Intelligent breeding behavior analysis system and method based on edge calculation and AI vision

Abstract

The invention provides an intelligent breeding behavior analysis system and method based on edge calculation and AI vision, relates to the technical field of intelligent breeding and computer vision, and solves the technical problems of low efficiency, poor behavior recognition accuracy and abnormal early warning lag caused by dependence on manual and cloud processing in existing breeding management. The system comprises an edge end device which is used for collecting a video stream in real time and carrying out target detection and key point identification on the video stream through a lightweight deep learning model to obtain a preliminary behavior classification result; the fog end server is used for gathering a plurality of preliminary behavior classification results, performing complex behavior recognition, behavior time sequence analysis and abnormal behavior detection, obtaining an abnormal behavior result and triggering real-time early warning; and the cloud platform is used for receiving and storing all data of the farm, performing long-term data mining, multi-source data association analysis and model training and updating, and generating decision support information. The method is used in the intelligent breeding process.

Description

Technical Field

[0001] This application relates to the fields of intelligent aquaculture and computer vision technology, and in particular to an intelligent aquaculture behavior analysis system and method based on edge computing and AI vision. Background Technology

[0002] Traditional livestock farming management relies primarily on manual inspections, which is inefficient, costly, and unable to achieve 24 / 7 monitoring, easily leading to missed anomalies. Existing video surveillance-based solutions mostly employ cloud-based processing, which suffers from high data transmission bandwidth consumption, high processing latency, and data privacy risks, failing to meet real-time requirements. Furthermore, the behavior recognition algorithms used in these systems are often general-purpose models, lacking accuracy and robustness in the specific context of livestock farming. They typically only perform single, superficial behavior recognition, lacking a deep analytical framework for analyzing behavioral temporal patterns, group dynamics, and their correlation with environmental factors. This results in delayed anomaly warnings, hindering early detection and scientific quantitative assessment of animal health and welfare issues. Summary of the Invention

[0003] This application provides an intelligent aquaculture behavior analysis system and method based on edge computing and AI vision, which solves the technical problems of low efficiency, poor behavior recognition accuracy and delayed abnormal early warning caused by the reliance on manual and cloud processing in existing aquaculture management.

[0004] To achieve the above objectives, this application adopts the following technical solution: Firstly, an intelligent aquaculture behavior analysis system based on edge computing and AI vision is provided, including: edge devices, fog server and cloud platform; The edge device is used to acquire video streams in real time, and to perform target detection and key point recognition on the video streams using a lightweight deep learning model to obtain preliminary behavior classification results. The fog server is used to aggregate several preliminary behavior classification results, perform complex behavior recognition, behavior time sequence analysis and abnormal behavior detection, obtain abnormal behavior results and trigger real-time early warnings; The cloud platform is used to receive and store all data from the farm, perform long-term data mining, multi-source data correlation analysis, model training and updating, and generate decision support information.

[0005] Based on the above technical solutions, the intelligent aquaculture behavior analysis system based on edge computing and AI vision provided in this application optimizes the entire process of aquaculture behavior analysis through clear division of labor and efficient collaboration in its architecture: localized real-time processing at the edge significantly reduces data transmission bandwidth and cloud dependence from the source, ensuring low-latency response and data privacy; deep analysis at the cloud gathers information from multiple sources, improving the accuracy and timeliness of complex behavior recognition and anomaly detection, and achieving precise early warning; while the long-term mining and model evolution capabilities in the cloud enable the system not only to support macro-level decision-making and correlation analysis, but also to have the ability to continuously learn and adaptively optimize, thus realizing a closed loop from real-time perception to intelligent decision-making, significantly improving the level of refinement, automation and intelligence in aquaculture management.

[0006] In conjunction with the first aspect above, in one possible implementation, the edge device includes a high-definition smart camera and a first AI processing module; The high-definition smart camera supports visible light and infrared imaging and is used to collect video data of the breeding area 24 hours a day. The first AI processing module is equipped with a lightweight deep learning model for real-time target detection, key point recognition, and preliminary behavior classification of video streams. The lightweight deep learning model is a model built based on a combination of MobileNetV3, YOLO-Tiny, and Lite-HRNet after model pruning, quantization, and knowledge distillation optimization.

[0007] In conjunction with the first aspect above, in one possible implementation, the fog server includes a data aggregation module and a second AI processing module; The data aggregation module is used to receive and fuse recognition results and keyframe images from multiple edge devices, and perform time synchronization to obtain a behavior sequence; The second AI processing module runs a complex behavior recognition model and a behavior time series analysis model, used for: Based on the improved DeepSORT algorithm and Re-ID technology, a unique ID is assigned to each poultry and continuously tracked to establish an individual behavior profile; Identify typical poultry behaviors using complex behavior recognition models; By modeling behavioral sequences using a temporal convolutional network (TCN) and a long short-term memory network (LSTM), the duration, frequency, and rhythmicity of behaviors are identified, and a behavior transition matrix is ​​constructed to obtain real-time analysis results.

[0008] In conjunction with the first aspect described above, in one possible implementation, the fog server further includes an anomaly warning module; wherein the anomaly warning module includes: A dynamically updated behavioral baseline model is established for each field based on historical data. Multiple anomaly detection methods, including statistical methods, isolated forest algorithm, LSTM autoencoder and DBSCAN clustering algorithm, are employed to compare real-time analysis results with behavioral baseline models to detect abnormal behaviors of individuals and groups, and obtain abnormal behavior results; wherein, the abnormal behavior results include the degree of abnormality and the duration. Based on the degree and duration of the anomaly, a tiered early warning mechanism is activated, which includes at least a Level 1 observation warning, a Level 2 manual inspection warning, and a Level 3 immediate action warning.

[0009] In conjunction with the first aspect above, in one possible implementation, the cloud platform includes a model training center and a data mining center; The model training center is used to collect difficult example samples from edge devices and fog servers, perform manual annotation and incremental learning, train an optimized AI model, and distribute it to edge devices and fog servers via OTA for model updates. The data mining center is used to integrate behavioral data, environmental data, production data, and health data to build a data warehouse. It uses correlation analysis, regression analysis, or causal inference methods to mine the correlation between behavioral patterns and environmental parameters and production performance indicators, and builds a correlation and prediction model of behavior-environment-production performance based on the gradient boosting tree (XGBoost) algorithm.

[0010] In conjunction with the first aspect described above, in one possible implementation, the cloud platform further includes an animal welfare assessment module; wherein, the animal welfare assessment module includes: Based on the five freedoms of animal welfare, a quantitative evaluation index system is constructed that includes physiological welfare, environmental welfare, health welfare, behavioral welfare, and psychological welfare. Based on the analysis results provided by the data mining center, a weighted scoring model is used to automatically generate a comprehensive welfare score. When the overall welfare score falls below a preset threshold, a welfare alert is triggered and improvement suggestions are generated.

[0011] In conjunction with the first aspect mentioned above, in one possible implementation, when the fog terminal server triggers the graded early warning mechanism, it also links with the environmental control system of the farm to automatically adjust at least one of the parameters of temperature, humidity, ventilation, or light.

[0012] In conjunction with the first aspect mentioned above, one possible implementation further includes: an early warning response terminal, which is connected to the abnormal early warning module of the fog server to receive early warning information; and pushes the early warning information via a mobile APP, WeChat, or SMS, along with the time, location, type, severity of the abnormality, and video clips from a period of time before and after the abnormality.

[0013] In conjunction with the first aspect above, in one possible implementation, the edge device, fog server, and cloud platform adopt a dynamic task scheduling strategy to dynamically allocate AI inference tasks based on network bandwidth, computing load, and task priority.

[0014] Secondly, a method for intelligent aquaculture behavior analysis based on edge computing and AI vision is provided, including: Video streams are captured in real time by edge devices, and image denoising, enhancement, and distortion correction are performed. A lightweight deep learning model is used to perform real-time object detection, tracking, and preliminary behavior classification on video streams, and output preliminary recognition results. The fog server aggregates preliminary identification results and keyframe images from several edge devices for data fusion and time synchronization. Complex AI models are used to perform multi-target continuous tracking, complex behavior recognition, behavior time series analysis, and abnormal behavior detection based on behavior baseline models on the aggregated data. When abnormal behavior is detected, an early warning is triggered according to the hierarchical early warning mechanism, and alarm information is pushed or linkage control operations are performed. The aggregated data and key video clips are periodically uploaded to the cloud platform via fog servers for cleaning and mining, and a data warehouse is established. Historical data is stored long-term, behavioral pattern analysis is performed, correlation analysis is conducted, and predictive modeling is carried out through a cloud platform. The AI ​​model is optimized by collecting difficult sample data through a cloud platform for incremental learning and then pushing it to edge devices and fog servers via OTA. Based on the data analysis results, breeding management analysis reports, animal welfare assessment reports, and decision-making recommendations are generated.

[0015] Thirdly, an electronic device is provided, comprising: a communication unit and a processing unit; the communication unit is used to acquire video streams in real time; the processing unit is used to perform target detection and key point recognition on the video streams using a lightweight deep learning model to obtain preliminary behavior classification results; aggregate several preliminary behavior classification results and perform complex behavior recognition, behavior temporal analysis and abnormal behavior detection to obtain abnormal behavior results and trigger real-time early warning; receive and store all data from the farm, perform long-term data mining, multi-source data association analysis and model training and updating, and generate decision support information.

[0016] Fourthly, this application provides an electronic device, including: a processor and a storage medium; the storage medium includes instructions, and the processor is configured to execute the instructions to implement the methods described in the first aspect and any possible implementation thereof. The electronic device may be an electronic device or a chip within an electronic device.

[0017] Fifthly, this application provides an intelligent aquaculture behavior analysis system based on edge computing and AI vision, comprising: edge devices, a cloud server, and a cloud platform; wherein, the edge devices are used to collect video streams in real time, perform target detection and key point recognition on the video streams using a lightweight deep learning model, and obtain preliminary behavior classification results; the cloud server is used to aggregate several preliminary behavior classification results, and perform complex behavior recognition, behavior temporal analysis, and abnormal behavior detection, obtain abnormal behavior results, and trigger real-time early warnings; the cloud platform is used to receive and store all data from the aquaculture farm, perform long-term data mining, multi-source data correlation analysis, and model training and updating, and generate decision support information.

[0018] In a sixth aspect, this application provides a computer-readable storage medium storing instructions that, when executed on an electronic device, cause the electronic device to perform the methods described in the first aspect and any possible implementation thereof.

[0019] In a seventh aspect, this application provides a computer program product containing instructions that, when run on an electronic device, cause the electronic device to perform the methods described in the first aspect and any possible implementation thereof.

[0020] This application provides an intelligent livestock behavior analysis system and method based on edge computing and AI vision. By constructing a distributed computing architecture that coordinates the edge, fog, and cloud, it achieves localized real-time processing and intelligent analysis of livestock behavior data, significantly reducing system latency and cloud bandwidth dependence while ensuring data privacy. By deploying a lightweight AI model optimized for livestock scenarios, the system can accurately identify various typical behaviors and, based on time-series analysis and multi-source data fusion, achieve early, accurate warnings and tiered responses to abnormal individual and group behaviors. Simultaneously, the system establishes for the first time an automated quantitative assessment system for animal welfare and can deeply explore the intrinsic relationships between behavior, environment, and production performance, providing comprehensive data support for livestock management from real-time monitoring to strategic decision-making, significantly improving the refinement, intelligence, and economic efficiency of livestock management.

[0021] It should be understood that the descriptions of technical features, technical solutions, beneficial effects, or similar language in this application do not imply that all features and advantages can be achieved in any single embodiment. Rather, it is understood that the description of a feature or beneficial effect means that a specific technical feature, technical solution, or beneficial effect is included in at least one embodiment. Therefore, the descriptions of technical features, technical solutions, or beneficial effects in this specification do not necessarily refer to the same embodiment. Furthermore, the technical features, technical solutions, and beneficial effects described in this embodiment can be combined in any suitable manner. Those skilled in the art will understand that embodiments can be implemented without one or more specific technical features, technical solutions, or beneficial effects of a particular embodiment. In other embodiments, additional technical features and beneficial effects may be identified in specific embodiments that do not embody all embodiments. Attached Figure Description

[0022] Figure 1 A system architecture diagram of an intelligent aquaculture behavior analysis system based on edge computing and AI vision provided for embodiments of this application; Figure 2 A schematic diagram illustrating a process for intelligent aquaculture behavior analysis based on edge computing and AI vision, provided for an embodiment of this application; Figure 3 A schematic diagram illustrating another intelligent aquaculture behavior analysis based on edge computing and AI vision provided for an embodiment of this application; Figure 4 A flowchart illustrating an intelligent aquaculture behavior analysis method based on edge computing and AI vision, provided for an embodiment of this application; Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application; Figure 6 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0023] In the description of this application, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. The "and / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "multiple" means two or more. The terms "first," "second," etc., do not limit the quantity or order of execution, and "first," "second," etc., do not necessarily imply differences.

[0024] It should be noted that, in this application, the terms "exemplary" or "for example" are used to indicate that something is being described as an example, illustration, or illustration. Any embodiment or design described as "exemplary" or "for example" in this application should not be construed as being more preferred or advantageous than other embodiments or design solutions. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0025] The intelligent aquaculture behavior analysis system based on edge computing and AI vision provided in this application embodiment can be applied to, for example... Figure 1 The intelligent aquaculture behavior analysis system 100 based on edge computing and AI vision shown below, such as Figure 1 As shown, the communication system includes: an edge device 10, a fog server 20, and a cloud platform 30; Among them, edge device 10 is used to collect video streams in real time, and to perform target detection and key point recognition on the video streams through a lightweight deep learning model to obtain preliminary behavior classification results; Fog server 20: Used to aggregate several preliminary behavior classification results, and perform complex behavior recognition, behavior time sequence analysis and abnormal behavior detection to obtain abnormal behavior results and trigger real-time early warning; Cloud Platform 30: Used to receive and store all data from the farm, perform long-term data mining, multi-source data correlation analysis, model training and updating, and generate decision support information.

[0026] To address the technical problems of low efficiency, poor accuracy of behavior recognition, and delayed anomaly warning caused by the reliance on manual and cloud processing in existing aquaculture management, this application provides an intelligent aquaculture behavior analysis system based on edge computing and AI vision. The system includes: edge devices, fog server, and cloud platform. The edge device is used to acquire video streams in real time, and to perform target detection and key point recognition on the video streams using a lightweight deep learning model to obtain preliminary behavior classification results. The fog server is used to aggregate several preliminary behavior classification results, perform complex behavior recognition, behavior time sequence analysis and abnormal behavior detection, obtain abnormal behavior results and trigger real-time early warnings; The cloud platform is used to receive and store all data from the farm, perform long-term data mining, multi-source data correlation analysis, model training and updating, and generate decision support information.

[0027] Based on this, the technical problems of low efficiency, poor accuracy of behavior recognition, and delayed abnormal early warning caused by the reliance on manual and cloud processing in existing aquaculture management have been solved.

[0028] The intelligent aquaculture behavior analysis system based on edge computing and AI vision provided in this application includes: Edge devices: used to acquire video streams in real time, and perform object detection and key point recognition on the video streams using lightweight deep learning models to obtain preliminary behavior classification results.

[0029] In some implementations, the edge device includes a high-definition smart camera and a first AI processing module; The high-definition smart camera supports visible light and infrared imaging and is used to collect video data of the breeding area 24 hours a day. The first AI processing module is equipped with a lightweight deep learning model for real-time target detection, key point recognition, and preliminary behavior classification of video streams. The lightweight deep learning model is a model built based on a combination of MobileNetV3, YOLO-Tiny, and Lite-HRNet after model pruning, quantization, and knowledge distillation optimization.

[0030] For example, taking the analysis of pig farming behavior as an example, the edge device consists of a high-definition smart camera supporting 1080P / 4K resolution and a first AI processing module equipped with an AI inference chip (such as NVIDIA Jetson, Huawei Ascend, or Rockchip RK3588). The camera has visible light and infrared night vision functions, is powered by PoE and achieves IP67 protection level. It is deployed in the building to achieve multi-angle, 24-hour uninterrupted video acquisition. The first AI processing module runs a lightweight deep learning model optimized by model pruning, quantization, and knowledge distillation techniques. The model uses MobileNetV3 as the backbone network, combines YOLO-Tiny for target detection, and uses Lite-HRNet to extract key points of pigs. Finally, it forms an embedded system with a size of only 10-50MB that can achieve 30fps real-time inference with ≤15W power consumption, and completes real-time detection, posture estimation, and preliminary behavior classification of pigs such as feeding and lying down.

[0031] Fog server: Used to aggregate several preliminary behavior classification results, and perform complex behavior recognition, behavior time sequence analysis and abnormal behavior detection to obtain abnormal behavior results and trigger real-time alerts.

[0032] In some implementations, the fog server includes a data aggregation module and a second AI processing module; The data aggregation module is used to receive and fuse recognition results and keyframe images from multiple edge devices, and perform time synchronization to obtain a behavior sequence; The second AI processing module runs complex behavior recognition models and behavior time-series analysis models, such as... Figure 2 As shown, it is used for: Based on the improved DeepSORT algorithm and Re-ID technology, a unique ID is assigned to each poultry and continuously tracked to establish an individual behavior profile; Identify typical poultry behaviors using complex behavior recognition models; By modeling behavioral sequences using a temporal convolutional network (TCN) and a long short-term memory network (LSTM), the duration, frequency, and rhythmicity of behaviors are identified, and a behavior transition matrix is ​​constructed to obtain real-time analysis results.

[0033] It should be noted that the typical behaviors of poultry include feeding, drinking, lying down, standing, walking, fighting, mounting, tail biting, scratching, and rooting. The improved DeepSORT algorithm enables continuous tracking of multiple targets and can solve problems such as poultry occlusion and crossover.

[0034] As an example, the complex behavior recognition model can identify 10+ typical pig behaviors: feeding (head entering the feed trough), drinking (mouth touching the waterer), lying down (body touching the ground), standing (limbs supporting the body), walking (positional movement), rooting (repeated action of nose touching the ground), fighting (intense interaction between two or more pigs), mounting (reproductive behavior), tail biting (aggressive behavior), and scratching (body rubbing against the wall or ground).

[0035] For example, the fog server uses an edge server equipped with an RTX 4060 GPU accelerator card, 2TB NVMe SSD storage, and a gigabit network switch, and is deployed in the building control room. Its data aggregation module can simultaneously process data from 16 cameras, constructing a complete behavioral sequence at a rate of 30 frames per second. The second AI processing module uses an improved DeepSORT algorithm to achieve a multi-target tracking accuracy of over 98% in intensive farming scenarios, and establishes an individual profile for each pig containing 12 typical behaviors (feeding, drinking, lying down, standing, walking, fighting, mounting, tail biting, etc.). By integrating the TCN and LSTM temporal analysis model, the system continuously analyzes behavioral patterns with a 5-second time window, accurately counting the duration (error ±0.5 seconds), frequency (counting accuracy >95%), and rhythmic characteristics of each behavior. The behavior transition matrix is ​​updated every 30 minutes. When the frequency of tail biting exceeds 8 times per hour or the continuous lying down time exceeds 4 hours, the system triggers a graded warning within 3 seconds and records 30 seconds of video evidence before and after the detection.

[0036] In some implementations, the fog server further includes an anomaly warning module; wherein, the anomaly warning module, such as... Figure 3 As shown, it includes: A dynamically updated behavioral baseline model is established for each field based on historical data. Multiple anomaly detection methods, including statistical methods, isolated forest algorithm, LSTM autoencoder and DBSCAN clustering algorithm, are employed to compare real-time analysis results with behavioral baseline models to detect abnormal behaviors of individuals and groups, and obtain abnormal behavior results; wherein, the abnormal behavior results include the degree of abnormality and the duration. Based on the degree and duration of the anomaly, a tiered early warning mechanism is activated, which includes at least a Level 1 observation warning, a Level 2 manual inspection warning, and a Level 3 immediate action warning.

[0037] It should be noted that a behavioral baseline model is established for each pen to statistically analyze behavioral distribution, diurnal rhythms, and individual differences under normal conditions. A sliding window method is used to dynamically update the baseline model, adapting to pig growth and seasonal changes.

[0038] As an example, a three-level warning system is set based on the degree and duration of the abnormality: Level 1 warning (mild abnormality, observation is the main approach): activity level decreases by 20-30%, feeding time decreases by 10-20 minutes; Level 2 warning (moderate abnormality, manual inspection): activity level decreases by 30-50%, frequent tail biting or fighting occurs, and lying down time increases abnormally; Level 3 warning (severe abnormality, immediate action is required): activity level decreases by more than 50%, prolonged periods without eating or drinking, abnormal postures or motor incoordination occur, and abnormal group behavior occurs.

[0039] For example, the anomaly warning module of the fog server establishes a dynamic behavioral baseline model for each pen based on 14 days of historical data. This model is automatically updated every 24 hours using a sliding window mechanism to adapt to changes in the pig's growth stage and environment. The module simultaneously runs four anomaly detection algorithms: a statistical method based on the 3σ principle to monitor deviations of single behavioral indicators in real time, an isolated forest algorithm to analyze multidimensional anomalies of behavioral features exceeding 15 dimensions, an LSTM autoencoder to detect temporal anomaly patterns in behavioral sequences, and a DBSCAN clustering algorithm to identify deviations in group behavior. When a 20-30% decrease in activity level is detected for 2 hours or a 15-minute reduction in feeding time, a level 1 warning is triggered. When a 40% decrease in activity level is accompanied by fighting more than 5 times per hour, a level 2 warning is initiated and manual verification is required. When a sudden drop in activity level of 60% is detected and there is no feeding for 3 consecutive hours, the system immediately triggers a level 3 warning and automatically notifies a veterinarian to intervene. The entire warning response delay is controlled within 3 seconds.

[0040] In some implementations, when the fog terminal server triggers the graded early warning mechanism, it also links with the environmental control system of the farm to automatically adjust at least one of the parameters of temperature, humidity, ventilation, or light.

[0041] For example, when the fog server triggers a level 2 warning (detecting pigs huddling together and activity decreasing by 35% for 1 hour), the system automatically activates the environmental control system while issuing a manual inspection alarm: raising the target pen temperature by 2°C, starting the wet curtain fan to increase ventilation by 15%, and maintaining this control until the behavioral data returns to normal; when a level 3 warning is triggered (monitoring group breathing rapidness and huddling for warmth), the system immediately performs comprehensive control: on the basis of raising the temperature by 3°C, simultaneously increasing the ventilation by 25% and turning on the auxiliary heating device, while temporarily adjusting the light intensity to 300 lux to promote pig activity, thereby actively intervening and alleviating abnormal behavior through precise automatic adjustment of environmental parameters.

[0042] In some implementations, the fog server further includes: an early warning response terminal, which communicates with the abnormal early warning module of the fog server to receive early warning information; and pushes the early warning information via a mobile APP, WeChat, or SMS, along with the time, location, type, severity of the abnormality, and video clips from a period of time before and after the abnormality.

[0043] For example, the early warning response terminal maintains real-time communication with the fog server through the API interface. When the system triggers a level 2 or higher warning, the terminal will send alarm information in parallel through three channels within 5 seconds: mobile APP push, WeChat official account template message, and SMS. The alarm content includes the precise time of the abnormality (accurate to the second), the specific column number, the type of abnormal behavior (such as "frequent tail biting"), and the severity level (such as "level 2 warning"). It will also automatically associate and push 30-second on-site video clips before and after the abnormal behavior (totaling 1 minute). Managers can directly view the video evidence on their mobile phones and start the handling process with one click, realizing closed-loop management from early warning to response.

[0044] Cloud platform: Used to receive and store all data from the farm, perform long-term data mining, multi-source data correlation analysis, model training and updating, and generate decision support information.

[0045] In some implementations, the cloud platform includes a model training center and a data mining center; The model training center is used to collect difficult example samples from edge devices and fog servers, perform manual annotation and incremental learning, train an optimized AI model, and distribute it to edge devices and fog servers via OTA for model updates. The data mining center is used to integrate behavioral data, environmental data, production data, and health data to build a data warehouse. It uses correlation analysis, regression analysis, or causal inference methods to mine the correlation between behavioral patterns and environmental parameters and production performance indicators, and builds a correlation and prediction model of behavior-environment-production performance based on the gradient boosting tree (XGBoost) algorithm.

[0046] It should be noted that the cloud platform is used for historical data analysis and mining, behavioral pattern learning, disease prediction model training, animal welfare assessment, production performance correlation analysis, report generation, and decision support. It establishes behavior-environment-production performance correlation models to guide precise management; optimizes environmental parameters to improve comfort; predicts production performance to optimize slaughter plans; and identifies optimal behavioral patterns for replication and promotion. Cloud-based sample collection continuously trains the model, while edge devices provide automatic OTA updates, continuously improving model accuracy and adapting to different breeds and scenarios, making the system increasingly intelligent with use.

[0047] As an example, correlation analysis, regression analysis, and causal inference were used to explore the relationships between behavior and other factors: the impact of environmental comfort on behavior (high temperatures lead to increased lying down and reduced activity); the impact of behavioral patterns on production performance (feeding duration is positively correlated with daily weight gain); and the association between behavioral abnormalities and disease occurrence (the probability of disease occurrence is 80% within 3 days of decreased activity). Predictive models were trained based on historical data: daily weight gain prediction (based on feeding behavior and activity level), slaughter time prediction, disease risk prediction, and feed conversion ratio prediction. Ensemble learning methods such as XGBoost and random forest were used to construct the predictive models.

[0048] For example, the cloud platform's model training center collects approximately 5,000 difficult sample cases from edge and fog devices each week. After manual annotation, the model is trained and optimized through incremental learning. Updated versions are deployed to farms every two weeks via OTA, continuously improving the model's accuracy in pig identification. The data mining center integrates over 18 months of behavioral data, environmental data (temperature and humidity, ammonia concentration, light intensity), production data (daily weight gain, feed conversion ratio), and health records to build a data warehouse of over 50TB. The predictive model built using the XGBoost algorithm accurately predicts daily weight gain (94% accuracy), slaughter time (error ±3 days), and disease risk (AUC value 0.92). It also discovered a correlation that pigs' lying time increases by 35% when the ambient temperature exceeds 28℃. Based on this, it automatically generates optimization suggestion reports to guide multiple pig farms in implementing standardized management.

[0049] In some implementations, the cloud platform further includes an animal welfare assessment module; wherein the animal welfare assessment module includes: Based on the five freedoms of animal welfare, a quantitative evaluation index system is constructed that includes physiological welfare, environmental welfare, health welfare, behavioral welfare, and psychological welfare. Based on the analysis results provided by the data mining center, a weighted scoring model is used to automatically generate a comprehensive welfare score. When the overall welfare score falls below a preset threshold, a welfare alert is triggered and improvement suggestions are generated.

[0050] It should be noted that the animal welfare assessment module automatically conducts welfare assessments in an objective, quantitative, and traceable manner; the accuracy of welfare scoring is guaranteed; welfare issues can be identified and addressed in a timely manner; brand image and market competitiveness are enhanced; and it aligns with future regulatory trends.

[0051] As an example, based on the EU's five freedoms of animal welfare principles, quantitative assessment indicators were constructed: physiological welfare (normal feeding and drinking behavior, good body condition); environmental welfare (comfortable lying down, suitable temperature and humidity); health welfare (no signs of disease, normal movement); behavioral welfare (able to express natural behaviors, no abnormal behaviors); and psychological welfare (no fear, no chronic stress). AI vision was used to automatically statistically analyze various indicators: lying down rate (normal 60-80%), feeding uniformity, drinking frequency, activity duration, fighting frequency, tail biting frequency, and environmental comfort behavior (clustering or dispersal). A weighted scoring model was used to generate a comprehensive welfare score (0-100 points), which was compared with industry benchmarks and historical data. Changes in the welfare score were monitored in real time, with automatic alerts when scores fell below a threshold, and improvement suggestions provided: adjusting temperature and humidity, increasing feeding frequency, improving pen environment, adjusting stocking density, and enrichment measures (toys, bedding).

[0052] For example, the cloud platform's animal welfare assessment module, based on the EU's Five Freedoms of Animal Welfare, has constructed a quantitative assessment system comprising 5 primary indicators and 18 secondary indicators. Using analysis results provided by the data mining center, it automatically calculates key indicators such as lying-down rate (normal range 60-80%), feed intake uniformity, and water frequency. An XGBoost weighted regression model is used to generate a comprehensive welfare score of 0-100 points every 4 hours. When a pen's score falls below the 70-point threshold twice consecutively, the system automatically triggers a welfare warning and generates specific improvement suggestions based on correlation analysis results, such as "lowering the ambient temperature from 32℃ to 26℃, increasing the daily feeding frequency to 6 times, and adding bedding to the northeast corner pen." These suggestions are then pushed to management for implementation, resulting in an increase in the average welfare score of the farm from 68 points to 82 points within four weeks of implementation.

[0053] In some implementations, a dynamic task scheduling strategy is adopted among the edge devices, fog servers, and cloud platforms to dynamically allocate AI inference tasks based on network bandwidth, computing load, and task priority.

[0054] For example, the edge, fog, and cloud are intelligently coordinated through a dynamic task scheduler: when the network bandwidth of an edge node is detected to be lower than 10Mbps, the system automatically migrates the complex behavior recognition tasks it is responsible for to the fog node for processing; when the GPU load of the fog server continuously exceeds 80%, some of its model training tasks are offloaded to the cloud; at the same time, the system sets priorities for various tasks, such as anomaly detection as the highest priority (P0), ensuring that the response can be completed within 100ms even when computing resources are scarce, while background tasks such as model updates are set to a low priority that can be flexibly scheduled (P2). This dynamic scheduling mechanism improves the overall resource utilization of the system by more than 40%, while ensuring the real-time performance of critical tasks.

[0055] Based on the above technical solutions, the intelligent aquaculture behavior analysis system based on edge computing and AI vision provided in this application achieves localized real-time processing and intelligent analysis of aquaculture behavior data by constructing a distributed computing architecture that coordinates the edge, fog, and cloud. This significantly reduces system latency and cloud bandwidth dependence, while ensuring data privacy. By deploying lightweight AI models optimized for aquaculture scenarios, the system can accurately identify various typical behaviors and, based on time-series analysis and multi-source data fusion, achieve early and accurate warnings and tiered responses to abnormal behaviors of individuals and groups. Simultaneously, the system establishes for the first time an automated quantitative assessment system for animal welfare and can deeply explore the intrinsic correlation between behavior, environment, and production performance, providing comprehensive data support for aquaculture management from real-time monitoring to strategic decision-making, significantly improving the refinement, intelligence, and economic benefits of aquaculture management.

[0056] In one possible implementation, embodiments of this application also provide an intelligent aquaculture behavior analysis method based on edge computing and AI vision, such as... Figure 4 As shown, it includes: Video streams are captured in real time by edge devices, and image denoising, enhancement, and distortion correction are performed. A lightweight deep learning model is used to perform real-time object detection, tracking, and preliminary behavior classification on video streams, and output preliminary recognition results. The fog server aggregates preliminary identification results and keyframe images from several edge devices for data fusion and time synchronization. Complex AI models are used to perform multi-target continuous tracking, complex behavior recognition, behavior time series analysis, and abnormal behavior detection based on behavior baseline models on the aggregated data. When abnormal behavior is detected, an early warning is triggered according to the hierarchical early warning mechanism, and alarm information is pushed or linkage control operations are performed. The aggregated data and key video clips are periodically uploaded to the cloud platform via fog servers for cleaning and mining, and a data warehouse is established. Historical data is stored long-term, behavioral pattern analysis is performed, correlation analysis is conducted, and predictive modeling is carried out through a cloud platform. The AI ​​model is optimized by collecting difficult sample data through a cloud platform for incremental learning and then pushing it to edge devices and fog servers via OTA. Based on the data analysis results, breeding management analysis reports, animal welfare assessment reports, and decision-making recommendations are generated.

[0057] For example, smart cameras deployed in pigsties collect 1080P video streams in real time at 30fps. At the edge, a lightweight model (a MobileNetV3-YOLO combined model compressed to 15MB) is used to complete pig detection and basic behavior classification. After the fog server aggregates 16 edge data streams, it achieves individual identity binding with 98% tracking accuracy through an improved DeepSORT algorithm. Simultaneously, a TCN-LSTM hybrid model is run to analyze behavioral temporal characteristics. When the frequency of tail biting in a pen exceeds 8 times / hour, the system initiates a level 2 warning within 3 seconds and links the environmental control system to increase ventilation by 15%. The cloud platform integrates more than 100,000 behavioral records and environmental data every week, achieves disease risk prediction with 94% accuracy through the XGBoost model, and pushes optimized models to terminal devices every half month via OTA. Finally, a weekly report is generated, which includes welfare scores (up from 78 to 85 points) and suggestions for optimizing the slaughter cycle, forming a complete closed loop from real-time perception to decision optimization.

[0058] Based on the above technical solution, a distributed architecture with edge-fog-cloud three-level collaboration was used to optimize and manage the entire process of aquaculture behavior analysis: the lightweight model and local processing at the edge ensured real-time performance and data privacy, the deep analysis and linkage control at the fog end enabled accurate early warning and immediate intervention, and the long-term mining and continuous learning in the cloud end enabled the system to have constantly evolving intelligence; ultimately, a complete technical closed loop was formed from real-time perception, accurate early warning to intelligent decision-making, which significantly improved the level of precision and economic benefits of aquaculture management.

[0059] The foregoing mainly describes the solutions of the embodiments of this application from the perspective of device implementation. It is understood that each device, such as an electronic device, includes at least one of the hardware structures and software modules corresponding to the execution of each function in order to achieve the above-mentioned functions. Those skilled in the art should readily recognize that, based on the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein, this application can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in a hardware or computer software-driven hardware manner depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0060] This application embodiment can divide the electronic device into functional units according to the above method example. For example, each function can be divided into a separate functional unit, or two or more functions can be integrated into one processing unit. The integrated unit can be implemented in hardware or as a software functional unit. It should be noted that the unit division in this application embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods.

[0061] When using integrated units, Figure 5 A possible structural schematic diagram of the electronic device (referred to as electronic device 50) involved in the above embodiments is shown. The electronic device 50 includes a processing unit 501 and a communication unit 502, and may also include a storage unit 503. Figure 5 The structural diagram shown can be used to illustrate the structure of the electronic device involved in the above embodiments.

[0062] when Figure 5 The schematic diagram shown is used to illustrate the structure of the electronic device involved in the above embodiments. The processing unit 501 is used to control and manage the operation of the electronic device, the communication unit 502 is used for the electronic device to communicate with other devices, and the storage unit 503 is used to store the program code and data of the electronic device.

[0063] For example, communication unit 502 is used for real-time acquisition of video streams; The processing unit 501 is used to perform target detection and key point recognition on the video stream using a lightweight deep learning model to obtain preliminary behavior classification results; aggregate several preliminary behavior classification results and perform complex behavior recognition, behavior time sequence analysis and abnormal behavior detection to obtain abnormal behavior results and trigger real-time early warning; receive and store all data from the farm, perform long-term data mining, multi-source data association analysis and model training and updating, and generate decision support information.

[0064] The processing unit 501 can be a processor or a controller, and the communication unit 502 can be a communication interface, transceiver, transceiver circuit, transceiver device, etc. The term "communication interface" is a general term and may include one or more interfaces. The storage unit 503 can be a memory. When the electronic device 50 is a chip, the processing unit 501 can be a processor or a controller, and the communication unit 502 can be an input interface and / or an output interface, pins, or circuits, etc. The storage unit 503 can be a storage unit within the chip (e.g., a register, cache, etc.) or a storage unit located outside the chip (e.g., read-only memory (ROM), random access memory (RAM, etc.).

[0065] The communication unit can also be called a transceiver unit. The antenna and control circuit with transceiver functions in the electronic device 50 can be considered as the communication unit 502 of the electronic device 50, and the processor with processing functions can be considered as the processing unit 501 of the electronic device 50. Optionally, the device in the communication unit 502 used to implement the receiving function can be considered as the communication unit. The communication unit is used to execute the receiving steps in the embodiments of this application, and the communication unit can be a receiver, a receiver circuit, etc. The device in the communication unit 502 used to implement the transmitting function can be considered as the transmitting unit. The transmitting unit is used to execute the transmitting steps in the embodiments of this application, and the transmitting unit can be a transmitter, a transmitter, a transmitting circuit, etc.

[0066] Figure 5 If the integrated units in the process are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solutions of the embodiments of this application, in essence, or the parts that contribute to the prior art, or all or part of the technical solutions, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute all or part of the steps of the methods described in the various embodiments of this application. Storage media for storing computer software products include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory, random access memory, magnetic disks, or optical disks.

[0067] Figure 5 The units in the process can also be called modules; for example, a processing unit can be called a processing module.

[0068] This application also provides a hardware structure diagram of an electronic device (denoted as electronic device 60), see [link to diagram]. Figure 6 The electronic device 60 includes a processor 601, and optionally, a memory 602 connected to the processor 601.

[0069] In the first possible implementation, see Figure 6 The electronic device 60 also includes a transceiver 603. The processor 601, memory 602, and transceiver 603 are connected via a bus. The transceiver 603 is used to communicate with other devices or communication networks. Optionally, the transceiver 603 may include a transmitter and a receiver. The device in the transceiver 603 that implements the receiving function can be considered as a receiver, which is used to perform the receiving steps in the embodiments of this application. The device in the transceiver 603 that implements the transmitting function can be considered as a transmitter, which is used to perform the transmitting steps in the embodiments of this application.

[0070] Based on the first possible implementation method Figure 6 The structural diagram shown can be used to illustrate the structure of the electronic device involved in the above embodiments.

[0071] in, Figure 6 This can also be illustrated by a system chip in an electronic device. In this case, the actions performed by the aforementioned electronic device can be implemented by this system chip; the specific actions performed can be found above and will not be repeated here.

[0072] In implementation, each step of the method provided in this embodiment can be completed by integrated logic circuits in the processor or by instructions in software form. The steps of the method disclosed in the embodiments of this application can be directly manifested as being executed by a hardware processor, or being executed by a combination of hardware and software modules in the processor.

[0073] The processor in this application may include, but is not limited to, at least one of the following: a central processing unit (CPU), a microprocessor, a digital signal processor (DSP), a microcontroller unit (MCU), or an artificial intelligence processor, etc., which are various computing devices that run software. Each computing device may include one or more cores for executing software instructions to perform calculations or processing. The processor may be a separate semiconductor chip or integrated with other circuits into a single semiconductor chip. For example, it may be integrated with other circuits (such as encoding / decoding circuits, hardware acceleration circuits, or various bus and interface circuits) to form a SoC (System-on-a-Chip), or it may be integrated as a built-in processor within an ASIC. The ASIC with the integrated processor may be packaged separately or together with other circuits. In addition to the cores for executing software instructions to perform calculations or processing, the processor may further include necessary hardware accelerators, such as field-programmable gate arrays (FPGAs), PLDs (programmable logic devices), or logic circuits that implement dedicated logic operations.

[0074] The memory in the embodiments of this application may include at least one of the following types: read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions; random access memory (RAM) or other types of dynamic storage devices capable of storing information and instructions; or electrically erasable programmable-only memory (EEPROM). In some scenarios, the memory may also be a compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0075] This application also provides a computer-readable storage medium including instructions that, when run on a computer, cause the computer to perform any of the methods described above.

[0076] This application also provides a computer program product containing instructions that, when run on a computer, cause the computer to perform any of the methods described above.

[0077] This application also provides a chip including a processor and an interface circuit. The interface circuit is coupled to the processor, which is used to run computer programs or instructions to implement the above-described method. The interface circuit is used to communicate with other modules outside the chip.

[0078] In the above embodiments, implementation can be achieved, in whole or in part, through software, hardware, firmware, or any combination thereof. When implemented using software programs, implementation can be, in whole or in part, in the form of a computer program product. This computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium accessible to a computer or a data storage device containing one or more servers, data centers, etc., that can be integrated with the medium. The available media can be magnetic media (e.g., floppy disks, hard disks, magnetic tapes), optical media (e.g., DVDs), or semiconductor media (e.g., solid-state disks (SSDs)).

[0079] Although this application has been described herein in conjunction with various embodiments, those skilled in the art, by reviewing the accompanying drawings, disclosure, and appended claims, will understand and implement other variations of the disclosed embodiments in carrying out the claimed application. In the claims, the word "comprising" does not exclude other components or steps, and "a" or "an" does not exclude multiple instances. A single processor or other unit can implement several functions listed in the claims. While different dependent claims may recite certain measures, this does not mean that these measures cannot be combined to produce good results.

[0080] Although this application has been described in conjunction with specific features and embodiments, it is obvious that various modifications and combinations can be made thereto without departing from the spirit and scope of this application. Accordingly, this specification and drawings are merely exemplary illustrations of this application as defined by the appended claims, and are considered to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Clearly, those skilled in the art can make various alterations and modifications to this application without departing from the spirit and scope of this application. Thus, if such modifications and modifications of this application fall within the scope of the claims of this application and their equivalents, this application is also intended to include such modifications and modifications.

Claims

1. A smart aquaculture behavior analysis system based on edge computing and AI vision, characterized in that, This includes edge devices, fog servers, and cloud platforms; The edge device is used to acquire video streams in real time, and to perform target detection and key point recognition on the video streams using a lightweight deep learning model to obtain preliminary behavior classification results. The fog server is used to aggregate several preliminary behavior classification results, perform complex behavior recognition, behavior time sequence analysis and abnormal behavior detection, obtain abnormal behavior results and trigger real-time early warnings; The cloud platform is used to receive and store all data from the farm, perform long-term data mining, multi-source data correlation analysis, model training and updating, and generate decision support information.

2. The intelligent aquaculture behavior analysis system based on edge computing and AI vision according to claim 1, characterized in that, The edge device includes a high-definition smart camera and a first AI processing module; The high-definition smart camera supports visible light and infrared imaging and is used to collect video data of the breeding area 24 hours a day. The first AI processing module is equipped with a lightweight deep learning model for real-time target detection, key point recognition, and preliminary behavior classification of video streams. The lightweight deep learning model is a model built based on a combination of MobileNetV3, YOLO-Tiny, and Lite-HRNet after model pruning, quantization, and knowledge distillation optimization.

3. The intelligent aquaculture behavior analysis system based on edge computing and AI vision according to claim 1, characterized in that, The fog server includes a data aggregation module and a second AI processing module. The data aggregation module is used to receive and fuse recognition results and keyframe images from multiple edge devices, and perform time synchronization to obtain a behavior sequence; The second AI processing module runs a complex behavior recognition model and a behavior time series analysis model, used for: Based on the improved DeepSORT algorithm and Re-ID technology, a unique ID is assigned to each poultry and continuously tracked to establish an individual behavior profile; Identify typical poultry behaviors using complex behavior recognition models; By modeling behavioral sequences using a temporal convolutional network (TCN) and a long short-term memory network (LSTM), the duration, frequency, and rhythmicity of behaviors are identified, and a behavior transition matrix is ​​constructed to obtain real-time analysis results.

4. The intelligent aquaculture behavior analysis system based on edge computing and AI vision according to claim 3, characterized in that, The fog server further includes an anomaly warning module; wherein the anomaly warning module includes: A dynamically updated behavioral baseline model is established for each field based on historical data. Multiple anomaly detection methods, including statistical methods, isolated forest algorithm, LSTM autoencoder and DBSCAN clustering algorithm, are employed to compare real-time analysis results with behavioral baseline models to detect abnormal behaviors of individuals and groups, and obtain abnormal behavior results; wherein, the abnormal behavior results include the degree of abnormality and the duration. Based on the degree and duration of the anomaly, a tiered early warning mechanism is activated, which includes at least a Level 1 observation warning, a Level 2 manual inspection warning, and a Level 3 immediate action warning.

5. The intelligent aquaculture behavior analysis system based on edge computing and AI vision according to claim 1, characterized in that, The cloud platform includes a model training center and a data mining center; The model training center is used to collect difficult example samples from edge devices and fog servers, perform manual annotation and incremental learning, train an optimized AI model, and distribute it to edge devices and fog servers via OTA for model updates. The data mining center is used to integrate behavioral data, environmental data, production data, and health data to build a data warehouse. It uses correlation analysis, regression analysis, or causal inference methods to mine the correlation between behavioral patterns and environmental parameters and production performance indicators, and builds a correlation and prediction model of behavior-environment-production performance based on the gradient boosting tree (XGBoost) algorithm.

6. The intelligent aquaculture behavior analysis system based on edge computing and AI vision according to claim 5, characterized in that, The cloud platform also includes an animal welfare assessment module; wherein, the animal welfare assessment module includes: Based on the five freedoms of animal welfare, a quantitative evaluation index system is constructed that includes physiological welfare, environmental welfare, health welfare, behavioral welfare, and psychological welfare. Based on the analysis results provided by the data mining center, a weighted scoring model is used to automatically generate a comprehensive welfare score. When the overall welfare score falls below a preset threshold, a welfare alert is triggered and improvement suggestions are generated.

7. The intelligent aquaculture behavior analysis system based on edge computing and AI vision according to claim 4, characterized in that, When the fog terminal server triggers the graded early warning mechanism, it also links with the environmental control system of the farm to automatically adjust at least one of the parameters of temperature, humidity, ventilation, or light.

8. The intelligent aquaculture behavior analysis system based on edge computing and AI vision according to claim 4, characterized in that, Also includes: The early warning response terminal is connected to the abnormal early warning module of the fog server to receive early warning information; Warning information is pushed out via mobile app, WeChat or SMS, along with the time, location, type, severity of the anomaly, and video clips from a period of time before and after it occurred.

9. The intelligent aquaculture behavior analysis system based on edge computing and AI vision according to claim 1, characterized in that, The edge devices, fog servers, and cloud platform employ a dynamic task scheduling strategy, dynamically allocating AI inference tasks based on network bandwidth, computing load, and task priority.

10. A method for intelligent aquaculture behavior analysis based on edge computing and AI vision according to claim 1, characterized in that, include: Video streams are captured in real time by edge devices, and image denoising, enhancement, and distortion correction are performed. A lightweight deep learning model is used to perform real-time object detection, tracking, and preliminary behavior classification on video streams, and output preliminary recognition results. The fog server aggregates preliminary identification results and keyframe images from several edge devices for data fusion and time synchronization. Complex AI models are used to perform multi-target continuous tracking, complex behavior recognition, behavior time series analysis, and abnormal behavior detection based on behavior baseline models on the aggregated data. When abnormal behavior is detected, an early warning is triggered according to the hierarchical early warning mechanism, and alarm information is pushed or linkage control operations are performed. The aggregated data and key video clips are periodically uploaded to the cloud platform via fog servers for cleaning and mining, and a data warehouse is established. Historical data is stored long-term, behavioral pattern analysis is performed, correlation analysis is conducted, and predictive modeling is carried out through a cloud platform. The AI ​​model is optimized by collecting difficult sample data through a cloud platform for incremental learning and then pushing it to edge devices and fog servers via OTA. Based on the data analysis results, breeding management analysis reports, animal welfare assessment reports, and decision-making recommendations are generated.

Images