A fish school automatic feeding control method based on visual recognition

Abstract

The present application relates to the technical field of intelligent fishery, in particular to a fish school automatic feeding control method based on visual identification, which comprises the following steps: adopting a camera with a holder, being equipped with an infrared / enhanced light source, being installed at a suitable position to cover the fish school feeding area, continuously collecting water surface image videos, obtaining fish school feeding water flower activity data, reducing noise through mean value, median value and Gaussian filtering, selecting a target detection model, judging the fish feeding state to determine the feeding operation, driving the feeding equipment operation according to the decision result, setting an alarm threshold, recording operation logs, generating a strategy report by using a big data algorithm, and obtaining the feeding operation, monitoring screen and strategy report. The present application solves the problem that in traditional breeding, due to the lack of effective monitoring means, it is difficult to accurately obtain fish school feeding water flower data and to realize intelligent and efficient breeding.

Description

Technical Field

[0001] This invention belongs to the field of smart fishery technology, specifically relating to a method for automatic feeding control of fish schools based on visual recognition. Background Technology

[0002] In the field of smart fisheries technology, precise feeding is a crucial step in improving aquaculture efficiency, reducing costs, and ensuring the healthy growth of fish. Traditional feeding methods rely heavily on human experience, leading to inaccurate feeding amounts and timing. This not only wastes feed and increases costs but can also cause water quality deterioration due to excessive feed, impacting the fish's habitat. While some automated feeding devices have been implemented, most lack effective monitoring and feedback mechanisms, failing to provide precise feeding based on the actual feeding behavior of the fish. For example, some devices operate only according to preset times and amounts, making it difficult to adapt to the dynamic changes in fish feeding needs. Existing technologies also have many shortcomings in monitoring fish feeding. Ordinary cameras struggle to capture clear surface images and videos under complex lighting conditions, such as at night or in low light, making it impossible to accurately obtain data on fish feeding activity. The acquired image and video data are often affected by noise, water surface reflection, and water wave interference, resulting in poor image quality and affecting the accuracy of subsequent analysis and processing. Meanwhile, there is a lack of efficient and accurate methods for detecting and tracking the location of splashes, making it difficult to determine the range of splash locations in real time and track their changes. Moreover, there is a lack of scientific and reasonable basis and standards for judging the feeding status of fish, making it impossible to accurately judge the feeding situation of fish based on changes in the range of splashes, and thus make reasonable feeding decisions.

[0003] Existing technologies suffer from several shortcomings: poor imaging performance of acquisition equipment in complex environments, limited image preprocessing effects, poor accuracy in water splash detection and tracking, inability to scientifically determine fish feeding status, difficulty in achieving precise feeding, and lack of a sound monitoring and strategy generation mechanism. Summary of the Invention

[0004] To address the aforementioned issues, this invention provides a visual recognition-based automatic fish feeding control method. This method solves the problems in traditional aquaculture where the lack of effective monitoring methods hinders accurate acquisition of fish feeding data, leading to unscientific and inaccurate feeding decisions. Furthermore, the lack of real-time monitoring and strategy generation mechanisms makes intelligent and efficient aquaculture difficult to achieve. To achieve the above objectives, this invention adopts the following technical solution: The aforementioned visual recognition-based automatic fish feeding control method includes the following steps: A pan-tilt-zoom camera, equipped with an infrared / enhanced light source, is installed at a suitable location to cover the fish feeding area, continuously acquiring water surface images and videos to obtain data on fish feeding splashes and other water activity, thus obtaining water surface image and video data; noise reduction is achieved through mean, median, and Gaussian filtering; contrast enhancement is performed using histogram equalization; water surface reflection is removed using image segmentation techniques; and water wave interference is removed using an image stabilization algorithm, resulting in preprocessed image and video data; a target detection model is selected, trained using labeled splash image data, and parameters are adjusted; the trained model is then applied to real-time data to detect the location and range of splashes in real time. The system marks the locations of the splashes, obtaining images of the marked splash areas. Kalman filtering and mean-shift algorithms are used to track the splash range across frames. Motion information and appearance features are used to infer and recover the range of obscured splashes, resulting in tracked splash range change data. The system continuously monitors the splash range change trend, sets a change threshold based on experience, compares the current splash range with the threshold, determines the fish feeding status, and decides on the feeding operation, obtaining the feeding decision result. Based on the decision result, the system drives the feeding equipment operation, establishes a real-time monitoring system to display data, sets alarm thresholds, records operation logs, and uses big data algorithms to generate a strategy report, resulting in feeding operations, monitoring images, and a strategy report.

[0005] Furthermore, the method of using a pan-tilt camera equipped with an infrared / enhanced light source, installed in a suitable location to cover the fish feeding area, continuously collecting water surface images and videos, and obtaining data on fish feeding splashes and other water surface image data includes the following steps: Using a camera with pan-tilt functionality and equipped with infrared and enhanced light sources, based on the actual layout of the aquaculture site and the fish activity patterns, the camera is precisely installed in a position that can fully cover the fish feeding area, ensuring no blind spots; by adjusting the pan-tilt angle and focal length, the camera is focused on the target area, and the camera is activated to continuously collect images and video information of the water surface area; the camera accurately captures the dynamic images of water splashes generated when the fish feed, extracting data that reflects the fish feeding activity, and obtaining complete and clear water surface image and video data.

[0006] Furthermore, the preprocessed image and video are obtained by using mean, median, and Gaussian filtering for noise reduction, histogram equalization to enhance contrast, image segmentation to remove water surface reflections, and image stabilization algorithms to remove water wave interference. This process includes the following steps: Mean filtering, median filtering, and Gaussian filtering are used to process each frame of the acquired image and video to remove noise points and smooth the image; histogram equalization is used to enhance image contrast, making the difference between water splashes and the background more obvious; image segmentation is used to accurately locate and extract the water surface reflection area, removing it from the image; and image stabilization algorithms are used to eliminate image distortion and blurring caused by water waves, extracting effective information from the original data to obtain a high-quality preprocessed image and video.

[0007] Furthermore, the selection of the target detection model, trained using labeled water splash image data, adjusting parameters, and applying the trained model to real-time data to detect and mark the location and range of water splashes in real time, resulting in an image with the marked water splash location and range, includes the following steps: selecting a target detection model and training the selected model using image data with labeled water splash location and range information; iteratively optimizing the model by adjusting parameters based on its performance on the training and validation sets to improve detection accuracy; analyzing real-time acquired image and video frames using the target detection model to extract water splash features, accurately locate the water splash location and determine its range, and presenting the detection results on the image in labeled form to obtain an image with the marked water splash location and range.

[0008] Furthermore, the method of using Kalman filtering and mean-shift algorithm to track the water splash range across frames, and using motion information and appearance features to infer and recover the range of the obscured water splash, to obtain the tracked water splash range change data, includes the following steps: using Kalman filtering algorithm, based on the motion information of the water splash's position and speed in the previous frame, predicting the possible position range in the current frame; using mean-shift algorithm, searching for regions matching the appearance features of the water splash within the predicted region of the current frame to achieve cross-frame tracking; when the water splash is obscured, extracting the motion trajectory, speed, and appearance features of color and shape before the water splash is obscured; using prediction and matching functions, inferring and recovering the range of the obscured water splash, continuously recording the changes in the water splash range, and obtaining the tracked water splash range change data.

[0009] Furthermore, the continuous monitoring of the splash range change trend, setting a change threshold based on experience, comparing the current splash range with the threshold, determining the fish's feeding status, and deciding on the feeding operation to obtain the feeding decision judgment result includes the following steps: Real-time continuous monitoring and tracking of the splash range; analyzing the dynamic change trend of different frame data to form a continuous time series curve; setting an initial threshold and a minimum threshold based on long-term aquaculture experience, corresponding to the critical states of the fish actively feeding and basically satiated, respectively; comparing the current splash range with the threshold, determining the fish's feeding status, and deciding on the feeding operation; comprehensively analyzing the results to output a clear feeding decision instruction, thus obtaining the feeding decision judgment result.

[0010] Furthermore, the process of driving the feeding equipment operation based on the decision results, establishing a real-time monitoring system to display data, setting alarm thresholds, recording operation logs, and generating strategy reports using big data algorithms to obtain feeding operations, monitoring screens, and strategy reports includes the following steps: using automated control equipment to precisely drive the motor of the feeding equipment to start based on the feeding decision results, completing the quantitative feeding operation; establishing a real-time monitoring module through sensor integration, synchronously displaying dynamic water splash images, equipment operating status, and aquaculture environment parameters on a visual interface, setting alarm thresholds for dissolved oxygen and water temperature parameters, and automatically pushing alarm information when triggered; recording the entire process of feeding time and water splash changes in the operation log, and mining historical data patterns through big data analysis algorithms to obtain dynamic strategy reports for feeding optimization and health warnings.

[0011] Furthermore, the mean-shift algorithm is used to search for regions matching the appearance features of the water splash within the current frame prediction area to achieve cross-frame tracking. When the water splash is obscured, the motion trajectory, speed, color, and shape appearance features of the water splash before obscuration are extracted, including the following steps: cross-frame tracking is performed using the mean-shift algorithm, and the current frame prediction area is determined based on the water splash position in the previous frame; a feature model is constructed by analyzing the appearance features of the water splash's color distribution and texture structure, and the region with the highest matching degree is searched within the current frame prediction area to complete the water splash location; when the water splash is obscured, the motion trajectory and speed information in the data from multiple frames before obscuration are immediately extracted, and the color value and shape outline appearance features of the water splash in a stable state are recorded, and the frame data is used to predict the possible location and shape of the obscured water splash.

[0012] Furthermore, the step of comparing the current splash range with a threshold to determine the fish's feeding status and decide on the feeding operation, and outputting a clear feeding decision instruction based on the comprehensive analysis results, to obtain the feeding decision judgment result, includes the following steps: comparing the splash range value detected in the current frame with a preset threshold; if it exceeds the initial threshold, it is determined that the fish are actively feeding and feeding should continue; if it is below the minimum threshold, it is determined that the fish are satiated and feeding should stop; and outputting a clear feeding decision instruction based on the comprehensive analysis results, to obtain the feeding decision judgment result.

[0013] In the technical solution provided by this invention, a pan-tilt camera equipped with an infrared / enhanced light source is used and installed at a suitable location to cover the feeding area of ​​the fish school, continuously acquiring water surface images and videos to obtain data on the fish feeding splash activity, thus obtaining water surface image and video data. Noise reduction is achieved through mean, median, and Gaussian filtering; contrast enhancement is achieved through histogram equalization; water surface reflection is removed using image segmentation technology; and water wave interference is removed using an image stabilization algorithm, resulting in preprocessed images and videos. A target detection model is selected, trained using labeled splash image data, and parameters are adjusted. The trained model is then applied to real-time data to detect and mark the location and range of splashes in real time, resulting in labeled data. The invention generates images of the water splash location range; it uses Kalman filtering and mean shift algorithms to track the water splash range across frames, and uses motion information and appearance features to infer and recover the range of obscured water splashes, obtaining data on the changes in the tracked water splash range; it continuously monitors the trend of water splash range changes, sets a change threshold based on experience, compares the current water splash range with the threshold to determine the fish feeding status and decides on feeding operations, obtaining the feeding decision judgment result; it drives the operation of the feeding equipment based on the decision result, establishes a real-time monitoring system to display data, sets alarm thresholds, records operation logs, and uses big data algorithms to generate strategy reports, obtaining feeding operations, monitoring screens, and strategy reports. This invention solves the problems in traditional aquaculture where the lack of effective monitoring methods makes it impossible to accurately obtain water splash data of fish feeding, leading to unscientific and inaccurate feeding decisions, as well as the lack of real-time monitoring and strategy generation mechanisms, making it difficult to achieve intelligent and efficient aquaculture. Attached Figure Description

[0014] Various other advantages and benefits will become apparent to those skilled in the art upon reading the following detailed description of preferred embodiments. The accompanying drawings are for illustrative purposes only and are not intended to limit the invention.

[0015] Figure 1 This is a schematic diagram of the first embodiment of a visual recognition-based automatic fish feeding control method according to the present invention.

[0016] Figure 2 This is a schematic diagram of a second embodiment of a visual recognition-based automatic fish feeding control method according to the present invention.

[0017] Figure 3 This is a schematic diagram of a third embodiment of a visual recognition-based automatic fish feeding control method according to the present invention.

[0018] Figure 4 This is a schematic diagram of the fourth embodiment of a visual recognition-based automatic feeding control method for fish swarms in this invention.

[0019] Figure 5 This is a schematic diagram of the fifth embodiment of a visual recognition-based automatic feeding control method for fish swarms in this invention. Detailed Implementation

[0020] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0021] Those skilled in the art will understand that, unless specifically stated otherwise, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the term “comprising” as used in this specification means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof.

[0022] A method for automatic feeding control of fish schools based on visual recognition, such as Figure 1 As shown, the process includes the following steps: A camera with a gimbal, equipped with an infrared / enhanced light source, is installed at a suitable location to cover the fish feeding area, continuously acquiring water surface images and videos to obtain data on fish feeding splashes and other water activity, thus obtaining water surface image and video data; noise reduction is achieved through mean, median, and Gaussian filtering; contrast is enhanced using histogram equalization; water surface reflection is removed using image segmentation techniques; and water wave interference is removed using an image stabilization algorithm, resulting in preprocessed images and videos; a target detection model is selected, trained using labeled splash image data, parameters are adjusted, and the trained model is applied to real-time data to detect and label the location and range of splashes in real time, obtaining labeled water splash data. The image shows the range of the splash area; Kalman filtering and mean shift algorithm are used to track the splash range across frames, and motion information and appearance features are used to infer and recover the range of obscured splashes, obtaining the splash range change data after tracking; the splash range change trend is continuously monitored, and a change threshold is set based on experience. The current splash range is compared with the threshold to determine the fish feeding status and decide on the feeding operation, obtaining the feeding decision judgment result; the feeding equipment operation is driven according to the decision result, a real-time monitoring system is established to display the data screen, alarm thresholds are set, operation logs are recorded, and a strategy report is generated using big data algorithms, obtaining the feeding operation, monitoring screen, and strategy report.

[0023] like Figure 2As shown, in this embodiment, a camera with a gimbal function and equipped with infrared and enhanced light sources is used. Based on the actual layout of the aquaculture site and the activity patterns of the fish, the camera is precisely installed in a position that can fully cover the feeding area of ​​the fish, ensuring no blind spots. By adjusting the angle and focal length of the gimbal, the camera is focused on the target area, and the camera is started to continuously collect images and video information of the water surface area. The camera accurately captures the dynamic images of water splashes generated when the fish feed, extracts data that reflects the feeding activities of the fish, and obtains complete and clear water surface image and video data.

[0024] Cameras equipped with gimbals and infrared and enhanced light sources can be precisely installed according to the site layout and fish activity patterns, providing comprehensive coverage of the feeding area, eliminating blind spots, and ensuring the integrity of data collection. By adjusting the gimbal angle and focus on the target area, dynamic images of water splashes when fish feed can be clearly captured, obtaining data reflecting their activity and producing high-quality water surface images and videos. This not only provides a reliable basis for accurate analysis of fish feeding status but also helps fish farmers to promptly grasp fish dynamics, rationally adjust breeding strategies, effectively improve breeding efficiency and quality, and reduce breeding costs.

[0025] like Figure 3 As shown, in this embodiment, mean filtering, median filtering, and Gaussian filtering are used to process the acquired images and videos frame by frame to remove noise points and make the image smoother. Histogram equalization technology is used to enhance the contrast of the image, making the difference between water splashes and the background more obvious. With the help of image segmentation technology, the water surface reflection area is accurately located and extracted, and the water surface reflection area is removed from the image. Image stabilization algorithm is used to eliminate image deformation and blur caused by water waves, extract effective information from the original data, and obtain high-quality preprocessed images and videos.

[0026] By processing images and videos frame by frame using mean, median, and Gaussian filtering, noise points can be effectively removed, making the image smoother and reducing interference factors. Histogram equalization technology enhances image contrast, making the difference between water splashes and the background more significant, facilitating subsequent identification and analysis. Image segmentation technology accurately locates and removes water surface reflection areas, preventing them from interfering with the judgment of water splashes. Image stabilization algorithms eliminate image distortion and blurring caused by water waves, extracting the most effective information from the original data. After a series of processing steps, high-quality pre-processed images and videos are obtained, providing a solid foundation for accurate detection and tracking of water splashes.

[0027] like Figure 4As shown, in this embodiment, a target detection model is selected, and image data with pre-labeled water splash location and range information is used to train the selected model. Through iterative optimization, the parameters are adjusted according to the model's performance on the training and validation sets to improve detection accuracy. The target detection model is used to analyze real-time acquired image and video frames, extract water splash features, accurately locate the water splash location and determine the range, and present the detection results on the image in the form of labels to obtain an image with labeled water splash location and range.

[0028] A target detection model was selected and trained using labeled water splash image data. Through iterative optimization and parameter adjustment, the model's accuracy in detecting water splashes was significantly improved, effectively adapting to the water splash detection needs in different scenarios. In practical applications, the model can quickly analyze real-time acquired image and video frames, accurately extract water splash features, determine their location and range, and present them intuitively in a labeled form. This not only provides a reliable basis for accurately tracking changes in water splashes but also allows aquaculture personnel to clearly understand the feeding situation of fish schools, make timely and reasonable decisions, and improve the scientific and precise nature of aquaculture management.

[0029] like Figure 5 As shown, in this embodiment, the Kalman filter algorithm is used to predict the possible location range of the splash in the current frame based on the position and speed motion information of the splash in the previous frame; the mean drift algorithm is used to search for regions that match the appearance features of the splash in the prediction area of ​​the current frame to achieve cross-frame tracking; when the splash is occluded, the motion trajectory, speed, and appearance features of the splash before occlusion are extracted; through the prediction and matching function, the range of the occluded splash is inferred and restored, and the changes in the splash range are continuously recorded to obtain the splash range change data after tracking.

[0030] The Kalman filter algorithm, leveraging motion information from the previous frame, can accurately predict the possible location range of splashes in the current frame, providing a basic direction for subsequent tracking. The mean-shift algorithm searches for matching appearance features within the prediction region, achieving efficient cross-frame tracking. Addressing the challenge of splash occlusion, the algorithm extracts motion and appearance features before occlusion and uses prediction and matching functions to infer and recover the occluded range, ensuring tracking continuity. Continuously recording changes in the splash range provides data that accurately reflects the feeding activity of fish, offering strong and reliable data support for accurately judging fish feeding behavior and formulating scientific feeding strategies.

[0031] In this embodiment, the range of splashing water is continuously monitored in real time. By analyzing the dynamic change trend of different frames of data, a continuous time series curve is formed. Based on long-term aquaculture experience, an initial threshold and a minimum threshold are set, which correspond to the critical states of fish actively feeding and basically being full, respectively. The current range of splashing water is compared with the threshold to determine the feeding status of the fish and decide on the feeding operation. The comprehensive analysis results output a clear feeding decision instruction and obtain the feeding decision judgment result.

[0032] Real-time, continuous monitoring and tracking of the water splash range generates a time-series curve, clearly presenting dynamic changes in the splash and accurately capturing subtle characteristics of fish feeding activities. Based on long-term aquaculture experience, initial and minimum thresholds are set, providing a scientifically sound standard for judging fish feeding status. Comparing the current splash range with the threshold accurately determines whether the fish are actively feeding or nearly full, thus allowing for appropriate feeding decisions. Comprehensive analysis outputs clear feeding instructions, effectively preventing overfeeding or underfeeding and improving feed utilization.

[0033] In this embodiment, an automated control device is used to precisely drive the motor of the feeding equipment based on the feeding decision results, thereby completing the quantitative feeding operation. A real-time monitoring module is established through sensor integration, which synchronously displays the dynamic water splash image, equipment operating status, and aquaculture environment parameters on a visual interface. Alarm thresholds for dissolved oxygen and water temperature parameters are set, and alarm information is automatically pushed when triggered. The entire process records the feeding time and water splash changes in the operation log. Through big data analysis algorithms, historical data patterns are mined to obtain dynamic strategy reports on feeding optimization and health warning.

[0034] The automated control equipment precisely drives the feeding motor based on feeding decisions, achieving quantitative feeding, avoiding human error, and improving feeding accuracy and efficiency. The real-time monitoring module, through sensor integration, synchronously presents key information on a visual interface. Setting alarm thresholds allows for timely alerts, ensuring aquaculture safety. The entire operation log is recorded, and big data analytics algorithms are used to uncover patterns in historical data. The resulting dynamic strategy reports not only optimize feeding plans and reduce aquaculture costs but also provide health warnings.

[0035] In this embodiment, the mean-shift algorithm is used for cross-frame tracking. The prediction area of ​​the current frame is determined based on the position of the splash in the previous frame. A feature model is constructed by analyzing the appearance features of the splash color distribution and texture structure. The area with the highest matching degree is searched within the prediction area of ​​the current frame to complete the splash location. When the splash is occluded, the motion trajectory and speed information in the data of multiple frames before the occlusion are immediately extracted. At the same time, the color value and shape outline appearance features of the splash in a stable state are recorded. The frame data is used to predict the possible position and shape of the occluded splash.

[0036] The mean-shift algorithm determines the prediction region for the current frame based on the splash position of the previous frame, narrowing the search range and improving tracking efficiency. By constructing appearance feature models such as splash color distribution and texture structure, it accurately searches and matches within the prediction region to achieve splash location and ensure tracking accuracy. When a splash is occluded, it extracts the motion trajectory, velocity, and stable appearance features from multiple frames before the occlusion, using this information to predict the position and shape of the occluded splash, effectively solving the occlusion problem and ensuring tracking continuity.

[0037] In this embodiment, the range of water splashes detected in the current frame is compared with a preset threshold. If the value exceeds the initial threshold, it is determined that the fish are actively feeding and feeding should continue. If the value is below the minimum threshold, it is determined that the fish are satiated and feeding should stop. The comprehensive analysis results output a clear feeding decision instruction, and the feeding decision judgment result is obtained.

[0038] The system compares the current frame's splash range value with a preset threshold, offering a simple and efficient way to quickly assess the feeding status of fish. If the splash exceeds the initial threshold, the system determines the fish are actively feeding and continues feeding to meet their growth needs and promote healthy development. If the splash falls below the minimum threshold, the system determines the fish are satiated and stops feeding, preventing feed waste and reducing aquaculture costs. The comprehensive analysis outputs clear feeding decision instructions, providing a clear operational guide for aquaculture personnel.

[0039] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely preferred examples and are not intended to limit the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A method for automatic feeding control of fish schools based on visual recognition, characterized in that, The automatic feeding control method for fish schools includes the following steps: Using a camera with a gimbal and equipped with an infrared / enhanced light source, it is installed in a suitable position to cover the feeding area of ​​fish and continuously collect water surface images and videos to obtain data on fish feeding splashes and water surface image and video data. The image and video are preprocessed by noise reduction through mean, median and Gaussian filtering, contrast enhancement by histogram equalization, removal of water surface reflection by image segmentation, and removal of water wave interference by image stabilization algorithm. Select an object detection model and train it using image data with pre-labeled water splash location and range information. By iteratively optimizing the model, parameters are adjusted based on the model's performance on the training and validation sets to improve detection accuracy. The target detection model analyzes real-time acquired image and video frames, extracts water splash features, accurately locates the water splash position and determines its range, and presents the detection results on the image in the form of markers, thus obtaining an image with marked water splash position ranges. The Kalman filter algorithm is used to predict the possible location range of water splashes in the current frame based on the position and velocity motion information of the water splashes in the previous frame. The mean-shift algorithm is used to search for regions that match the appearance features of water splashes within the prediction region of the current frame to achieve cross-frame tracking. When water splashes are obscured, the motion trajectory, speed, color, and shape appearance features of the water splashes before obscuration are extracted. By using the prediction and matching functions, the range of water splashes in the obscured part is inferred and restored, and the changes in the range of water splashes are continuously recorded to obtain the tracked water splash range change data. The system employs real-time continuous monitoring to track the range of water splashes, and analyzes the dynamic changing trends of different frames of data to form a continuous time series curve. Based on long-term aquaculture experience, initial and minimum thresholds are set, corresponding to the critical states of fish actively feeding and basically being full, respectively. The current water splash range is compared with the threshold to determine the fish's feeding status and decide on the feeding operation. The comprehensive analysis results output a clear feeding decision instruction, and the feeding decision judgment result is obtained. The feeding equipment is operated based on the feeding decision judgment results. A real-time monitoring system is established to display data, alarm thresholds are set, operation logs are recorded, and a strategy report is generated using big data algorithms, resulting in feeding operation, monitoring screen, and strategy report. The method employs a mean-shift algorithm to search for regions matching the appearance features of water splashes within the current frame's prediction area, enabling cross-frame tracking. When a water splash is obscured, the method extracts its trajectory, speed, color, and shape features before obscuration, including the following steps: The mean-shift algorithm is used to perform cross-frame tracking, and the prediction area of ​​the current frame is determined based on the position of the water splash in the previous frame. By analyzing the color distribution and texture structure of the water splashes, a feature model is constructed. The area with the highest matching degree is searched within the prediction area of ​​the current frame to complete the location of the water splashes. When a splash appears to be obstructed, the motion trajectory and speed information in the data from multiple frames before the obstruction are immediately extracted. At the same time, the color value and shape outline of the splash in a stable state are recorded. The frame data is used to predict the possible location and shape of the obstructed splash.

2. The method for automatic feeding control of fish schools based on visual recognition according to claim 1, characterized in that, The method employs a pan-tilt camera equipped with an infrared / enhanced light source, installed in a suitable location to cover the fish feeding area, continuously acquiring water surface images and videos, obtaining data on fish feeding splashes, and obtaining water surface image and video data, including the following steps: A camera with a gimbal function and equipped with infrared and enhanced light source is used. Based on the actual layout of the breeding site and the activity pattern of the fish, the camera is precisely installed in a position that can fully cover the feeding area of ​​the fish to ensure that there are no blind spots in the shooting. By adjusting the angle and focal length of the gimbal, the camera is focused on the target area, and the camera is started to continuously collect images and video information of the water surface area. The camera accurately captures the dynamic images of water splashes produced when fish feed, extracts data that reflects the feeding activities of the fish, and obtains complete and clear water surface image and video data.

3. The method for automatic feeding control of fish schools based on visual recognition according to claim 1, characterized in that, The process of denoising using mean, median, and Gaussian filters, enhancing contrast with histogram equalization, removing water surface reflections using image segmentation techniques, and removing water wave interference using image stabilization algorithms to obtain the preprocessed image / video includes the following steps: Mean filtering, median filtering, and Gaussian filtering are used to process the acquired images and videos frame by frame to remove noise points in the images and make the picture smoother. Histogram equalization is used to enhance image contrast, making the difference between water splashes and background more obvious. With the help of image segmentation technology, the water surface reflection area is accurately located and extracted, and then removed from the image. Image stabilization algorithms are used to eliminate image distortion and blurring caused by water waves, extract effective information from the original data, and obtain high-quality preprocessed images and videos.

4. The method for automatic feeding control of fish schools based on visual recognition according to claim 1, characterized in that, The process of driving the feeding equipment operation based on the feeding decision judgment result, establishing a real-time monitoring system to display data, setting alarm thresholds, recording operation logs, and generating strategy reports using big data algorithms to obtain feeding operations, monitoring screens, and strategy reports includes the following steps: The automated control equipment is used to precisely drive the motor of the feeding equipment to start based on the feeding decision judgment results, so as to complete the quantitative feeding operation. A real-time monitoring module is established by integrating sensors, which synchronously displays dynamic water splashes, equipment operating status and aquaculture environment parameters on a visual interface. Alarm thresholds for dissolved oxygen and water temperature parameters are set, and alarm information is automatically pushed when triggered. The entire process is recorded in logs, including feeding time and water splash changes. By using big data analysis algorithms to uncover patterns in historical data, dynamic strategy reports are generated, including optimized feeding plans and health warnings.

5. The method for automatic feeding control of fish schools based on visual recognition according to claim 1, characterized in that, The process of comparing the current water splash range with a threshold to determine the fish's feeding status and decide on the feeding operation, and outputting a clear feeding decision instruction based on the comprehensive analysis results, thus obtaining the feeding decision judgment result, includes the following steps: The range of water splashes detected in the current frame is compared with a preset threshold. If the value exceeds the initial threshold, it is determined that the fish are actively feeding and feeding should continue. When the fish population falls below the minimum threshold, it is determined that the fish are satiated and feeding should be stopped. The comprehensive analysis results output a clear feeding decision instruction, thus obtaining the feeding decision judgment result.

Images