection dataset when the number of historical underwater images of the fish to be detected meets a preset requirement; An image crawling unit, configured to obtain underwater image information and crawled images of the fish to be detected when the number of historical underwater images of the fish to be detected does not meet the preset requirement; An image conversion unit, configured to convert the crawled images based on the underwater image information and use the converted images as the historical underwater images of the fish to be detected.

[0056] Optionally, the crawled images include crawled underwater images and other crawled images; the image conversion unit includes: A first image conversion unit, configured to perform at least one of geometric conversion, scale conversion, and inter-domain conversion on the crawled underwater images based on the underwater image information when the sum of the number of the historical underwater images and the crawled underwater images meets the preset requirement; A second image conversion unit, configured to perform at least one of geometric conversion, scale conversion, and inter-domain conversion on the crawled underwater images and the other crawled images based on the underwater image information when the sum of the number of the historical underwater images and the crawled underwater images does not meet the preset requirement.

[0057] Optionally, the convolutional neural network architecture includes a fully convolutional masked autoencoder framework FCMAE; the loss function of the object detection model is a bounding box regression loss function Wise-IoU; the model training module includes: An image region random masking unit, configured to perform region random masking processing on the input images in the fish disease detection dataset based on the FCMAE; A first feature map determination unit, configured to extract the global features and local features of the input images based on the masking processing result to obtain a first feature map; An object detection model training unit, configured to train the object detection model based on the first feature map and the Wise-IoU.

[0058] Optionally, the object detection model further includes a large-scale separable convolutional attention module LSKA and a feature extraction module BiFPN; the object detection model training unit includes: A feature fusion unit, configured to perform feature fusion on the first feature map based on the BiFPN to obtain a high-low layer fusion feature map; A feature extraction unit, configured to perform feature extraction on the high-low layer fusion feature map based on the LSKA to obtain an attention feature map; A model parameter adjustment unit, configured to adjust the parameters of the target detection model based on the Wise-IoU during the process of training the target detection model based on the FCMAE, the BiFPN, and the LSKA.

[0059] Optionally, the fish disease detection device further includes: A feedback result acquisition module, configured to acquire the feedback results of the fish disease type information and the fish disease location information; the feedback results include feedback information and a feedback image; A feedback image annotation module, configured to annotate the feedback image based on the feedback information; A fish disease detection dataset update module, configured to add the annotated feedback image to the fish disease detection dataset, and return to the step of training the target detection model based on the fish disease detection dataset.

[0060] Figure 6 An example of the physical structure diagram of an electronic device is shown as Figure 6 shown. The electronic device may include: a processor 610, a communication interface 620, a memory 630, and a communication bus 640. Among them, the processor 610, the communication interface 620, and the memory 630 communicate with each other through the communication bus 640. The processor 610 can call the logical instructions in the memory 630 to execute the fish disease detection method, which includes: annotating the historical underwater images of the fish to be detected to obtain a fish disease detection dataset; training a target detection model based on the fish disease detection dataset; the backbone structure of the target detection model is a convolutional neural network architecture; inputting the underwater image to be detected of the fish to be detected into the trained target detection model to obtain fish disease type information and fish disease location information; determining the fish disease detection result of the fish to be detected based on the fish disease type information and the fish disease location information.

[0061] In addition, when the logical instructions in the above-mentioned memory 630 are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0062] On the other hand, the present invention also provides a computer program product. The computer program product includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the fish disease detection method provided by the above-mentioned various methods. The method includes: annotating historical underwater images of fish to be detected to obtain a fish disease detection data set; training a target detection model based on the fish disease detection data set; the backbone structure of the target detection model is a convolutional neural network architecture; inputting the underwater image to be detected of the fish to be detected into the trained target detection model to obtain fish disease type information and fish disease location information; based on the fish disease type information and the fish disease location information, determining the fish disease detection result of the fish to be detected.

[0063] On another aspect, the present invention also provides a non-transitory computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it is implemented to execute the fish disease detection method provided by the above-mentioned various methods. The method includes: annotating historical underwater images of fish to be detected to obtain a fish disease detection data set; training a target detection model based on the fish disease detection data set; the backbone structure of the target detection model is a convolutional neural network architecture; inputting the underwater image to be detected of the fish to be detected into the trained target detection model to obtain fish disease type information and fish disease location information; based on the fish disease type information and the fish disease location information, determining the fish disease detection result of the fish to be detected.

[0064] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0065] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the essence of the above technical solution or the part that contributes to the prior art can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0066] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements for some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of each embodiment of the present invention.

Claims

1. A fish disease detection method, characterized in that, Including: Annotating historical underwater images of fish to be detected to obtain a fish disease detection dataset; Training an object detection model based on the fish disease detection dataset; The backbone structure of the object detection model is a convolutional neural network architecture; Inputting the underwater image to be detected of the fish to be detected into the trained object detection model to obtain fish disease type information and fish disease location information; Based on the fish disease type information and the fish disease location information, determining the fish disease detection result of the fish to be detected.

2. The fish disease detection method according to claim 1, characterized in that The annotating the historical underwater images of the fish to be detected to obtain a fish disease detection dataset includes: When the number of historical underwater images of the fish to be detected meets the preset requirements, annotating the historical underwater images of the fish to be detected to obtain a fish disease detection dataset; When the number of historical underwater images of the fish to be detected does not meet the preset requirements, obtaining the underwater image information and crawled images of the fish to be detected; Based on the underwater image information, converting the crawled images, and using the converted images as the historical underwater images of the fish to be detected.

3. The fish disease detection method according to claim 2, characterized in that, The crawled images include crawled underwater images and other crawled images; the converting the crawled images based on the underwater image information includes: When the sum of the number of the historical underwater images and the crawled underwater images meets the preset requirements, performing at least one of geometric transformation, scale transformation, and domain transformation on the crawled underwater images based on the underwater image information; When the sum of the number of the historical underwater images and the crawled underwater images does not meet the preset requirements, performing at least one of geometric transformation, scale transformation, and domain transformation on the crawled underwater images and the other crawled images based on the underwater image information.

4. The fish disease detection method according to claim 1, characterized in that, The convolutional neural network architecture includes a fully convolutional masked autoencoder framework FCMAE; the loss function of the object detection model is a bounding box regression loss function Wise-IoU; the training the object detection model based on the fish disease detection dataset includes: Performing regional random masking on the input images in the fish disease detection dataset based on the FCMAE; Extracting the global features and local features of the input images based on the masking result to obtain a first feature map; Training the object detection model based on the first feature map and the Wise-IoU.

5. The fish disease detection method according to claim 4, characterized in that, The object detection model further includes a large-scale separable convolutional attention module LSKA and a feature extraction module BiFPN; The training the object detection model based on the first feature map and the Wise-IoU includes: Performing feature fusion on the first feature map based on the BiFPN to obtain a high-low layer fusion feature map; Performing feature extraction on the high-low layer fusion feature map based on the LSKA to obtain an attention feature map; During the process of training the object detection model based on the FCMAE, the BiFPN, and the LSKA, adjusting the parameters of the object detection model based on the Wise-IoU.

6. The fish disease detection method according to claim 1, characterized in that Based on the fish disease type information and the fish disease location information, determining the fish disease detection result of the fish to be detected, which includes the following steps: Obtaining the feedback results of the fish disease type information and the fish disease location information; the feedback results include feedback information and feedback images; Annotating the feedback images based on the feedback information; Adding the annotated feedback images to the fish disease detection data set, and returning to the step of training the object detection model based on the fish disease detection data set.

7. A fish disease detection device, characterized in that, It includes: An image annotation module, configured to annotate historical underwater images of the fish to be detected to obtain a fish disease detection data set; A model training module, configured to train an object detection model based on the fish disease detection data set; the backbone structure of the object detection model is a convolutional neural network architecture; A fish disease detection module, configured to input the underwater image to be detected of the fish to be detected into the trained object detection model to obtain fish disease type information and fish disease location information; A fish disease detection result determination module, configured to determine the fish disease detection result of the fish to be detected based on the fish disease type information and the fish disease location information.

8. An electronic device, comprising a memory, a processor, and a computer program stored on the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the fish disease detection method according to any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the fish disease detection method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the fish disease detection method according to any one of claims 1 to 6.

Images