Attitude analysis-based ingestion assessment feeding method, system and equipment and medium

Abstract

The invention relates to a feeding evaluation feeding method, system and device based on attitude analysis and a medium. The method comprises the steps that a preprocessed feeding area video stream and water quality environment data corresponding to the feeding area video stream are acquired; performing attitude estimation on the preprocessed feeding area video stream to obtain fish body key point time sequence data; according to the fish body key point time sequence data, microcosmic motion features of fish body behavior postures are extracted, a dense optical flow field is generated based on the preprocessed feeding area video stream to calculate motion confusion entropy, the microcosmic motion features and the motion confusion entropy are fused, and a feeding desire feature sequence is obtained; performing time sequence analysis on the ingestion desire feature sequence to obtain an ingestion desire index; and generating a corresponding feeding control instruction according to the ingestion desire index and the water quality environment data, and performing feeding control on the feeding machine according to the feeding control instruction. The device has the effect of improving the feeding accuracy.

Description

Technical Field

[0001] This invention belongs to the technical field of intelligent management of aquaculture, and in particular relates to a feeding assessment and feeding method, system, equipment and medium based on posture analysis. Background Technology

[0002] Currently, feeding operations in aquaculture are usually carried out continuously in open water environments such as ponds and cages. The feeding process is affected by various factors such as the growth stage of the fish, the environmental conditions such as dissolved oxygen and water temperature, and water surface disturbance. On-site, it is necessary to ensure aquaculture efficiency while taking into account feed utilization and water quality stability.

[0003] Existing feeding control methods mostly rely on human experience or feed fish at fixed times and in fixed amounts, and make simple adjustments based on the level of activity in the video footage. This can easily lead to misjudgments due to water surface reflections, wave disturbances, or fish swimming, resulting in a mismatch between feeding control and the fish's actual feeding desire, leading to feed waste and increased water quality risks. Summary of the Invention

[0004] The purpose of this invention is to provide a feeding assessment method, system, device, and medium based on posture analysis to solve the technical problem that existing feeding control methods deviate from the actual needs because the judgment criteria are inconsistent with the feeding desire of the fish.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: In a first aspect, the present invention provides a feeding assessment method based on posture analysis, the method comprising: Acquire the pre-processed video stream of the feeding area and the corresponding water quality environmental data; Pose estimation is performed on the preprocessed video stream of the feeding area to obtain temporal data of key points on the fish body; Microscopic motion features of fish behavior and posture are extracted based on the time-series data of key points on the fish body, and a dense optical flow field is generated based on the preprocessed video stream of the feeding area to calculate motion disorder entropy. The microscopic motion features and the motion disorder entropy are fused to obtain a feeding desire feature sequence, and then the feeding desire feature sequence is subjected to time-series analysis to obtain a feeding desire index. Based on the feeding desire index and the water quality data, a corresponding feeding control command is generated, and the feeding machine is controlled according to the feeding control command.

[0006] By adopting the above technical solutions, and by acquiring the pre-processed video stream of the feeding area and the corresponding water quality environmental data, synchronous visual and environmental information can be provided for feeding desire assessment, thereby improving the adaptability of feeding decisions to changes in the aquaculture environment. By performing posture estimation on the pre-processed video stream of the feeding area to obtain the time-series data of key points of the fish body, the fish behavior changes can be stably depicted from the structured posture information of the fish body, thereby reducing the influence of water surface reflection and background disturbance on feeding judgment. By extracting micro-motion features and generating a dense optical flow field to calculate the motion disorder entropy and fuse them to obtain the feeding desire index, the difference in feeding state between individual action intensity and group motion disorder can be reflected simultaneously, thereby improving the accuracy of distinguishing between normal cruising and competing for feeding. By generating feeding control commands based on the feeding desire index and water quality environmental data and executing feeding control, the feeding intensity can be dynamically adjusted according to changes in feeding desire, thereby reducing the risk of feed waste and water quality deterioration caused by overfeeding.

[0007] In one example, the present invention can be further configured as follows: acquiring the preprocessed video stream of the feeding area and the water quality environmental data corresponding to the video stream of the feeding area includes: Acquire video images of the feeding area and generate a video stream of the feeding area; perform image enhancement processing on the video stream of the feeding area to obtain enhanced video frames; Denoising is performed on the enhanced video frame to obtain a preprocessed video frame, and a preprocessed video stream of the feeding area is generated based on the preprocessed video frame. Dissolved oxygen and water temperature in the feeding area are collected to generate the water quality environmental data, and the collection time is added to the water quality environmental data. The water quality environmental data is correlated with the preprocessed feeding area video stream according to the acquisition time to obtain the water quality environmental data corresponding to the preprocessed feeding area video stream.

[0008] By adopting the above technical solutions, and by acquiring video images of the feeding area, generating a video stream of the feeding area, and performing image enhancement processing, the recognizability of fish outlines and key details in the feeding area video can be improved, thereby enhancing the stability of subsequent posture estimation and motion analysis. By performing noise reduction processing on the enhanced video frames and generating a pre-processed video stream of the feeding area, noise interference introduced by water turbidity and wave textures can be suppressed, thereby reducing the impact of false detections and missed detections on feeding assessment results. By acquiring dissolved oxygen and water temperature to generate water quality environmental data and attaching acquisition time markers, the environmental state at the time of feeding can be accurately reflected, thereby ensuring that the basis for feeding control is consistent with the on-site water quality conditions. By associating the water quality environmental data with the pre-processed video stream of the feeding area according to the acquisition time, time alignment and synchronous input of multi-source data can be achieved, thereby improving the real-time performance and consistency of feeding desire assessment and feeding control.

[0009] In one example, the present invention can be further configured as follows: performing pose estimation on the preprocessed video stream of the feeding area to obtain temporal data of key points on the fish body includes: The fish target is located in the preprocessed video stream of the feeding area to obtain the fish target region in each video frame; The target area of ​​the fish body is subjected to pose estimation processing to obtain the head key points, dorsal fin key points and tail key points; The head key points, dorsal fin key points and tail key points in adjacent video frames are temporally correlated to obtain key point trajectories, and then the fish body key point temporal data is generated based on the key point trajectories.

[0010] By employing the above technical solutions, the target area of ​​the fish can be obtained by locating the fish target in the preprocessed feeding area video stream. This allows the analysis scope to be focused on the effective area of ​​the fish, thereby reducing the interference of background fluctuations and floating objects on key point recognition. By performing pose estimation on the target area of ​​the fish, key points of the head, dorsal fin, and tail can be obtained, forming structured features that reflect the fish's orientation and posture changes, thus improving the ability to describe the fish's behavior and posture. By temporally associating key points in adjacent video frames to obtain key point trajectories and generating temporal data of fish key points, a continuous and stable posture change sequence can be obtained, thus providing reliable input for subsequent micro-motion feature extraction and improving the accuracy of feeding desire assessment.

[0011] In one example, the present invention can be further configured as follows: extracting micro-motion features of fish behavior and posture based on the time-series data of key points on the fish body, generating a dense optical flow field based on the preprocessed video stream of the feeding area to calculate motion disorder entropy, fusing the micro-motion features with the motion disorder entropy to obtain a feeding desire feature sequence, and then performing time-series analysis on the feeding desire feature sequence to obtain a feeding desire index, including: The motion sequence of key points is calculated based on the time-series data of key points on the fish body, and a micro motion feature sequence is generated based on the motion sequence of key points. Dense optical flow calculations are performed on adjacent video frames in the preprocessed video stream of the feeding area to obtain a dense optical flow field, and a motion disorder entropy sequence is generated based on the dense optical flow field; The micro-motion feature sequence and the motion disorder entropy sequence are fused to obtain the feeding desire feature sequence. The food desire feature sequence is segmented based on a preset time sliding window to obtain a windowed feature sequence, and time series analysis is performed on the windowed feature sequence to generate the food desire index.

[0012] By employing the above technical solutions, and by calculating the motion sequence of key points based on the time-series data of key points on the fish body and generating a micro-motion feature sequence, it is possible to quantify changes in micro-movements such as sudden stops, accelerations, and turns of the fish body, thereby improving the ability to recognize feeding intentions. By performing dense optical flow calculations on adjacent video frames to obtain a dense optical flow field and generating a motion disorder entropy sequence, it is possible to quantify the consistency and disorder of the fish group's movement direction, thereby reducing the risk of misjudgment caused by relying solely on pixel changes. By fusing the micro-motion feature sequence and the motion disorder entropy sequence to obtain a feeding desire feature sequence, it is possible to comprehensively express both individual and group movement characteristics, thereby improving the differentiation between normal gathering and cruising and competing for food. By segmenting the feeding desire feature sequence based on a preset time sliding window and performing time-series analysis to generate a feeding desire index, it is possible to depict the continuous evolution trend of the feeding state, thereby reducing the judgment lag caused by single-frame fluctuations and improving the real-time performance of feeding control.

[0013] In one example, the present invention can be further configured as follows: calculating the key point motion sequence based on the time-series data of the fish body key points, and generating a micro-motion feature sequence based on the key point motion sequence, includes: Based on the time-series data of the fish's key points, the displacement changes of the head key points, dorsal fin key points, and tail key points are calculated to obtain the key point displacement sequence. Calculate the velocity change of key points based on the key point displacement sequence, and generate a key point velocity sequence. The instantaneous acceleration sequence is calculated based on the velocity sequence of the key points, and the high-frequency steering sequence is calculated based on the relative directional changes of the head key points and the tail key points. The instantaneous acceleration sequence and the high-frequency steering sequence are combined to obtain the microscopic motion feature sequence.

[0014] By employing the above technical solutions, a key point displacement sequence is obtained by calculating the displacement changes of key points on the head, dorsal fin, and tail. This sequence accurately characterizes the fish's movement amplitude and posture swaying within a short period, providing fundamental data for motion intensity analysis. A key point velocity sequence is generated by calculating the key point velocity changes based on the key point displacement sequence, reflecting the changing trends of the fish's movement state over continuous time, thus improving the ability to depict changes in feeding activity. Instantaneous acceleration sequences and high-frequency turning sequences are calculated based on the key point velocity sequences, enabling the identification of explosive acceleration and frequent turning behaviors during feeding competition, thereby enhancing the ability to discern the fish's feeding intentions. Finally, a micro-motion feature sequence is obtained by combining the instantaneous acceleration sequence and the high-frequency turning sequence, forming a more discriminative expression of micro-movement features, thereby improving the precision of feeding desire assessment.

[0015] In one example, the present invention can be further configured as follows: performing dense optical flow calculations on adjacent video frames in the preprocessed feed area video stream to obtain a dense optical flow field, and generating a motion disorder entropy sequence based on the dense optical flow field, including: The motion vectors in the dense optical flow field are statistically analyzed to generate a motion direction histogram, and the direction distribution probability is determined based on the motion direction histogram. The information entropy value is calculated based on the directional distribution probability to obtain the motion disorder entropy corresponding to the adjacent video frame; The motion disorder entropy corresponding to consecutive video frames is arranged in a temporal sequence to obtain the motion disorder entropy sequence.

[0016] By employing the above technical solutions, and generating motion direction histograms and determining direction distribution probabilities through directional statistics of motion vectors in a dense optical flow field, the motion direction information of a group can be transformed into a quantifiable probability distribution representation, thereby improving the interpretability of group motion state analysis. By calculating the information entropy value based on the direction distribution probability, motion disorder entropy can be obtained, quantifying the degree of disorder in the fish's motion direction, thus effectively distinguishing between a consistent cruising state and a disordered competition state. Furthermore, by temporally arranging the motion disorder entropy corresponding to consecutive video frames, a motion disorder entropy sequence can be obtained, reflecting the changing pattern of group disorder over time, thus providing a stable input for feeding desire feature fusion and temporal analysis.

[0017] In one example, the present invention can be further configured as follows: generating a corresponding feeding control command based on the feeding desire index and the water quality environment data, and executing feeding control on the feeder according to the feeding control command, includes: The feeding conditions are determined based on the dissolved oxygen in the water quality environmental data. If the determination result does not meet the feeding conditions, a pause feeding instruction is generated. If the determination result meets the feeding conditions, a control quantity for the feeding motor speed is generated based on the feeding desire index to obtain the feeding control command; The feeding machine is activated by the feeding control command, and the speed of the feeding motor is dynamically adjusted. If the feeding desire index meets the conditions for stopping feeding, a stop feeding command is generated, and the current round of feeding control ends based on the stop feeding command.

[0018] By adopting the above technical solutions, the feeding conditions are determined based on dissolved oxygen in the water quality data, and a pause feeding command is generated. This can suppress feeding behavior under unfavorable water quality conditions, thereby reducing the risk of hypoxia and water quality deterioration. By generating a control quantity for the feeding motor speed based on the feeding desire index, the feeding control command can be obtained, ensuring that the feeding intensity matches the feeding needs of the fish, thus reducing feed waste caused by overfeeding. By executing the feeding induction start control and dynamically adjusting the feeding motor speed, a closed-loop adjustment can be made in real time to follow changes in feeding desire during the feeding process, thereby improving the response speed and feeding accuracy of the feeding process. By generating a stop feeding command and ending the current round of feeding control, feeding can be stopped in time when the feeding desire declines, thus avoiding uneaten feed sinking to the bottom and water pollution caused by delayed feeding.

[0019] In a second aspect, the present invention provides a feeding assessment and evaluation system based on posture analysis, the system comprising: The data acquisition module is used to acquire the preprocessed video stream of the feeding area and the water quality environment data corresponding to the video stream of the feeding area; The attitude estimation module is used to perform attitude estimation on the preprocessed video stream of the feeding area to obtain the time series data of key points of the fish. The desire assessment module is used to extract the micro-motion features of the fish's behavior and posture based on the time-series data of the key points of the fish, and generate a dense optical flow field based on the preprocessed video stream of the feeding area to calculate the motion disorder entropy. The micro-motion features and the motion disorder entropy are fused to obtain the feeding desire feature sequence, and then the feeding desire feature sequence is subjected to time-series analysis to obtain the feeding desire index. The feeding control module is used to generate corresponding feeding control instructions based on the feeding desire index and the water quality environment data, and to execute feeding control on the feeder according to the feeding control instructions.

[0020] By adopting the above technical solutions, and by acquiring the pre-processed video stream of the feeding area and the corresponding water quality environmental data, synchronous visual and environmental information can be provided for feeding desire assessment, thereby improving the adaptability of feeding decisions to changes in the aquaculture environment. By performing posture estimation on the pre-processed video stream of the feeding area to obtain the time-series data of key points of the fish body, the fish behavior changes can be stably depicted from the structured posture information of the fish body, thereby reducing the influence of water surface reflection and background disturbance on feeding judgment. By extracting micro-motion features and generating a dense optical flow field to calculate the motion disorder entropy and fuse them to obtain the feeding desire index, the difference in feeding state between individual action intensity and group motion disorder can be reflected simultaneously, thereby improving the accuracy of distinguishing between normal cruising and competing for feeding. By generating feeding control commands based on the feeding desire index and water quality environmental data and executing feeding control, the feeding intensity can be dynamically adjusted according to changes in feeding desire, thereby reducing the risk of feed waste and water quality deterioration caused by overfeeding.

[0021] In a third aspect, the present invention provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the feeding assessment and feeding method based on posture analysis.

[0022] In a fourth aspect, the present invention provides a storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of the feeding assessment and feeding method based on posture analysis. Attached Figure Description

[0023] The accompanying drawings, which form part of this specification, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an undue limitation of the invention. In the drawings: Figure 1 This is a flowchart of a feeding assessment and feeding method based on posture analysis in an embodiment of the present invention; Figure 2 This is a structural block diagram of the feeding assessment and evaluation system based on posture analysis according to an embodiment of the present invention; Figure 3 This is a structural block diagram of an electronic device according to an embodiment of the present invention. Detailed Implementation

[0024] The present invention will now be described in detail with reference to the accompanying drawings and embodiments. It should be noted that, unless otherwise specified, the embodiments and features described herein can be combined with each other.

[0025] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this invention is for describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0026] Example 1 like Figure 1 As shown, this invention discloses a feeding assessment method based on posture analysis, which specifically includes the following steps: S10: Acquire the pre-processed video stream of the feeding area and the corresponding water quality environmental data.

[0027] Specifically, the pre-processed video stream of the feeding area is acquired while maintaining the temporal integrity of the continuous frame sequence. At the same time, water quality environmental data such as dissolved oxygen and water temperature are acquired and the acquisition time is added to the water quality environmental data. Then, the water quality environmental data and the video stream of the feeding area are established to form an input dataset at the same acquisition time. This enables the subsequent feeding desire assessment to utilize both visual and environmental information and adapt to the real-time feeding constraints of complex aquaculture scenarios such as net cages and ponds.

[0028] S20: Perform attitude estimation on the preprocessed video stream of the feeding area to obtain the time series data of key points on the fish body.

[0029] Specifically, the preprocessed video stream of the feeding area is subjected to frame-by-frame pose estimation and the position results of the fish key points in each video frame are output. Then, the key point position results in adjacent video frames are correlated in time order to form key point trajectories. Based on the key point trajectories, fish key point time series data that can characterize the fish behavior and posture changes over time are generated.

[0030] S30: Extract the micro-motion features of fish behavior and posture based on the time series data of key points on the fish body, and generate a dense optical flow field based on the preprocessed video stream of the feeding area to calculate the motion disorder entropy. The micro-motion features and motion disorder entropy are fused to obtain the feeding desire feature sequence, and then the feeding desire feature sequence is subjected to time series analysis to obtain the feeding desire index.

[0031] Specifically, micro-motion features that describe the evolution of micro-movements of fish are formed based on the time-series data of key points on the fish body, and motion disorder entropy that describes the movement state of the fish group is formed based on the pre-processed video stream of the feeding area. The micro-motion features and motion disorder entropy are then fused to obtain the feeding desire feature sequence, and the feeding desire feature sequence is subjected to time-series analysis to output the feeding desire index. This makes the evaluation of feeding desire no longer limited to static or coarse-grained indicators such as pixel duty cycle or activity in adjacent frames, but can distinguish the difference between normal agglomeration and cruising and competing for food.

[0032] S40: Generates corresponding feeding control instructions based on the feeding desire index and water quality data, and executes feeding control on the feeder according to the feeding control instructions.

[0033] Specifically, the feeding desire index and water quality data are used as joint inputs for feeding control to generate feeding control instructions. These instructions are used to instruct the feeding machine to start and stop and adjust the feeding intensity, so that the feeding process is synchronized with the changes in the fish's feeding desire and the risk of water quality deterioration caused by feed sinking and rotting at the bottom is reduced.

[0034] In one embodiment, step S10, namely acquiring the preprocessed video stream of the feeding area and the corresponding water quality environmental data, includes: S11: Acquire video images of the feeding area and generate a video stream of the feeding area. Perform image enhancement processing on the video stream of the feeding area to obtain enhanced video frames.

[0035] Specifically, video images of the feeding area covering the feeding point and its surrounding water surface are collected and formed into a feeding area video stream at a high frame rate. Brightness equalization and contrast enhancement are performed on each video frame in the feeding area video stream to alleviate the loss of details caused by insufficient lighting in rainy weather. At the same time, local overexposure areas caused by water surface reflection are suppressed and the distinguishability of fish body edges and key point textures is improved. In the enhancement process, the feeding point area is given higher weight to reduce the interference of non-core feeding area patrolling fish and floating objects on subsequent analysis.

[0036] S12: Perform denoising processing on the enhanced video frame to obtain a preprocessed video frame, and generate a preprocessed feed area video stream based on the preprocessed video frame.

[0037] Specifically, spatial denoising and temporal smoothing are performed on the enhanced video frames to reduce particle noise caused by water turbidity and texture jitter caused by wind and waves, while maintaining the continuity and stability of the fish outline structure between adjacent video frames. The denoised video frames are then reassembled in chronological order to obtain a preprocessed video frame sequence. Based on the preprocessed video frame sequence, a preprocessed feeding area video stream is generated for attitude estimation and dense optical flow calculation.

[0038] S13: Collect dissolved oxygen and water temperature data in the feeding area to generate water quality environmental data, and add a collection time marker to the water quality environmental data.

[0039] Specifically, dissolved oxygen and water temperature measurements in the feeding area are collected and packaged into water quality environmental data frames. At the same time, a collection time marker is added to the water quality environmental data frames to represent the time position of the water quality status. The continuously collected water quality environmental data frames are then arranged in chronological order to obtain a water quality environmental data sequence for subsequent feeding condition determination and feeding control.

[0040] S14: Associate the water quality environmental data with the pre-processed feeding area video stream according to the acquisition time to obtain the water quality environmental data corresponding to the pre-processed feeding area video stream.

[0041] Specifically, the acquisition time information of each video frame in the preprocessed feeding area video stream is read and used as the correlation benchmark. The water quality environment data frame with the same or closest acquisition time is retrieved in the water quality environment data sequence and a mapping relationship is established. Then, the mapped water quality environment data frame is bound to the corresponding video frame to obtain the water quality environment data corresponding to the preprocessed feeding area video stream, thereby ensuring that the feeding desire assessment and feeding control are based on the real environmental state at the same time.

[0042] In one embodiment, step S20, namely, performing pose estimation on the preprocessed feeding area video stream to obtain temporal data of key points on the fish body, includes: S21: Locate the fish target in the preprocessed feeding area video stream to obtain the fish target region in each video frame.

[0043] Specifically, the fish target is located and its spatial range is determined in each video frame of the preprocessed feeding area video stream. The area containing the fish outline and key structures is cropped into the fish target area, and the fish target area is continuously tracked frame by frame to maintain the consistency of the same fish on the time axis. At the same time, the fish target area is stably updated under the conditions of fish occlusion, cross swimming and water surface fluctuation to reduce the impact of mislocation on the temporal association of subsequent key points.

[0044] S22: Perform pose estimation processing on the target area of ​​the fish body to obtain the head key points, dorsal fin key points and tail key points.

[0045] Specifically, pose estimation processing is performed within the target area of ​​the fish body, and the position results of the head key points, dorsal fin key points and tail key points corresponding to the fish body structure are output. The position results are used to characterize the orientation of the fish body and the morphology of the spine line and to provide a geometric basis for the calculation of micro-motion features. At the same time, the position results of each key point are bound with the temporal information of the corresponding video frame to form a set of key points that can be used for temporal correlation.

[0046] S23: Perform temporal correlation on the head key points, dorsal fin key points and tail key points in adjacent video frames to obtain key point trajectories, and then generate fish body key point temporal data based on the key point trajectories.

[0047] Specifically, the head key points, dorsal fin key points and tail key points of the same fish in adjacent video frames are matched and cross-fish body misassociations are eliminated. The matched key point positions are connected in chronological order to form key point trajectories. Then, based on the key point trajectories, fish body key point time series data containing key point coordinate sequences and time sequences are generated, so that subsequent calculations can characterize micro-movement patterns such as sudden stops, bursts of acceleration and frequent turns within a continuous time window.

[0048] In one embodiment, in step S30, micro-motion features of fish behavior and posture are extracted based on the time-series data of key points on the fish body, and a dense optical flow field is generated based on the preprocessed video stream of the feeding area to calculate the motion disorder entropy. The micro-motion features and motion disorder entropy are fused to obtain a feeding desire feature sequence, and then a time-series analysis is performed on the feeding desire feature sequence to obtain a feeding desire index, including: S31: Calculate the motion sequence of key points based on the time series data of key points on the fish body, and generate a micro motion feature sequence based on the motion sequence of key points.

[0049] Specifically, based on the time-series data of key points on the fish body, the displacement and direction changes of key points at adjacent sampling times are calculated to form a key point motion sequence. The second-order changes and direction changes of the key point motion sequence are quantified within a continuous time window to generate a micro motion feature sequence. The micro motion feature sequence is used to characterize the intensity changes of frequent sudden stops and explosive accelerations during feeding competition, as well as the turning changes of frequent angle adjustments when competing for feed particles.

[0050] S32: Perform dense optical flow calculation on adjacent video frames in the preprocessed feed area video stream to obtain the dense optical flow field, and generate a motion disorder entropy sequence based on the dense optical flow field.

[0051] Specifically, pixel-by-pixel motion estimation is performed on adjacent video frames in the preprocessed feeding area video stream to obtain a dense optical flow field. Then, the motion vectors in the dense optical flow field are mapped to a motion direction histogram to form a direction probability distribution. Based on the direction probability distribution, motion disorder entropy is calculated to quantify the anisotropy of the fish's movement direction. The motion disorder entropy corresponding to consecutive video frames is arranged in chronological order to generate a motion disorder entropy sequence. The motion disorder entropy sequence is used to distinguish between a consistent cruising state and a highly disordered, intense competition state.

[0052] S33: The micro-motion feature sequence and the motion disorder entropy sequence are fused to obtain the feeding desire feature sequence.

[0053] Specifically, the micro-motion feature sequence and the motion disorder entropy sequence are time-aligned and the features of different dimensions are normalized. Then, a combination of feature splicing and weighted fusion is used to form a fused feature vector, which is then arranged in chronological order to obtain the feeding desire feature sequence. This allows the feeding desire feature sequence to simultaneously contain information on the intensity of individual micro-movements and the macro-disorder of the group, thereby reducing false activity interference caused by wind, waves, splashes, and floating objects.

[0054] S34: The food desire feature sequence is segmented based on a preset time sliding window to obtain a windowed feature sequence, and time series analysis is performed on the windowed feature sequence to generate a food desire index.

[0055] Specifically, a time sliding window is constructed and its window duration is set to a preset range of seconds to cover the time span from triggering to the evolution of the feeding action. The feeding desire feature sequence is continuously segmented according to the sliding window to obtain a windowed feature sequence. The windowed feature sequence is then input into a time-series deep learning model to learn the evolution trend of the micro-acceleration feature sequence and the macro-entropy value sequence and output the feeding desire index. The feeding desire index is denoted as FDI and limited to a scalar value within a continuous interval to achieve real-time output and continuous updates.

[0056] In one embodiment, step S31, namely calculating the motion sequence of key points based on the time-series data of key points on the fish body, and generating a micro-motion feature sequence based on the motion sequence of key points, includes: S311: Calculate the displacement changes of the head key point, dorsal fin key point and tail key point based on the time series data of key points of the fish body, and obtain the key point displacement sequence.

[0057] Specifically, the coordinates of the head key point, dorsal fin key point, and tail key point in the time series data of fish body key points are read at adjacent sampling times, and the displacement change Δp is calculated. t =p t -p t-1 , where p t Let Δp represent the spatial coordinates of the keypoint corresponding to frame t. t The displacement vector between adjacent frames is represented by the displacement change of consecutive frames arranged in chronological order to obtain the key point displacement sequence, which reflects the movement and swaying amplitude of the fish body in a short period of time.

[0058] S312: Calculate the velocity change of key points based on the displacement sequence of key points, and generate the velocity sequence of key points.

[0059] Specifically, the change in velocity at key points is calculated based on the key point displacement sequence and the interval Δt between adjacent sampling times. , where v tLet t represent the velocity vector of the key point corresponding to the t-th frame, and Δt represent the time interval between adjacent frames. Then, the velocity changes of each key point in consecutive frames are arranged in chronological order to generate a key point velocity sequence for further calculation of instantaneous acceleration and steering characteristics.

[0060] S313: Calculate the instantaneous acceleration sequence based on the velocity sequence of key points, and calculate the high-frequency steering sequence based on the relative directional changes of the head key points and the tail key points.

[0061] Specifically, the velocity sequence at key points is subjected to second-order difference to calculate the instantaneous acceleration sequence. , where a t This represents the instantaneous acceleration vector corresponding to frame t and is used to characterize sudden stops and explosive acceleration features. Simultaneously, the fish's orientation is defined by connecting the head and tail keypoints, and the orientation angle θ is calculated. t The rate of change is used to obtain the angular velocity. Then, the curvature κ of the spinal line is calculated by combining the offset of the key points of the dorsal fin relative to the connecting line. t And take the rate of change of curvature Δκ t =κ t -κ{ t-1 This leads to the formation of high-frequency turning sequences to characterize the frequent angle adjustment features during feeding competition.

[0062] S314: Combine the instantaneous acceleration sequence with the high-frequency steering sequence to obtain the microscopic motion characteristic sequence.

[0063] Specifically, amplitude normalization and time alignment are performed on the instantaneous acceleration sequence and the high-frequency steering sequence, respectively. Then, the normalized instantaneous acceleration features, angular velocity features, and curvature change features are combined to form a microscopic feature vector f. t = [||at||, ||t||, ||Δκt||] and arrange them in chronological order to obtain the micro-motion feature sequence, so that the micro-motion feature sequence can simultaneously express the changes in motion intensity and the changes in turning frequency and be used for subsequent feeding desire feature fusion.

[0064] In one embodiment, step S32 involves performing dense optical flow calculations on adjacent video frames in the preprocessed feed area video stream to obtain a dense optical flow field, and generating a motion disorder entropy sequence based on the dense optical flow field, including: S321: Perform direction statistics on the motion vectors in the dense optical flow field, generate a motion direction histogram, and determine the direction distribution probability based on the motion direction histogram.

[0065] Specifically, the dense optical flow field is represented as a motion vector field F(x,y) = (u(x,y), v(x,y)) on the set of pixels, where u(x,y) and v(x,y) represent the motion components of the pixel (x,y) in the horizontal and vertical directions, respectively. Then, the direction angle φ(x,y) of the motion vector of each pixel is calculated and divided into K direction buckets according to a preset direction interval. The number of vectors in each direction bucket is counted to obtain the motion direction histogram h. k Then normalize it to obtain the directional distribution probability. , where p k The probability value of the k-th directional interval is used as the probability input for calculating the information entropy.

[0066] S322: Calculate the information entropy value based on the directional distribution probability to obtain the motion disorder entropy corresponding to the adjacent video frames.

[0067] Specifically, information entropy is calculated based on the probability distribution of directions. The information entropy H is used as the motion disorder entropy corresponding to adjacent video frames, where K represents the number of directional intervals, and p k The probability of the k-th direction interval is represented by the motion disorder entropy, which is used to quantify the consistency and disorder of the fish's movement direction and to reflect the difference between the lower entropy value in the cruising state and the higher entropy value in the fighting state.

[0068] S323: Arrange the motion disorder entropy corresponding to consecutive video frames in a temporal sequence to obtain the motion disorder entropy sequence.

[0069] Specifically, the motion disorder entropy calculated from consecutive adjacent video frames is arranged in chronological order to form a motion disorder entropy sequence E = {H1, H2, ..., H...}. t}, where H t Let represent the motion disorder entropy corresponding to the t-th time position. The motion disorder entropy sequence is smoothed to reduce the entropy value jitter caused by short-term optical flow noise, so as to obtain a motion disorder entropy sequence that can stably reflect the change of group disorder over time.

[0070] In one embodiment, step S40, namely generating a corresponding feeding control command based on the feeding desire index and water quality environmental data, and executing feeding control on the feeder according to the feeding control command, includes: S41: Determine feeding conditions based on dissolved oxygen in water quality environmental data, and generate a pause feeding instruction if the determination result does not meet the feeding conditions.

[0071] Specifically, the dissolved oxygen measurement value in the water quality environmental data is read and compared with the safety threshold to obtain the feeding condition judgment result. When the dissolved oxygen is lower than the safety threshold, regardless of the feeding desire index output by the visual analysis, a pause feeding instruction is generated and the feeding desire index is forcibly set to FDI=0. At the same time, the pause feeding instruction is sent to the feeder to execute the stop feeding control, thereby avoiding the risk of continued feeding and aggravation of water body load under the risk of hypoxia such as low dissolved oxygen on cloudy or rainy days.

[0072] S42: If the judgment result meets the feeding conditions, the control quantity of the feeding motor speed is generated based on the feeding desire index to obtain the feeding control command.

[0073] Specifically, when the feeding conditions are met, the feeding desire index is mapped to the feeding motor speed control quantity and a feeding control command is generated. When the feeding desire index is at an initial low level, a lower speed control quantity is generated to induce feeding and conduct exploratory feeding. When the feeding desire index rises rapidly and exceeds a high threshold, a full-load speed control quantity is generated to respond to the feed demand during the intense competition period. When the feeding desire index enters the maintenance and decay stage, a speed control quantity that dynamically adjusts with the index change is generated to achieve on-demand feeding and precise stopping of feeding.

[0074] S43: Perform feed-inducing start control on the feeder according to the feeding control command, and perform dynamic adjustment control on the speed of the feeding motor.

[0075] Specifically, the feeding machine is driven to start feeding according to the feeding control command and to throw feed tentatively at a low speed. During the feeding process, the machine continuously receives updated feeding control commands and dynamically adjusts the speed of the feeding motor in combination with the changing trend of the feeding desire index. When the feeding desire index rises explosively, the speed is rapidly increased to ensure the supply of pellets. When the feeding desire index gradually falls, the speed is reduced simultaneously to reduce overfeeding and prevent feed from sinking and rotting, which would cause an increase in ammonia nitrogen.

[0076] S44: Generate a stop feeding command when the appetite index meets the stop feeding conditions, and end the current round of feeding control based on the stop feeding command.

[0077] Specifically, when the feeding desire index is below the cutoff threshold for a continuous period of time and remains below the preset duration, a stop feeding command is generated and sent to the feeder to shut down the feeding output. Then, based on the stop feeding command, the current round of feeding control ends and the feeding desire index curve characteristics and feeding duration of this round of feeding are recorded. In subsequent feedings, the mapping coefficient between the feeding desire index and the rotation speed control is dynamically fine-tuned based on the historical curve to adapt to the differences in feeding patterns caused by changes in the fish's growth cycle and environment.

[0078] Furthermore, based on the characteristics of the feeding desire index curve recorded during this feeding cycle, the induction intensity and speed adjustment response of the next feeding cycle are calibrated at the parametric level, while maintaining the consistency and interpretability of the feeding desire index output at different breeding stages.

[0079] Example 2 like Figure 2 As shown, based on the same inventive concept as the above embodiments, the present invention also provides a feeding assessment and feeding system based on posture analysis, comprising: The data acquisition module is used to acquire the pre-processed video stream of the feeding area and the corresponding water quality environmental data. The attitude estimation module is used to estimate the attitude of the preprocessed video stream of the feeding area to obtain the time-series data of key points of the fish. The desire assessment module is used to extract the micro-motion features of fish behavior and posture based on the time series data of key points on the fish body, and generate a dense optical flow field based on the preprocessed video stream of the feeding area to calculate the motion disorder entropy. The micro-motion features and motion disorder entropy are fused to obtain the feeding desire feature sequence, and then the feeding desire feature sequence is analyzed in time series to obtain the feeding desire index. The feeding control module is used to generate corresponding feeding control instructions based on the feeding desire index and water quality environment data, and to execute feeding control on the feeder according to the feeding control instructions.

[0080] Optionally, the data acquisition module includes: The video acquisition submodule is used to acquire video images of the feeding area and generate a video stream of the feeding area, and to perform image enhancement processing on the video stream of the feeding area to obtain enhanced video frames. The denoising preprocessing submodule is used to perform denoising processing on the enhanced video frames to obtain preprocessed video frames, and to generate a preprocessed feed area video stream based on the preprocessed video frames; The water quality acquisition submodule is used to collect dissolved oxygen and water temperature data in the feeding area to generate water quality environmental data, and to add a collection time identifier to the water quality environmental data. The time-association submodule is used to associate water quality environmental data with the pre-processed feeding area video stream according to the acquisition time, so as to obtain water quality environmental data corresponding to the pre-processed feeding area video stream.

[0081] Optionally, the attitude estimation module includes: The target localization submodule is used to locate the fish target in the preprocessed feeding area video stream and obtain the fish target area in each video frame; The key point extraction submodule is used to perform pose estimation processing on the target area of ​​the fish body to obtain head key points, dorsal fin key points and tail key points; The trajectory generation submodule is used to perform temporal correlation of head key points, dorsal fin key points and tail key points in adjacent video frames to obtain key point trajectories, and then generate fish body key point temporal data based on the key point trajectories.

[0082] Optional, the desire assessment module includes: The micro-motion extraction submodule is used to calculate the motion sequence of key points based on the time series data of key points on the fish body, and generate a micro-motion feature sequence based on the motion sequence of key points. The optical flow entropy operator module is used to perform dense optical flow calculations on adjacent video frames in the preprocessed feed area video stream to obtain a dense optical flow field, and generate a motion disorder entropy sequence based on the dense optical flow field. The feature fusion submodule is used to fuse the micro-motion feature sequence and the motion disorder entropy sequence to obtain the feeding desire feature sequence; The time series analysis submodule is used to segment the food desire feature sequence based on a preset time sliding window to obtain a windowed feature sequence, and to perform time series analysis on the windowed feature sequence to generate a food desire index.

[0083] Optional, the micro-motion extraction submodule includes: The displacement calculation unit is used to calculate the displacement changes of the head key point, dorsal fin key point and tail key point based on the time series data of fish key points, and obtain the key point displacement sequence. The velocity calculation unit is used to calculate the velocity change of key points based on the key point displacement sequence and generate the key point velocity sequence. The acceleration extraction unit is used to calculate the instantaneous acceleration sequence based on the velocity sequence of key points, and to calculate the high-frequency steering sequence based on the relative directional changes of the head key points and the tail key points; The micro-motion combination unit is used to combine instantaneous acceleration sequences and high-frequency steering sequences to obtain micro-motion characteristic sequences.

[0084] Optionally, the optical flow entropy operator module includes: The orientation statistics unit is used to perform orientation statistics on motion vectors in a dense optical flow field, generate a motion orientation histogram, and determine the orientation distribution probability based on the motion orientation histogram. The entropy calculation unit is used to calculate the information entropy value based on the direction distribution probability to obtain the motion disorder entropy corresponding to the adjacent video frames; The entropy ordering unit is used to temporally arrange the motion disorder entropy corresponding to consecutive video frames to obtain a motion disorder entropy sequence.

[0085] Optionally, the feeding control module includes: The condition determination submodule is used to determine the feeding conditions based on dissolved oxygen in the water quality environmental data, and to generate a pause feeding command if the determination result does not meet the feeding conditions. The instruction generation submodule is used to generate a control quantity for the feeding motor speed based on the feeding desire index when the judgment result meets the feeding conditions, and thus obtain the feeding control instruction. The dynamic adjustment submodule is used to perform feeding start control on the feeder according to the feeding control command, and to perform dynamic adjustment control on the speed of the feeding motor; The termination control submodule is used to generate a stop feeding command when the appetite index meets the stop feeding conditions, and to end the current round of feeding control based on the stop feeding command.

[0086] Example 3 like Figure 3 As shown, the present invention also provides an electronic device 100 for implementing a feeding assessment and feeding method based on posture analysis; The electronic device 100 includes a memory 101, at least one processor 102, a computer program 103 stored in the memory 101 and executable on at least one processor 102, and at least one communication bus 104.

[0087] The memory 101 can be used to store the computer program 103. The processor 102 implements the steps of the feeding assessment method based on posture analysis in Embodiment 1 by running or executing the computer program stored in the memory 101 and calling the data stored in the memory 101.

[0088] The memory 101 may primarily include a program storage area and a data storage area. The program storage area may store the operating system, application programs required for at least one function (such as sound playback function, image playback function, etc.), etc.; the data storage area may store data created based on the use of the electronic device 100 (such as audio data), etc. In addition, the memory 101 may include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0089] At least one processor 102 may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Processor 102 may be a microprocessor or any conventional processor. Processor 102 is the control center of electronic device 100, connecting various parts of electronic device 100 via various interfaces and lines.

[0090] The memory 101 in the electronic device 100 stores multiple instructions to implement a feeding assessment method based on posture analysis, and the processor 102 can execute multiple instructions to achieve the following: Acquire the pre-processed video stream of the feeding area and the corresponding water quality environmental data; Pose estimation was performed on the preprocessed video stream of the feeding area to obtain temporal data of key points on the fish body; Microscopic motion features of fish behavior and posture are extracted from the time series data of key points on the fish body. A dense optical flow field is generated based on the preprocessed video stream of the feeding area to calculate the motion disorder entropy. The microscopic motion features and motion disorder entropy are fused to obtain the feeding desire feature sequence. Then, the feeding desire feature sequence is analyzed in time series to obtain the feeding desire index. Feeding control instructions are generated based on the feeding desire index and water quality data, and the feeding machine is controlled according to the feeding control instructions.

[0091] Example 4 If the modules / units integrated in the electronic device 100 are implemented as software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or system capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, and read-only memory (ROM).

[0092] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied 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.

[0093] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A system that specifies functions in one or more boxes.

[0094] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including an instruction set implemented in a process. Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0095] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0096] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0097] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the scope of protection of the claims of the present invention.

Claims

1. A method of feeding based on a posture analysis of an eating assessment, characterized in that, The method comprises: acquiring a pre-processed feeding area video stream and water quality environment data corresponding to the feeding area video stream; performing pose estimation on the pre-processed feeding area video stream to obtain fish key point time series data; extracting micro-motion features of fish behavior postures according to the fish key point time series data, generating a dense optical flow field based on the pre-processed feeding area video stream to calculate motion chaos entropy, fusing the micro-motion features and the motion chaos entropy to obtain a feeding desire feature sequence, and then performing time series analysis on the feeding desire feature sequence to obtain a feeding desire index; generating a corresponding feeding control instruction according to the feeding desire index and the water quality environment data, and performing feeding control on a feeding machine according to the feeding control instruction.

2. The gesture analysis based assessment of intake feeding method according to claim 1, wherein, The acquisition of the pre-processed feeding area video stream and the water quality environment data corresponding to the feeding area video stream comprises: collecting feeding area video images and generating a feeding area video stream, performing image enhancement processing on the feeding area video stream to obtain enhanced video frames; performing denoising processing on the enhanced video frames to obtain pre-processed video frames, and generating a pre-processed feeding area video stream based on the pre-processed video frames; collecting dissolved oxygen and water temperature of the feeding area to generate the water quality environment data, and attaching a collection time point identifier to the water quality environment data; associating the water quality environment data with the pre-processed feeding area video stream according to the collection time point to obtain the water quality environment data corresponding to the pre-processed feeding area video stream.

3. The gesture analysis based assessment of intake feeding method according to claim 1, wherein, The pose estimation on the pre-processed feeding area video stream to obtain fish key point time series data comprises: locating a fish target in the pre-processed feeding area video stream to obtain a fish target region in each video frame; performing pose estimation processing on the fish target region to obtain head key points, dorsal fin key points and tail key points; performing time series association on the head key points, the dorsal fin key points and the tail key points in adjacent video frames to obtain key point trajectories, and then generating the fish key point time series data based on the key point trajectories.

4. The gesture analysis based assessment of intake feeding method according to claim 1, wherein, The extraction of micro-motion features of fish behavior postures according to the fish key point time series data, the generation of a dense optical flow field based on the pre-processed feeding area video stream to calculate motion chaos entropy, the fusion of the micro-motion features and the motion chaos entropy to obtain a feeding desire feature sequence, and the time series analysis on the feeding desire feature sequence to obtain a feeding desire index comprise: calculating a key point motion sequence based on the fish key point time series data, and generating a micro-motion feature sequence according to the key point motion sequence; performing dense optical flow calculation on adjacent video frames in the pre-processed feeding area video stream to obtain a dense optical flow field, and generating a motion chaos entropy sequence based on the dense optical flow field; fusing the micro-motion feature sequence and the motion chaos entropy sequence to obtain a feeding desire feature sequence; The feeding desire feature sequence is segmented based on a preset time sliding window to obtain a windowed feature sequence, and the windowed feature sequence is subjected to time series analysis to generate the feeding desire index.

5. The gesture analysis based assessment of intake feeding method according to claim 4, wherein, The key point motion sequence is calculated based on the fish key point time series data, and a micro motion feature sequence is generated according to the key point motion sequence, including: The displacement change amount of the head key point, the dorsal fin key point and the tail key point is calculated based on the fish key point time series data to obtain a key point displacement sequence; The key point velocity change amount is calculated according to the key point displacement sequence to generate a key point velocity sequence; The instantaneous acceleration sequence is calculated according to the key point velocity sequence, and a high-frequency turning sequence is calculated based on the relative direction change of the head key point and the tail key point; The instantaneous acceleration sequence and the high-frequency turning sequence are combined to obtain the micro motion feature sequence.

6. The gesture analysis based assessment of intake feeding method according to claim 4, wherein, The dense optical flow field is generated based on the dense optical flow field, including: The direction distribution probability is determined based on the motion direction histogram generated by direction statistics of the motion vector in the dense optical flow field; The information entropy value is calculated according to the direction distribution probability to obtain the motion confusion entropy corresponding to the adjacent video frames; The motion confusion entropy corresponding to the continuous video frames is arranged in time series to obtain the motion confusion entropy sequence.

7. The gesture analysis based assessment of intake feeding method according to claim 1, wherein, The corresponding feeding control instruction is generated according to the feeding desire index and the water quality environment data, and the feeding control instruction is used to control the feeding machine, including: The feeding condition is determined based on the dissolved oxygen in the water quality environment data, and a feeding pause instruction is generated if the determination result does not meet the feeding condition; If the determination result meets the feeding condition, the control amount of the feeding motor speed is generated based on the feeding desire index to obtain the feeding control instruction; The feeding control instruction is used to control the feeding machine to start the feeding control, and the dynamic adjustment control is performed on the feeding motor speed; If the feeding desire index meets the stop feeding condition, a stop feeding instruction is generated, and the current feeding control is ended based on the stop feeding instruction.

8. A feeding assessment and feeding system based on gesture analysis, characterized in that, The system includes: A data acquisition module is configured to acquire a preprocessed feeding area video stream and water quality environment data corresponding to the feeding area video stream; An attitude estimation module is configured to perform attitude estimation on the preprocessed feeding area video stream to obtain fish key point time series data; A desire evaluation module is configured to extract micro motion features of fish behavior postures based on the fish key point time series data, generate a dense optical flow field based on the preprocessed feeding area video stream to calculate a motion confusion entropy, fuse the micro motion features and the motion confusion entropy to obtain a feeding desire feature sequence, and perform time series analysis on the feeding desire feature sequence to obtain a feeding desire index. A feeding control module is configured to generate a corresponding feeding control instruction according to the feeding desire index and the water quality environment data, and to perform feeding control on a feeding machine according to the feeding control instruction.

9. An electronic device, comprising: The computer readable storage medium stores at least one instruction, which is executed by the processor to implement the steps of the feeding method based on the posture analysis for the feeding desire evaluation according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores at least one instruction, which is executed by the processor to implement the steps of the feeding method based on the posture analysis for the feeding desire evaluation according to any one of claims 1 to 7.

Images