Intelligent breeding environment data analysis method and system based on dynamic perception

Abstract

The invention relates to the field of breeding management, and discloses an intelligent breeding environment data analysis method and system based on dynamic perception, and the method comprises the steps: dynamically collecting pig herd data through a mobile multi-mode perception terminal, and carrying out the data correction and segmentation through an anti-interference algorithm, so as to improve the accuracy; then, constructing a deep learning model, and fusing the physiological behaviors and the environmental data to generate a pig herd comprehensive comfort index; and finally, active regulation and control are realized through time sequence prediction, and instructions are issued to form a closed loop after the safety of the instructions is simulated and verified in the digital twin environment. The invention also provides a system for realizing the method. Through active prediction and simulation verification, the problems that in the prior art, sensing data are not accurate, a decision model is simple, regulation and control are lagged, and potential safety hazards exist are solved, the breeding environment can be accurately and safely optimized, and the intelligent level of breeding management is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of breeding management, in particular to a smart breeding environment data analysis method and system based on dynamic sensing. BACKGROUND

[0002] With the rapid development of large-scale pig breeding, the shortcomings of traditional environmental control mode in precision, dynamic adaptability and intelligence level are increasingly prominent. The current pig house ventilation system mostly uses linear control method based on single temperature index, which is difficult to adapt to the nonlinear dynamic relationship between pig physiological needs and environmental parameters.

[0003] Especially in the precise ventilation mode, the differentiated configuration of different fans, air pipes and spraying equipment makes it difficult to accurately adapt to the environmental control parameters. The existing system highly depends on manual experience adjustment, which has problems such as low efficiency and response lag, directly affecting the pig comfort and production performance, thereby causing the following problems:

[0004] 1. Insufficient data collection accuracy: phenomena such as pig body surface pollutant residue and group superposition shielding increase the recognition error of computer vision algorithm, making it difficult to accurately reflect the individual physiological state;

[0005] 2. Limited model adaptability: traditional environmental control model takes temperature as the core control target, ignoring the comprehensive influence of air quality, wind speed distribution and other multi-parameter on pig comfort, and linear control method cannot effectively cope with nonlinear time-varying system;

[0006] 3. Low system intelligence level: the existing environmental control device has limited computing power, which cannot support real-time operation of deep learning model, and the fusion processing and closed-loop control ability of multi-modal data are weak. SUMMARY

[0007] In view of the shortcomings of the prior art, the present application provides a smart breeding environment data analysis method and system based on dynamic sensing, which solves the problem that the existing technology cannot accurately, actively and safely meet the real needs of breeding objects due to inaccurate sensing data, simple decision-making model, lagging control response and safety hazards in automatic execution.

[0008] To achieve the above purpose, the present application realizes the following technical scheme: a smart breeding environment data analysis method and system based on dynamic sensing, comprising:

[0009] Step a, using a mobile multi-modal sensing terminal to dynamically collect multi-modal data including pig physiological behavior data and environmental parameter data;

[0010] Step b, anti-interference data processing is performed on the multi-modal data to improve data accuracy;

[0011] Step c, based on the anti-interference data processing data, through a nonlinear model to calculate the pig physiological behavior data and the coupling relationship between the environmental parameter data reflecting the pig group comprehensive comfort index;

[0012] Step d, based on the pig group comprehensive comfort index, generate control instructions and issue to the environment control equipment, form a closed loop control.

[0013] Preferably, in step a, the pig physiological behavior data dynamically collected by the mobile multi-modal sensing terminal includes body surface temperature data collected by a thermal imaging camera, lying or standing posture data collected by a depth camera, and respiratory or cough acoustic data collected by an acoustic acquisition device.

[0014] Preferably, the anti-interference data processing in step b includes at least one of the following steps:

[0015] For the influence of body surface dirt, a dirt area mask is generated by combining near-infrared spectrum analysis or image feature recognition, and the dirt area is excluded or compensated when calculating the body surface temperature;

[0016] For group superposition shielding, a skeleton key point-shape contour linkage analysis mechanism is used to identify and segment the pig individuals that are shielded.

[0017] For environmental heat source interference, a thermal radiation compensation model is established to correct the collected body surface temperature data.

[0018] Preferably, in step c, the nonlinear model is a neural network model; the pig group comprehensive comfort index is calculated by nonlinearly weighting and fusing the normalized physiological behavior feature vector and the environmental feature vector through the neural network model.

[0019] Preferably, at the initial stage of system operation, an initial control instruction is generated based on a classical model of pig group heat balance; and the nonlinear model is trained using the continuously dynamically collected multi-modal data to gradually replace the classical model.

[0020] Preferably, before the control instruction is issued to the environment control equipment in step d, the method further comprises:

[0021] A digital twin model corresponding to the physical environment of the breeding house is constructed;

[0022] The control instruction is simulated and run in the digital twin model to evaluate the possible environmental change effect;

[0023] Only after the control instruction passes the simulation verification, it is issued to the environment control equipment for execution.

[0024] Preferably, the step a further comprises: identifying abnormal calling or fierce fighting key priority events of the pig group, and automatically attaching semantic labels to the multi-modal data collected during the key priority events as high-value samples for training of the nonlinear model.

[0025] Preferably, further comprising: periodically analyzing the influence weight of each environmental factor in the environmental parameter data on the pig group comprehensive comfort index, and optimizing the input features of the nonlinear model according to the analysis result to eliminate or reduce the weight of low-influence factors.

[0026] The intelligent breeding environment data analysis system based on dynamic perception comprises:

[0027] The mobile multi-modal perception terminal is configured to dynamically collect multi-modal data including pig physiological behavior data and environmental parameter data.

[0028] The intelligent environment controller is in communication connection with the mobile multi-modal perception terminal and the environmental regulation equipment, and is configured to execute the intelligent breeding environment data analysis method.

[0029] The environmental regulation equipment connected with the intelligent environment controller is configured to receive and execute the regulation instructions generated by the intelligent environment controller.

[0030] The present application provides an intelligent breeding environment data analysis method and system based on dynamic perception. The present application has the following beneficial effects:

[0031] 1. The present application significantly improves the accuracy and comprehensiveness of physiological state perception of the breeding object by using a mobile multi-modal perception terminal and combining a targeted anti-interference data processing algorithm. The dynamic perception method overcomes the defects of limited monitoring range of fixed sensors and susceptibility to distance and angle, and the processing of dirt correction and superposition segmentation purifies the data from the source, so that the body temperature, posture, acoustic information and other information obtained by the system can more truly reflect the physiological health status of the pig group, providing a high-quality data basis for subsequent accurate decision-making.

[0032] 2. The present application realizes deep understanding and evaluation of the welfare level of the breeding object by constructing a physiological-environment nonlinear model based on deep learning and quantifying it into a pig group comprehensive comfort index. This method discards the traditional simple control logic which relies on single environmental parameter threshold, and can explore and utilize the complex coupling relationship between multi-dimensional physiological behavior data and environmental factors, so as to make a more realistic and comprehensive judgment on the overall comfort state of the pig group.

[0033] 3. This invention introduces a proactive control mechanism based on time-series prediction, transforming environmental control from a traditional passive response mode to a proactive management mode with foresight. The system can learn the historical variation patterns of pig herd comfort indicators and predict their future trends, thereby intervening and controlling the environment in advance and smoothly before environmental conditions deviate from the optimal range. This effectively avoids stress responses to pig herds caused by drastic changes in environmental parameters, enhancing the stability and timeliness of environmental control.

[0034] 4. By setting up a digital twin simulation verification step before issuing control commands, the safety and reliability of automated control decisions are greatly improved. Any preliminary control command must be pre-rendered in a virtual environment that is highly consistent with the physical pigsty to assess its potential impact on wind speed, temperature, and other fields. This effectively identifies and avoids potential adverse operating conditions, ensuring that every control operation is safe and effective.

[0035] 5. By establishing a complete intelligent system encompassing data collection, processing, modeling and analysis, and closed-loop control, the adaptability and intelligence level of the livestock farming environment management scheme have been enhanced. This system possesses the capability to iteratively upgrade from an initial model to a neural network model, and the model parameters can adaptively learn and optimize based on continuously acquired data, enabling it to adapt to the needs of pig herds of different ages, seasons, and health conditions, significantly reducing the complexity and frequency of manual intervention. Attached Figure Description

[0036] Figure 1 This is a schematic diagram of the system method flow of the present invention. Detailed Implementation

[0037] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] Please see the appendix Figure 1 This invention provides a method and system for analyzing smart aquaculture environment data based on dynamic perception, including:

[0039] A mobile multimodal sensing terminal is used to dynamically acquire multimodal data;

[0040] The collected raw multimodal data undergoes systematic anti-interference data processing.

[0041] Based on the processed data, a nonlinear model was constructed and the overall comfort index of the pig herd was calculated.

[0042] Based on the comprehensive comfort index of pig herds, and combined with prediction and simulation, closed-loop control instructions are generated and executed.

[0043] Specifically, in this embodiment, the dynamic acquisition step of multimodal data is the data source and foundation of the entire data analysis method. Its purpose is to comprehensively, accurately, and without interference acquire the original data stream that can reflect the true physiological state of the aquaculture object and its microenvironment.

[0044] This step is achieved through a mobile multimodal sensing terminal deployed inside the pigsty. To ensure stable operation and accurate positioning in the complex ground environment of the pigsty (such as the presence of standing water, manure, etc.), the terminal can adopt a track-type structure, moving back and forth along a pre-set track in key areas of the pigsty (e.g., above the passageways between different pig pens). Compared to fixed sensors, this deployment method can cover a wider monitoring range with limited hardware costs and can observe pigs from different angles and at closer distances, thereby acquiring data with a higher signal-to-noise ratio.

[0045] The mobile multimodal sensing terminal integrates a set of cooperating sensors, forming a compact multimodal sensing unit. This unit may specifically include:

[0046] Thermal imaging camera: This device is used to non-contactly capture the surface thermal radiation information of individual pigs and groups of pigs, and convert it into a visualized thermal imaging map. Each pixel on the thermal imaging map corresponds to a temperature value, which provides direct data input for subsequent analysis of the pig's body temperature, identification of inflammatory areas, and assessment of cold and heat stress.

[0047] Depth cameras: These devices, such as Time-of-Flight (ToF) or structured light cameras, can simultaneously acquire three-dimensional depth information of a scene. Their functions are multifaceted: First, they can accurately measure the distance between the terminal and the pigs, providing crucial parameters for subsequent thermal radiation compensation correction; second, they can acquire the three-dimensional contours and volume information of the pigs, used to distinguish individuals and determine whether the pigs are lying down or standing; finally, during system initialization or when the environment changes, depth cameras can be used to perform a three-dimensional scan of the entire pigsty, providing fundamental geometric data for building a digital twin model.

[0048] RGB High-Definition Camera: This device is used to capture high-definition visible light color images. These images are crucial for individual identification (such as through skin markings), analyzing for injuries or skin lesions, and helping to determine if the skin is covered in mud, feces, or other contaminants. The rich texture and color information it provides is an important supplement to thermal imaging and depth information.

[0049] Array-type acoustic acquisition devices: Unlike single microphones, array-type devices consist of multiple microphones arranged in a specific geometric configuration. By employing signal processing techniques such as beamforming, this device can focus the sound pickup in a specific direction or area, such as the pigpen directly below. This allows it to suppress noise interference from other directions in a noisy environment and more clearly capture the sounds emitted by pigs in the target area, such as breathing sounds, coughing sounds, and abnormal noises caused by fighting or discomfort.

[0050] The terminal's data acquisition task can be executed in two complementary working modes within a complete operating cycle:

[0051] Periodic inspection mode: In this mode, the terminal moves along the track at a set speed according to the pre-planned path and schedule in the system, performing a systematic and comprehensive data scan of the entire pigsty or designated responsibility area. This mode ensures the globality and regularity of data collection, and can provide the system with background data reflecting the overall status changes of the pig herd at different time periods (such as early morning, after feeding, midday rest, and evening).

[0052] Event-Driven Mode: This mode aims to enhance the system's responsiveness to sudden anomalies and the targeted nature of data capture. This mode is triggered when subsequent data processing or model analysis modules (such as in step S2 or S3) identify a preset key priority event. For example, when the acoustic analysis module detects an abnormal increase in cough frequency in a certain area within a short period, or the visual analysis module identifies continuous, intense fighting behavior, the system generates a high-priority dynamic instruction. This instruction temporarily interrupts the terminal's periodic inspection task, schedules it to quickly move to the target area where the event occurred, and performs enhanced data acquisition. Enhanced acquisition may include: short-distance back-and-forth movement within the area to observe from different angles, increasing the camera's frame rate, and extending the dwell time, thereby capturing richer and more detailed data about the anomaly.

[0053] In any of the above modes, the data streams acquired by all sensors are strictly synchronized. This means that every frame of thermal imaging, depth image, RGB image, and every segment of audio data will be precisely appended with a timestamp of the acquisition time and the three-dimensional spatial location information of the terminal in the pigsty coordinate system. This spatiotemporal alignment is a prerequisite for subsequent multimodal data fusion and analysis, ensuring that data from different sources can be accurately correlated to the same event or individual at the same time and in the same space.

[0054] The collected and labeled multimodal data streams constitute the raw input of the method described in this embodiment. This data is transmitted in real-time or in batches to the subsequent anti-interference data processing module, providing comprehensive and detailed raw information for subsequent data purification, feature extraction, and intelligent decision-making. Through this step, the present invention ensures that the source of its data analysis is dynamic, multidimensional, and intelligently focused on key events, thus laying a solid foundation for the effectiveness of the entire method.

[0055] In this embodiment, the anti-interference data processing step involves a series of purification and correction operations performed on the raw multimodal data collected in the preceding steps. This step is designed to address various physical and environmental interference factors present in real-world aquaculture environments. If these factors are not processed, they will severely affect the accuracy of the data and may lead to biases in subsequent model analysis and decision-making. This step introduces a targeted algorithm model to extract more effective information from the raw data that better reflects the true physiological and behavioral states of the aquaculture subjects.

[0056] This step may specifically include the following aspects of processing:

[0057] When using a thermal imaging camera for non-contact body temperature measurement, the raw temperature reading T is obtained. raw The apparent temperature is typically not the pure physiological surface temperature of the pig, but rather a mixture of various interfering factors. To obtain accurate physiological indicators, this embodiment designs a systematic correction process. First, the influence of contaminants on the pig's body surface (such as dry mud, moist feces, etc.) is addressed. These contaminants, due to their different temperature, emissivity, and heat capacity characteristics compared to clean skin, will form low-temperature or high-temperature artifact regions on the thermal imaging spectrum. Therefore, the system comprehensively utilizes high-resolution RGB images spatiotemporally aligned with the thermal imaging data. By analyzing the color, texture, and edge features in the RGB images, and combining this with near-infrared spectral analysis technology (if supported by the sensor terminal), a semantic segmentation model (e.g., U-Net or its variants) can be trained. This model can identify different regions in the image, such as "clean skin," "dry contaminants," and "moist contaminants," at the pixel level. The segmentation result is generated into a binary dirt mask M with the same size as the thermal imaging image. mask In this mask, pixels identified as clean skin are assigned a value of 1, while pixels identified as contaminants are assigned a value of 0.

[0058] Secondly, the interference from thermal radiation from the breeding environment is addressed. The floors, walls, and heating equipment (such as heat lamps) in the pigsty are all sources of thermal radiation. The infrared radiation emitted by these sources is reflected from the pig's body surface and received by the thermal imaging camera, leading to an overestimation of the measured temperature. This implementation establishes an environmental thermal radiation compensation model. The inputs to this model include: the three-dimensional spatial distance *d* from each point on the pig's body surface to the main environmental heat sources, measured in real time by a depth camera; and the average background temperature *T* measured by a fixed sensor. env The model outputs a thermal radiation compensation value T corresponding to each pixel of the image. comp It can be expressed by the following formula:

[0059] T comp =f rad (d,T env );

[0060] Among them, f rad It is a function established based on the fundamental laws of thermal radiation (such as the Stefan-Boltzmann law) and combined with empirical parameters.

[0061] Combining the two corrections mentioned above, the final corrected effective body surface temperature T is obtained. adj The calculation method for each pixel is as follows:

[0062] T adj =(T raw -T comp )~M mask ;

[0063] Through this processing, the system can use only validated, effective temperature data that represents the true skin condition when calculating key physiological indicators such as average body surface temperature and body temperature dispersion of pigs, thus providing reliable data input for subsequent health and comfort assessments.

[0064] In intensive farming, pigs, especially nursery and fattening pigs, often exhibit huddled lying-down behavior, leading to severe body occlusion between individuals. This occlusion makes it difficult for traditional contour- or shape-based image segmentation methods to accurately count individuals and analyze poses.

[0065] Using depth information from a depth camera, the image is preprocessed. By setting height thresholds and other methods, the clump of pigs is segmented from the background, forming a large "pig herd region" mask. Subsequently, a specially trained, lightweight animal pose estimation algorithm (e.g., a simplified version of DeepLabCut or HRNet) is deployed within this "pig herd region" mask. The model's training objective is not to locate all the skeletal points of the animal, but rather to focus on identifying key anatomical landmarks with high visibility and discriminative power even under partial occlusion, such as the center of the head, the midpoint of the spine, and the base of the tail. These key points combine to form a simplified "skeleton" that represents the core torso and orientation of the individual pig.

[0066] Once the model detects multiple sets of such "skeletons" within the pig herd area, the system can accurately count the number of stacked pigs based on the number of "skeletons." Simultaneously, by analyzing the length, curvature, and relative position of each "skeleton," the system can further determine the lying posture of each individual (e.g., lying on their side or prone) and assess the degree of crowding in the group. This mechanism deconstructs the complex problem of individual segmentation into a more manageable keypoint detection problem, significantly improving the ability to analyze individuals under high-density rearing conditions.

[0067] By performing the aforementioned anti-interference data processing steps, this invention can extract purer, more accurate, and more information-rich structured data from raw sensory data that is full of noise and uncertainty. This processed data lays a solid technical foundation for subsequently constructing high-precision physiological-environmental coupling models and making timely and effective regulatory decisions.

[0068] In this embodiment, the nonlinear modeling and calculation step of the pig herd comprehensive comfort index (S3) is a key link connecting data processing and execution control. The core task of this step is to integrate and quantify the purified and corrected multi-dimensional, heterogeneous data produced in the aforementioned step (S2) through a unified mathematical model, and finally generate an index that can intuitively and comprehensively reflect the overall welfare level of the pig herd, namely the "pig herd comprehensive comfort index (Cl)," and establish a nonlinear mapping relationship between this index and environmental control parameters.

[0069] In a preferred implementation, to ensure stable system operation at different stages, this step employs a hybrid startup and model iteration strategy. In the initial stage of system deployment, due to a lack of sufficient labeled historical data to train complex nonlinear models, the system first uses a classical model based on swine herd heat balance to generate initial, fundamental environmental control parameters. This initial model primarily calculates the total heat production Q of the swine herd based on thermodynamic principles. total Its sensible heat Q sWith latent heat Q l sum:

[0070] Q total =Q s +Q l ;

[0071] Among them, sensible heat Q s It is a function related to the total weight of the pig herd, feed intake, and ambient temperature, while the latent heat Q l This is a function related to the average respiratory rate, total mass, and relative humidity of the pig herd. This model provides a steady-state environmental control baseline that ensures the basic survival needs of the pig herd. Simultaneously, the system continuously collects and stores high-quality data processed by the aforementioned steps. When the accumulated data reaches a certain scale, it will be used to train a more complex deep neural network model capable of capturing nonlinear relationships. Once the performance of this deep neural network model is validated and proven to reflect the needs of the pig herd more precisely than the initial model, the system will smoothly transition the decision-making power for environmental control from the initial model to this neural network model.

[0072] The core of this deep neural network model lies in its ability to mathematically concretize and quantify the fuzzy concept of "herd comfort." In this embodiment, the "Herd Overall Comfort Index (CI)" is defined as a scalar or vector that integrates multidimensional information from physiological, behavioral, and environmental factors. Its calculation can be performed using a nonlinear fusion function F. fuse To express:

[0073]

[0074] To clearly illustrate this formula, its components are defined as follows:

[0075] P is a normalized physiological-behavioral feature vector.

[0076] The construction of this vector is based on the structured data output from the preceding steps. Its elements may include: the corrected average body surface temperature of the group, the standard deviation of the group's body surface temperature (this value reflects the temperature uniformity within the group; poor uniformity may indicate that some individuals are experiencing heat or cold stress), the average respiratory rate of the group obtained through acoustic analysis, the frequency of high-frequency coughing events, the proportion of pigs in relaxed postures such as lying on their sides or prone positions obtained through posture analysis, and the proportion of highly active pigs (such as those continuously walking or fighting) calculated through individual segmentation and tracking. ε is a normalized environmental feature vector. The elements of this vector are directly derived from environmental parameter data collected by fixed sensors, and may include: the average temperature, average relative humidity, average wind speed, and ammonia (NH4+) levels in the pigs' activity area. 3) Concentration, carbon dioxide (CO) 2) Concentration, etc.

[0077] w p and w e These are weight vectors corresponding to physiological and behavioral characteristics and environmental characteristics, respectively. Unlike traditional methods that require manual setting of weights, in this embodiment, these two weight vectors exist as trainable parameters of the neural network model.

[0078] During model training, using backpropagation and gradient descent, the model automatically learns and adjusts the contribution of each feature (e.g., whether temperature or humidity is more important) to the final comfort index based on the actual correlations inherent in the data. This adaptive weighting method allows the model to better adapt to the dynamic changes in the needs of pig herds at different ages, seasons, and health conditions. fuse It is a non-linear fusion function, which can be implemented as a multilayer perceptron consisting of one or more fully connected layers. This function receives two weighted sets of feature information as input, and performs deep fusion and transformation on this information through its non-linear activation function, finally outputting one or a set of values, which is the pig herd comprehensive comfort index (CI).

[0079] This indicator is a standardized value; for example, its range can be between 0 and 1. A higher value indicates that the overall condition of the pig herd is closer to the ideal comfort state. The training process of this deep neural network model is a supervised or semi-supervised learning process. Historical data segments are labeled by experts, or negative samples are generated by associating certain extreme physiological indicators (such as widespread high fever, group diarrhea, etc.) to provide the model with learning objectives. The optimization objective of the model is to minimize the difference between its predicted comfort index and the actual (or labeled) comfort index, which can be measured by loss functions such as mean squared error or cross-entropy.

[0080] Through this step, the present invention successfully transforms previously fragmented, multi-source, and heterogeneous data into a unified, quantifiable core indicator that dynamically reflects the overall welfare level of the pig herd. This indicator is not only a comprehensive snapshot of the current state of the pig herd, but its underlying nonlinear model also reveals the deep coupling relationship between complex environmental factors and the physiological and behavioral responses of the pig herd, providing a solid decision-making basis for subsequent precise and intelligent environmental regulation.

[0081] In this embodiment, the closed-loop control step is the final execution and implementation stage of the method described in this invention. This step receives the pig herd comprehensive comfort index (CI) calculated in the preceding steps, and uses it as the core decision-making basis. Through a series of intelligent processing steps, it generates specific, safe, and forward-looking control instructions, which are finally sent to the environmental control equipment in the pig house, forming a complete closed-loop control system of "perception-analysis-decision-execution".

[0082] In a preferred implementation, to overcome the response lag and potential risks present in traditional control logic, this step does not simply compare the comfort index with a fixed threshold, but integrates two core functional modules: active prediction and simulation verification.

[0083] The system-related content includes:

[0084] This system is the physical carrier for implementing the aforementioned methods. Through the organic synergy of its various components, it forms a complete technical closed loop from data acquisition, processing, decision-making to execution.

[0085] In a preferred embodiment, the system mainly includes: one or more mobile multimodal sensing terminals, an intelligent environmental controller as the core of the system, and an environmental control device connected to the intelligent environmental controller.

[0086] Mobile multimodal sensing terminal: This terminal is the system's "sensing organ," and its detailed composition and functions have been elaborated in the description of the aforementioned method step S1. Its core responsibility is to act as a mobile data acquisition platform, dynamically and closely acquiring comprehensive physiological and behavioral data of pigs and related environmental micro-data within the pigsty. The terminal integrates a thermal imaging camera, a depth camera, an RGB high-definition camera, and an array-type acoustic acquisition device, which work together to transmit the acquired raw multimodal data stream with precise spatiotemporal tags to the intelligent environmental controller for processing in real-time or near real-time via wireless (e.g., Wi-Fi, 5G) or wired communication. One or more such terminals can be deployed in the system to cover a wide breeding area or to monitor pigsties at different production stages in parallel.

[0087] Intelligent Environmental Controller: This device is the "brain" and "nerve center" of the entire system, responsible for all data processing, model computation, and decision generation tasks. In a typical physical implementation, this intelligent environmental controller can be an edge computing server deployed locally on the farm, or an industrial-grade computer with powerful computing capabilities. Its internal hardware configuration preferably includes a high-performance multi-core central processing unit (CPU) and graphics processing unit (GPU) to meet the needs of computationally intensive tasks such as deep learning model training and inference, as well as digital twin simulation. Its internal software architecture encapsulates the core algorithm modules of the method described in this invention.

[0088] Specifically, the intelligent environmental controller integrates the following functional units:

[0089] Data receiving and preprocessing unit: responsible for receiving data from mobile multimodal sensing terminals and fixed environmental sensors, and performing data alignment, format conversion and preliminary cleaning.

[0090] Anti-interference processing unit: This unit solidifies all the algorithms in step b, including the dirt and thermal radiation correction model for body surface temperature, and the group overlay individual segmentation model based on the linkage analysis of "skeletal key points - body contour". It processes the raw, noisy data stream into clean, accurate structured data.

[0091] Nonlinear Modeling and Analysis Unit: This unit is the core decision engine of the system, and it deploys the deep neural network model described in step S3. It receives structured data output from the anti-interference processing unit, calculates and updates the pig herd's overall comfort index (CI) in real time. Simultaneously, this unit is also responsible for the continuous training and iteration of the model, constantly optimizing the model parameters using newly collected data.

[0092] Time Series Prediction Unit: This unit deploys a short-term prediction model g based on recurrent neural networks such as LSTM. pred It continuously receives and analyzes the time series of CI indicators, outputs predictions of future comfort levels, and provides decision-making inputs for proactive regulation.

[0093] Digital Twin and Simulation Verification Unit: This unit stores a digital twin model of the breeding shed and integrates a fast CFD simulation engine. It receives initial control commands generated by the system and performs safety and effectiveness simulation verification in a virtual environment, acting as a "safety valve" to ensure the reliability of the control commands.

[0094] Command Generation and Issuance Unit: This unit integrates the current and predicted values ​​of the CI index with simulation verification results to generate final, safe, and refined control commands. These commands are then issued to the controllers of the environmental control equipment via standard industrial communication protocols (such as Modbus and OPC UA).

[0095] Human-Computer Interaction and Visualization Interface Unit: This unit provides a graphical user interface for pig farm managers. On this interface, managers can intuitively view real-time monitoring footage of the pigsty, curves showing changes in the overall comfort index of the pig herd, historical data on various environmental parameters, and records of automatically executed system controls. Simultaneously, managers are also allowed to manually intervene or adjust the system's control strategies when necessary.

[0096] Environmental control equipment: This part is the system's "effect organs" or "limbs," the ultimate executor of changes in the physical environment. It includes all automated environmental control equipment installed within the livestock sheds, such as:

[0097] Ventilation system: including variable frequency fans, air inlets, baffles, etc., used to control the airflow rate, air exchange rate and air distribution in the building.

[0098] Temperature control system: including water curtain, sprinkler cooling system, heating lamp, floor heating, etc., used to regulate the temperature inside the building.

[0099] Humidity control system: including spray or humidification equipment, used to regulate air humidity.

[0100] Other equipment, such as automatic feeders and automatic manure removal systems, although not direct environmental control devices, will have their operating status (such as feeding time) input into the intelligent environmental controller to help determine the behavioral rhythm of the pig herd.

[0101] These devices all communicate with intelligent environmental controllers via their controllers (such as PLCs). They passively receive instructions from the intelligent environmental controllers and execute them precisely, such as adjusting the fan speed to a specified percentage or turning on the sprinkler system in a designated area for a certain number of seconds.

[0102] In summary, the system described in this invention achieves a complete closed loop from perception to execution through the close collaboration of the three components. The mobile multimodal sensing terminal provides unprecedentedly rich dynamic data input; the intelligent environmental controller transforms the data into profound insights and intelligent decisions through a series of advanced algorithms; and the environmental control equipment precisely translates these decisions into modifications to the physical world. The entire system, as an organic whole, can continuously and adaptively maintain the aquaculture environment in an optimized state most conducive to the health and growth of the aquaculture organisms.

[0103] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A smart aquaculture environment data analysis method based on dynamic perception, characterized in that, include: Step a: Use a mobile multimodal sensing terminal to dynamically collect multimodal data, including pig physiological behavior data and environmental parameter data; Step b: Perform anti-interference data processing on the multimodal data to improve data accuracy; Step c: Based on the data after anti-interference data processing, calculate the comprehensive comfort index of the pig herd through a nonlinear model, which reflects the coupling relationship between the physiological behavior data of the pigs and the environmental parameter data. Step d: Based on the comprehensive comfort index of the pig herd, generate control instructions and send them to the environmental control equipment to form a closed-loop control.

2. The intelligent aquaculture environment data analysis method based on dynamic perception according to claim 1, characterized in that, In step a, the physiological behavior data of pigs dynamically collected by the mobile multimodal sensing terminal includes body surface temperature data collected by a thermal imaging camera, lying or standing posture data collected by a depth camera, and breathing or coughing acoustic data collected by an acoustic acquisition device.

3. The intelligent aquaculture environment data analysis method based on dynamic perception according to claim 1, characterized in that, The anti-interference data processing in step b includes at least one of the following steps: To address the impact of dirt on the body surface, a mask for the dirty area is generated by combining near-infrared spectral analysis or image feature recognition, and the dirty area is removed or compensated when calculating the body surface temperature. To address the issue of overlapping occlusion in groups, a skeletal key point-body contour linkage analysis mechanism is used to identify and segment occluded individual pigs. To address environmental heat source interference, a thermal radiation compensation model was established to correct the collected body surface temperature data.

4. The intelligent aquaculture environment data analysis method based on dynamic perception according to claim 1, characterized in that, In step c, the nonlinear model is a neural network model; the comprehensive comfort index of the pig herd is obtained by nonlinear weighted fusion calculation of normalized physiological behavior feature vectors and environmental feature vectors through the neural network model.

5. The intelligent aquaculture environment data analysis method based on dynamic perception according to claim 1, characterized in that, In the initial stage of system operation, the classical model based on the heat balance of the pig herd is used to generate initial control commands; and the nonlinear model is trained using the continuously dynamically collected multimodal data to gradually replace the classical model.

6. The intelligent aquaculture environment data analysis method based on dynamic perception according to claim 1, characterized in that, Before generating the control command in step d, the method further includes: Based on historical multimodal data from storage computing, learn and establish temporal behavioral patterns of pig herds; Based on the aforementioned temporal behavior pattern, the trend of changes in the comfort level of the pig herd in the near future is predicted. When it is predicted that the overall comfort index of the pig herd will deviate from the preset comfort range, a control strategy is generated and preloaded in advance to actively intervene in the environment.

7. The intelligent aquaculture environment data analysis method based on dynamic perception according to claim 1, characterized in that, Before issuing the control command to the environmental control device in step d, the method further includes: Construct a digital twin model corresponding to the physical environment of the livestock shed; The control commands are simulated and run in the digital twin model to evaluate their potential environmental effects. The control command is only sent to the environmental control equipment for execution after it has been verified by simulation.

8. The intelligent aquaculture environment data analysis method based on dynamic perception according to claim 1, characterized in that, Step a further includes: identifying key priority events such as abnormal vocalizations or violent fighting in the pig herd, and automatically attaching semantic labels to the multimodal data collected during the key priority events, so as to use them as high-value samples for training the nonlinear model.

9. The intelligent aquaculture environment data analysis method based on dynamic perception according to claim 1, characterized in that, Further steps include: periodically analyzing the influence weights of each environmental factor in the environmental parameter data on the overall comfort index of the pig herd, and optimizing the input features of the nonlinear model based on the analysis results to eliminate or reduce the weights of low-impact factors.

10. A smart aquaculture environment data analysis system based on dynamic perception, comprising the smart aquaculture environment data analysis method based on dynamic perception as described in any one of claims 1-9, characterized in that, include: The mobile multimodal sensing terminal is configured to dynamically collect multimodal data, including pig physiological behavior data and environmental parameter data. An intelligent environmental controller is communicatively connected to the mobile multimodal sensing terminal and the environmental control equipment, and the intelligent environmental controller is configured to perform the method steps as described in any one of claims 1 to 9; The environmental control equipment connected to the intelligent environmental controller is configured to receive and execute control commands generated by the intelligent environmental controller.

Images