Fish feeding method, device, computer equipment and storage medium

Abstract

The present invention provides a method, device, computer equipment and storage medium for feeding fish schools, which belongs to the field of aquaculture technology. The method includes: constructing a training sample of fish school feeding behavior; constructing a fish school feeding prediction model; obtaining a behavior image of a target fish school; inputting the behavior image into the fish school feeding prediction model to obtain a predicted feeding intensity of the target fish school; and feeding the target fish school according to the predicted feeding intensity. In this way, the behavior of the fish school is predicted according to the fish school feeding model to obtain the feeding intensity of the fish school, and the feeding machine is adjusted in real time according to the feeding intensity of the fish school to feed the fish school. In this way, the intelligent feeding method can monitor and analyze the feeding behavior of the fish school in real time, realize the real-time adjustment of the feeding strategy, improve the accuracy and efficiency of feeding, and can adapt to different breeding scenarios and needs. In addition, it can also reduce the feeding cost and labor cost, and improve the breeding efficiency and economic benefits.

Description

Technical Field

[0001] The invention belongs to the technical field of aquaculture, and in particular relates to a method and device for feeding fish, a computer device and a storage medium. Background Art

[0002] In the current fish farming industry, feeding is a very important part of the farming process, which directly affects the growth and development of fish and the farming efficiency. However, the current traditional feeding methods have many limitations. For example, manual feeding requires a lot of manpower and material resources. In order to save manpower, feeding machines are usually used for feeding. However, traditional feeding machines can only feed at set times, resulting in low farming efficiency.

[0003] In order to improve the breeding efficiency, the existing technology also proposes to use convolutional neural networks to identify the feeding behavior of fish schools and feed them according to the identification results. However, only using convolutional neural networks to identify the feeding behavior of fish schools has, firstly, a weak ability to capture local information, which may result in errors in the identification of fish school behavior. Secondly, when there are many fish schools, the complexity of the calculation will increase rapidly, and the reaction time required will be long, which is not conducive to real-time feeding of the fish schools. Therefore, there is still a problem of low breeding efficiency. Summary of the invention

[0004] In order to solve the above-mentioned problem of low fish farming efficiency, the present invention provides a fish feeding method, device, computer equipment and storage medium.

[0005] In order to achieve the above object, the present invention provides the following technical solutions:

[0006] A method for feeding a school of fish, comprising:

[0007] Constructing a fish feeding prediction model, specifically comprising: adding a focused linear attention FLatten module to a head module of an EfficientVit model to obtain an improved EfficientVit model; the improved EfficientVit model comprises an input module, a feature extraction backbone module and a head module; training the improved EfficientVit model with fish feeding behavior training samples to obtain a fish feeding prediction model;

[0008] Obtain behavioral images of target fish schools;

[0009] The behavior image is input into the fish school feeding prediction model, the input behavior image is received by the input module, multi-scale feature extraction is performed on the received behavior image by the feature extraction backbone module, and the extracted multi-scale features are fused to obtain target features, and the target features are classified by the head module to obtain the predicted feeding intensity of the target fish school;

[0010] The target fish population is fed according to the predicted feeding intensity.

[0011] Optionally, the feature extraction backbone module includes a preprocessing module, multiple MBConv modules and EfficientVit Module modules and a feature fusion module; the preprocessing module includes a Conv module and a DSConv module, and the input image is preliminarily extracted by the preprocessing module to obtain an original feature map; the original feature map is further feature extracted by multiple MBConv modules to obtain multiple MBConv feature maps; multi-scale feature extraction is performed on multiple MBConv feature maps by multiple EfficientVitModule modules to obtain a multi-scale feature pyramid; the feature fusion module uses a 1×1 convolution kernel and upsampling to adjust the spatial size and number of channels of the multi-scale feature pyramid, and fuses the multi-scale feature pyramid by addition to obtain the target feature.

[0012] Optionally, a multi-scale linear attention module Multi-Scale LinearAttention is introduced into the EfficientVit Module, and the multi-scale linear attention module is used to perform high-resolution dense prediction on the MBConv feature map to achieve the global receptive field; the ReLU linear attention mechanism is used to capture contextual information to achieve the global receptive field; in addition, a combination of FFN+DWConv is introduced to capture local information and enhance the ReLU linear attention.

[0013] Optionally, classifying the target features by the head module includes:

[0014] The target feature is classified into eating intensity by the head module, and the eating intensity is 0-10, where 0 means no desire to eat and 10 means strong desire to eat.

[0015] Optionally, the improved EfficientVit model is trained by using fish feeding behavior training samples to obtain a fish feeding prediction model including:

[0016] Acquire images of the target fish school's feeding behavior in three stages before, during and after historical feeding, and construct a fish school feeding behavior training sample based on the images; the fish school feeding behavior training sample includes the target fish school's training behavior image and the corresponding actual feeding intensity;

[0017] Inputting the training behavior image into the improved EfficientVit model to obtain the predicted feeding intensity of the target fish school;

[0018] The improved EfficientVit model was trained with the goal of minimizing the deviation between the predicted feeding intensity and the actual feeding intensity to obtain the fish feeding prediction model.

[0019] Optionally, obtaining the behavior image of the target fish school includes:

[0020] The behavior images of the target fish school are acquired at a preset frequency through an image acquisition device at a fixed position.

[0021] Optionally, feeding the target fish school according to the predicted feeding intensity includes:

[0022] According to the dynamic changes of the predicted feeding intensity of the target fish school, the feeding frequency and feeding amount of the bait throwing machine are synchronously and dynamically adjusted until the predicted feeding intensity of the fish school is 0, and the bait throwing machine stops working.

[0023] A fish feeding device, comprising:

[0024] A construction module is used to construct a fish feeding prediction model, specifically comprising: adding a focused linear attention FLatten module to a head module of an EfficientVit model to obtain an improved EfficientVit model; the improved EfficientVit model comprises an input module, a feature extraction backbone module and a head module; the improved EfficientVit model is trained by a fish feeding behavior training sample to obtain a fish feeding prediction model;

[0025] An acquisition module is used to acquire the behavior images of the target fish school;

[0026] A prediction module, used for inputting the behavior image into the fish school feeding prediction model, receiving the input behavior image through the input module, performing multi-scale feature extraction on the received behavior image through the feature extraction backbone module, and fusing the extracted multi-scale features to obtain target features, and classifying the target features through the head module to obtain the predicted feeding intensity of the target fish school;

[0027] A feeding module is used to feed the target fish school according to the predicted feeding intensity.

[0028] A computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the fish feeding method is implemented.

[0029] A computer device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the fish feeding method is implemented when the processor executes the program.

[0030] The fish feeding method provided by the present invention has the following beneficial effects:

[0031] First, the focused linear attention module is added to the head module of the EfficientVit model, so that high-resolution dense prediction can be performed, the global receptive field can be realized, and then the multi-scale feature extraction of the overall situation of the fish school can be performed, which is conducive to improving the accuracy of the prediction of the feeding behavior of the fish school. Moreover, the focused linear attention module can achieve higher expressiveness and lower computational complexity by balancing the efficiency and expression ability when solving the shortcomings of the linear attention mechanism performance degradation and introducing additional computational overhead to deal with the mapping function, so that the response speed and accuracy of the model are improved. Then, the behavior of the fish school is predicted according to the fish school feeding model, and the feeding intensity of the fish school is obtained. According to the feeding intensity of the fish school, the feeding machine is adjusted in real time to feed the fish school. In this way, the intelligent feeding method can monitor and analyze the feeding behavior of the fish school in real time, realize the real-time adjustment of the feeding strategy, improve the accuracy and efficiency of feeding, and can adapt to different breeding scenarios and needs, which is conducive to improving breeding efficiency. In addition, it can also reduce the feeding cost and labor cost, and improve the breeding efficiency and economic benefits. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the embodiment of the present invention and its design scheme, the following briefly introduces the drawings required for this embodiment. The drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0033] Figure 1 The present invention is a schematic flow chart of a method for feeding a school of fish according to an exemplary embodiment of the present invention.

[0034] Figure 2 The present invention is a structural block diagram of a fish feeding prediction model provided according to an exemplary embodiment of the present invention.

[0035] Figure 3 The present invention is a schematic diagram of an intelligent fish feeding system provided according to an exemplary embodiment.

[0036] Figure 4 The present invention is a block diagram of a fish feeding device provided according to an exemplary embodiment.

[0037] Explanation of the reference numerals: 1-circulating water aquaculture pond; 2-intelligent bait throwing machine; 3-camera; 4-power supply device; 5-PLC (Programmable Logic Controller); 6-microcomputer; 7-bracket. DETAILED DESCRIPTION

[0038] In order to enable those skilled in the art to better understand the technical solution of the present invention and implement it, the present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. The following embodiments are only used to more clearly illustrate the technical solution of the present invention, and cannot be used to limit the scope of protection of the present invention.

[0039] With the rapid development of deep learning technology, visual models have made remarkable achievements in the field of computer vision. The application of deep learning technology in visual models has achieved remarkable results. Many deep learning models have performed well in image recognition, target detection, and behavior analysis, providing a solid technical foundation for intelligent fish feeding methods. Deep learning technology can automatically learn and extract features from large-scale data, and has strong generalization and adaptability. By training visual models, real-time monitoring and analysis of fish behavior can be achieved, so as to more accurately control the feeding strategy. Based on this, the visual model combined with deep learning technology provides a feasible solution for intelligent fish feeding. In addition, the development of deep learning technology has also brought more efficient computing and learning algorithms, which enables visual models to have better performance and accuracy when processing large-scale data and complex scenes. At the same time, with the continuous advancement of hardware equipment, such as the application of GPU acceleration, edge computing and other technologies, the computing speed and efficiency of visual models have been further improved, providing strong support for the realization of intelligent fish feeding methods.

[0040] The intelligent fish feeding method based on visual model mainly includes three steps: data collection, model training and real-time feeding control. First, the aquaculture waters are monitored in real time by installing sensors such as cameras to collect fish behavior data. Then, the visual model is trained using the collected fish behavior data, and the collected data is analyzed and learned to achieve the recognition and prediction of fish behavior. Finally, according to the output results of the visual model, intelligent feeding control is achieved to adjust the feeding amount and frequency to meet the needs of different fish and environmental changes.

[0041] Compared with traditional feeding methods, intelligent feeding methods based on visual models have many advantages. Intelligent feeding methods can monitor and analyze the feeding behavior of fish in real time, realize real-time adjustment of feeding strategies, improve the accuracy and efficiency of feeding, and can adapt to different farming scenarios and needs. In addition, it can also reduce feeding costs and labor costs, and improve farming benefits and economic benefits.

[0042] The technical solutions provided by various embodiments of the present invention are described in detail below in conjunction with the accompanying drawings.

[0043] First, the present invention provides a method for feeding fish schools, specifically as follows: Figure 1 As shown, the following steps are included:

[0044] S101. Construct a fish feeding prediction model.

[0045] Specifically, the focused linear attention FLatten module is added to the head module of the EfficientVit model to obtain an improved EfficientVit model; the improved EfficientVit model includes an input module, a feature extraction backbone module and a head module. Figure 2 As shown; the input module receives the input image, the feature extraction backbone module performs multi-scale feature extraction on the received image, and the extracted multi-scale features are fused to obtain the target features, and the target features are classified by the head module; the improved EfficientVit model is trained by the fish feeding behavior training sample to obtain the fish feeding prediction model.

[0046] Among them, the feature extraction backbone module includes a preprocessing module, multiple MBConv modules and EfficientVitModule modules and a feature fusion module. The preprocessing module includes a Conv module and a DSConv module. The input image is initially extracted by the preprocessing module to obtain an original feature map; the original feature map is further extracted by multiple MBConv modules to obtain multiple MBConv feature maps; multi-scale feature extraction is performed on multiple MBConv feature maps by multiple EfficientVit Module modules to obtain a multi-scale feature pyramid; the feature fusion module uses a 1×1 convolution kernel and upsampling to adjust the spatial size and number of channels of the multi-scale feature pyramid, and fuses the multi-scale feature pyramid by addition to obtain the target feature.

[0047] For example, the feature extraction backbone module may include a preprocessing module, stage 1, stage 2, stage 3 and stage 4, the preprocessing module includes 1 Conv module and 1 DSConv module; stage 1 includes 1 MBConv module; stage 2 includes 1 MBConv module; stage 3 includes 1 and MBConv module and 1 EfficientVit Module module; stage 4 includes 1 and MBConv module and 1 EfficientVit Module module.

[0048] A multi-scale linear attention module Multi-ScaleLinearAttention is introduced in the EfficientVit Module, through which high-resolution dense prediction of MBConv feature maps is performed to achieve a global receptive field; the ReLU linear attention mechanism is used to capture contextual information to achieve a global receptive field; in addition, since the ReLU linear attention has the limitation of not being able to generate a concentrated attention map and thus has a weak ability to capture local information, a combination of FNN+DWConv is introduced to capture local information and enhance the ReLU linear attention.

[0049] In addition, in order to enable the fish school feeding prediction model to accurately predict the feeding intensity of the fish school, the fish school feeding prediction model needs to be trained.

[0050] Specifically, images of the target fish school's feeding behavior in three historical stages, before, during and after feeding, are obtained, and a fish school feeding behavior training sample is constructed based on the images.

[0051] The fish school feeding behavior training sample includes the training behavior image of the target fish school and the corresponding actual feeding intensity.

[0052] Specifically, a historical video of fish feeding can be obtained, and the obtained historical video is framed, and the feeding behavior pictures of the fish at each stage obtained after the frame processing are screened, and similar feeding behavior pictures are deleted to obtain a fish feeding behavior data set; the final fish feeding behavior data set is subjected to image classification processing. After the data set classification processing is completed, the pictures in the data set are randomly allocated in a ratio of 7:2:1 to form a training set, a validation set and a test set, and then the picture data in the training set is subjected to data enhancement processing.

[0053] For example, the image data can be enhanced by using the Cutout method to randomly cut off some areas in the sample and fill them with 0 pixel values, and at the same time, the image is processed by combining translation, rotation and scaling to obtain the final training set.

[0054] In one embodiment, the training behavior image is input into an improved EfficientVit model to obtain a predicted feeding intensity of a target fish school; the improved EfficientVit model is trained with the goal of minimizing the deviation between the predicted feeding intensity and the actual feeding intensity to obtain a fish school feeding prediction model.

[0055] For example, EfficientVit is used as the backbone network to extract image features, and the linear attention mechanism FLatten module is added to the head module to better focus on the key information in the image, thereby improving the recognition accuracy. The improved EfficientVit model parameters are initialized using the EfficientVit classification task weights, and the improved EfficientVit model is trained using the training set. After each round of training, the validation set is used to adjust the trained model until the prediction curve converges, thereby constructing an improved EfficientVit model suitable for the bait casting task; then the test set is used to test the effect of the improved EfficientVit model; if the test is qualified, the improved EfficientVit model is determined to be a fish feeding prediction model.

[0056] It should be noted that the target features are classified into feeding intensity through the head module. Specifically, the feeding intensity of the fish is classified according to the feeding state of the fish when the bait is fed. The feeding intensity of the fish is divided into 0 to 10, where 0 means no feeding desire and 10 means strong feeding desire.

[0057] S102: Obtaining a behavior image of the target fish school.

[0058] In this step, the behavior image of the target fish school can be obtained at a preset frequency by an image acquisition device at a fixed position. The image acquisition device can be a video camera or a camera.

[0059] For example, the behavior image of the target fish school is obtained through a camera; the camera used is located N m above the fish pond, and the camera collects images of the fish school behavior in the fish pond at a rate of 20 frames per second; then the collected video is extracted at a frequency of 1 frame per second to obtain the required image data set.

[0060] S103, inputting the behavior image into the fish school feeding prediction model to obtain the predicted feeding intensity of the fish school.

[0061] Specifically, the acquired target fish school behavior image is input into the fish school feeding prediction model, the input behavior image is received through the input module, multi-scale feature extraction is performed on the received behavior image through the feature extraction backbone module, and the extracted multi-scale features are fused to obtain the target features, and the target features are classified through the head module, so as to obtain the predicted feeding intensity of the target fish school in real time.

[0062] S104: feeding the target fish school according to the predicted feeding intensity.

[0063] Specifically, according to the dynamic changes of the predicted feeding intensity of the target fish school, the feeding frequency and feeding amount of the bait throwing machine are synchronously and dynamically adjusted until the predicted feeding intensity of the fish school is 0, and then the bait throwing machine stops working.

[0064] By adopting the above method, a fish feeding model is obtained according to the feeding behavior of the fish school, and then the fish school behavior is predicted according to the feeding model to obtain the feeding intensity of the fish school. According to the feeding intensity of the fish school, the feeding machine is adjusted in real time to feed the fish school. In this way, the intelligent feeding method can monitor and analyze the feeding behavior of the fish school in real time, realize the real-time adjustment of the feeding strategy, improve the accuracy and efficiency of feeding, and adapt to different breeding scenarios and needs. In addition, it can also reduce the feeding cost and labor cost, and improve the breeding efficiency and economic benefits.

[0065] In addition, the present invention also provides a fish school intelligent feeding system, specifically as follows Figure 3 As shown, the hardware components include a circulating water aquaculture pond 1, an intelligent bait throwing machine 2, a camera 3, a power supply device 4, a PLC 5, a microcomputer 6 and a bracket 7.

[0066] The feeding machine 2 is installed on a springboard 3 to 4 meters offshore next to the breeding pond, 0.4 meters above the highest water surface of the pond. The direction of the feeding port of the feeding machine 2 is consistent with the direction of the monitoring window of the camera 3. The microcomputer 6 is connected to the PLC5 and the camera 3 at the same time. The camera 3 is used to collect the behavior video of the fish school and transmit the collected video data to the microcomputer 6. The microcomputer 6 is used to analyze and process the video data according to the steps of the fish feeding method of the present invention, and then transmit the processing results to the PLC5. The PLC5 controls the feeding amount and feeding frequency of the feeding machine 2 according to the processing results of the microcomputer 6.

[0067] Secondly, the present invention also provides a fish feeding device, such as Figure 4 As shown, including:

[0068] The construction module 401 is used to construct a fish feeding prediction model, which specifically includes: adding the focused linear attention FLatten module to the head module of the EfficientVit model to obtain an improved EfficientVit model; the improved EfficientVit model includes an input module, a feature extraction backbone module and a head module; the improved EfficientVit model is trained with fish feeding behavior training samples to obtain a fish feeding prediction model.

[0069] The acquisition module 402 is used to acquire the behavior image of the target fish school.

[0070] The prediction module 403 is used to input the behavior image into the fish feeding prediction model, receive the input behavior image through the input module, perform multi-scale feature extraction on the received behavior image through the feature extraction backbone module, and fuse the extracted multi-scale features to obtain target features, and classify the target features through the head module to obtain the predicted feeding intensity of the target fish school.

[0071] The feeding module 404 is used to feed the target fish school according to the predicted feeding intensity.

[0072] By using the above device, the focused linear attention module is first added to the head module of the EfficientVit model, so that high-resolution dense prediction can be performed, the global receptive field is realized, and then the multi-scale feature extraction of the overall situation of the fish school is performed, which is conducive to improving the accuracy of the prediction of the feeding behavior of the fish school. Moreover, the focused linear attention module can achieve higher expressiveness and lower computational complexity by balancing the efficiency and expression ability when solving the shortcomings of the linear attention mechanism performance degradation and introducing additional computational overhead to process the mapping function, so that the reaction speed and accuracy of the model are improved. Then, the behavior of the fish school is predicted according to the fish school feeding model, and the feeding intensity of the fish school is obtained. According to the feeding intensity of the fish school, the feeding machine is adjusted in real time to feed the fish school. In this way, the intelligent feeding method can monitor and analyze the feeding behavior of the fish school in real time, realize the real-time adjustment of the feeding strategy, improve the accuracy and efficiency of feeding, and can adapt to different breeding scenarios and needs, which is conducive to improving breeding efficiency. In addition, it can also reduce the feeding cost and labor cost, and improve the breeding efficiency and economic benefits.

[0073] The present invention also provides a computer-readable storage medium, which stores a computer program, which can be used to execute the above Figure 1 Steps for the fish feeding method provided.

[0074] The present invention also provides a computer device. At the hardware level, the computer device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory. Of course, it may also include hardware required for other services. The processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to achieve the above Figure 1 Steps for the fish feeding method provided.

[0075] It will be appreciated by those skilled in the art that embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0076] The present invention is described with reference to flowcharts and / or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as a combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0077] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0078] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0079] It should be noted that the above specific implementation method can enable those skilled in the art to understand the invention more comprehensively, but does not limit the invention in any way. Therefore, although the present invention has been described in detail in this specification, those skilled in the art should understand that the invention can still be modified or replaced by equivalents; and all technical solutions and improvements that do not deviate from the spirit and scope of the invention are included in the protection scope of the patent for the invention. Any figure mark in the claims should not be regarded as limiting the claims involved.

Claims

1. A method for feeding fish, characterized in that: The method comprises: Constructing a fish feeding prediction model: adding the focused linear attention FLatten module to the head module of the EfficientVit model to obtain an improved EfficientVit model; the improved EfficientVit model includes an input module, a feature extraction backbone module and a head module; training the improved EfficientVit model with fish feeding behavior training samples to obtain a fish feeding prediction model; The training process includes: obtaining images of the target fish school's feeding behavior in three stages before, during and after feeding, and constructing a fish school feeding behavior training sample based on the images; the fish school feeding behavior training sample includes the target fish school's training behavior image and the corresponding actual feeding intensity; Inputting the training behavior image into the improved EfficientVit model to obtain the predicted feeding intensity of the target fish school; The improved EfficientVit model was trained with the goal of minimizing the deviation between the predicted feeding intensity and the actual feeding intensity to obtain the fish feeding prediction model. Obtain behavioral images of target fish schools; The behavior image is input into the fish school feeding prediction model, the input behavior image is received by the input module, multi-scale feature extraction is performed on the received behavior image by the feature extraction backbone module, and the extracted multi-scale features are fused to obtain target features, and the target features are classified by the head module to obtain the predicted feeding intensity of the target fish school; feeding the target fish group according to the predicted feeding intensity; The feature extraction backbone module includes a preprocessing module, multiple MBConv modules, an EfficientVit Module module, and a feature fusion module; the preprocessing module includes a Conv module and a DSConv module, and the input image is preliminarily extracted by the preprocessing module to obtain an original feature map; the original feature map is further feature extracted by multiple MBConv modules to obtain multiple MBConv feature maps; multiple EfficientVit Module modules are used to extract multi-scale features from multiple MBConv feature maps to obtain a multi-scale feature pyramid; the feature fusion module uses a 1×1 convolution kernel and upsampling to adjust the spatial size and number of channels of the multi-scale feature pyramid, and fuses the multi-scale feature pyramid by addition to obtain the target feature; A multi-scale linear attention module Multi-Scale Linear Attention is introduced in the EfficientVit Module, which performs high-resolution dense prediction on the MBConv feature map to achieve the global receptive field; the ReLU linear attention mechanism is used to capture contextual information to achieve the global receptive field; in addition, the combination of FFN+DWConv is introduced to capture local information and enhance the ReLU linear attention.

2. The fish feeding method according to claim 1, characterized in that: Classifying target features by the head module includes: The target feature is classified into eating intensity by the head module, and the eating intensity is 0-10, where 0 means no desire to eat and 10 means strong desire to eat.

3. The fish feeding method according to claim 1, characterized in that: The obtaining of the behavior image of the target fish school comprises: The behavior images of the target fish school are acquired at a preset frequency through an image acquisition device at a fixed position.

4. The fish feeding method according to claim 1, characterized in that: Feeding the target fish population according to the predicted feeding intensity includes: According to the dynamic changes of the predicted feeding intensity of the target fish school, the feeding frequency and feeding amount of the bait throwing machine are synchronously and dynamically adjusted until the predicted feeding intensity of the fish school is 0, and the bait throwing machine stops working.

5. A fish feeding device, characterized in that: The device comprises: A construction module is used to construct a fish feeding prediction model: adding a focused linear attention FLatten module to the head module of an EfficientVit model to obtain an improved EfficientVit model; the improved EfficientVit model includes an input module, a feature extraction backbone module and a head module; the improved EfficientVit model is trained by using fish feeding behavior training samples to obtain a fish feeding prediction model; An acquisition module is used to acquire the behavior images of the target fish school; A prediction module, used for inputting the behavior image into the fish school feeding prediction model, receiving the input behavior image through the input module, performing multi-scale feature extraction on the received behavior image through the feature extraction backbone module, and fusing the extracted multi-scale features to obtain target features, and classifying the target features through the head module to obtain the predicted feeding intensity of the target fish school; A feeding module, used for feeding the target fish school according to the predicted feeding intensity; The feature extraction backbone module includes a preprocessing module, multiple MBConv modules and EfficientVit Module modules and a feature fusion module; the preprocessing module includes a Conv module and a DSConv module; The prediction module is also used to perform preliminary extraction on the input image through the preprocessing module to obtain an original feature map; further perform feature extraction on the original feature map through multiple MBConv modules to obtain multiple MBConv feature maps; perform multi-scale feature extraction on multiple MBConv feature maps through multiple EfficientVit Module modules to obtain a multi-scale feature pyramid; adjust the spatial size and channel number of the multi-scale feature pyramid by using a 1×1 convolution kernel and upsampling through the feature fusion module, and fuse the multi-scale feature pyramid by addition to obtain the target feature; The building module is also used to introduce a multi-scale linear attention module Multi-Scale Linear Attention in the EfficientVit Module module; The prediction module is also used to perform high-resolution dense prediction on the MBConv feature map through the multi-scale linear attention module to achieve a global receptive field; the ReLU linear attention mechanism is used to capture context information to achieve a global receptive field; in addition, a combination of FFN+DWConv is introduced to capture local information and enhance ReLU linear attention; The construction module is also used to obtain images of the three stages of feeding behavior of the target fish school before, during and after historical feeding, and construct fish school feeding behavior training samples based on the images; the fish school feeding behavior training samples include the training behavior images of the target fish school and the corresponding actual feeding intensity; the training behavior images are input into the improved EfficientVit model to obtain the predicted feeding intensity of the target fish school; the improved EfficientVit model is trained with the goal of minimizing the deviation between the predicted feeding intensity and the actual feeding intensity to obtain a fish school feeding prediction model.

6. A computer-readable storage medium, characterized in that: The storage medium stores a computer program, and when the computer program is executed by the processor, the fish feeding method according to any one of claims 1 to 4 is implemented.

7. A computer device, characterized in that: The method comprises a memory, a processor and a computer program stored in the memory and executable on the processor, wherein the fish feeding method according to any one of claims 1 to 4 is implemented when the processor executes the program.

Images