A poultry health status monitoring method, device and medium

Abstract

This application discloses a method, device, and medium for monitoring the health status of poultry, executed by a data processing system deployed in a poultry house. The method includes: controlling an impedance imaging sensor array and a multimodal acquisition device group deployed in the poultry house to synchronously acquire monitoring data, and adding timestamps and corresponding spatial location information to the monitoring data to obtain a spatiotemporally aligned raw multimodal data stream; extracting poultry multimodal feature vectors based on the raw multimodal data stream; analyzing the poultry multimodal feature vectors and comparing them with a pre-established health status benchmark pattern to determine whether the poultry flock's health status is abnormal; and, when an abnormality is determined, generating monitoring results for the poultry health status by combining the correlation analysis results between the impedance spatiotemporal features, environmental features, acoustic features, and visual behavioral features.

Description

Technical Field

[0001] This application relates to the field of computer technology, and in particular to a method, device and medium for monitoring the health status of poultry. Background Technology

[0002] Intensive poultry farming is an important part of modern animal husbandry. In this production model, thousands or even tens of thousands of individuals live together in the same enclosed or semi-enclosed space. While this high-density farming increases production efficiency, it also significantly amplifies the risks and difficulties of managing the health of the population: once an infectious disease occurs, it can spread rapidly in a short period, causing severe economic losses; changes in environmental factors (such as temperature, humidity, ventilation, and harmful gases) can also quickly trigger population stress, affecting growth performance, welfare levels, and even immune status.

[0003] Therefore, the core value of real-time and accurate monitoring of the overall health status of poultry flocks lies in achieving early warning and disease identification at the "group" level. Traditional individual diagnosis (such as random sampling) is neither economical nor practical in large-scale farms, and often by the time obviously diseased individuals are discovered, the epidemic may have already been lurking or spreading in the flock, potentially leading to unavoidable losses. Summary of the Invention

[0004] This specification provides one or more embodiments of a method, device, and medium for monitoring the health status of poultry, which are used to solve the technical problems mentioned in the background art.

[0005] One or more embodiments of this specification employ the following technical solutions: This specification provides one or more embodiments of a method for monitoring the health status of poultry, the method comprising: Control the electrical impedance imaging sensor array and multimodal acquisition device group deployed in the poultry house to synchronously collect monitoring data, and add timestamps and corresponding spatial location information to the monitoring data to obtain a spatiotemporally aligned raw multimodal data stream. The monitoring data includes at least electrical impedance imaging measurement voltage data, environmental parameter data, audio data and video data. Based on the original multimodal data stream, a multimodal feature vector of poultry is extracted. The multimodal feature vector of poultry includes: extracting spatiotemporal impedance features based on the electrical impedance imaging voltage measurement data, extracting environmental features based on the environmental parameter data, and extracting acoustic features and visual behavioral features based on the audio data and video data. The spatiotemporal impedance features are the impedance distribution image of the poultry house cross section reconstructed based on the electrical impedance imaging voltage measurement data, and the relevant features characterizing the spatial distribution and dynamic activities of the poultry flock are extracted from the sequence impedance distribution image. The poultry multimodal feature vectors are analyzed and compared with a pre-established health status benchmark pattern to determine whether the poultry flock's health status is abnormal. When an anomaly is detected, the monitoring results of poultry health status are generated by combining the correlation analysis results between the impedance spatiotemporal characteristics, the environmental characteristics, the acoustic characteristics and the visual behavioral characteristics.

[0006] It should be noted that this method, by simultaneously acquiring and spatiotemporally aligning multimodal data such as impedance imaging, environmental data, audio, and video, first constructs an impedance distribution image and its dynamic characteristics that non-invasively and penetratingly reflects the internal spatial distribution and collective activity state of poultry flocks. This overcomes the limitations of traditional visual monitoring, which is easily obstructed and has difficulty in quantifying internal states, laying a data foundation for early anomaly detection at the flock level. Furthermore, by intelligently comparing the extracted multimodal features with pre-established health benchmark patterns, the system can keenly identify subtle deviations in flock behavior patterns and issue early warnings before individual clinical manifestations appear. Crucially, when an anomaly is detected, the method does not make isolated judgments but further integrates and analyzes the correlation between impedance anomaly patterns and environmental parameters, audio-visual features, thereby effectively distinguishing whether the root cause of the anomaly lies in environmental stress or potential disease risk. This allows poultry farmers to receive not just a single "abnormal alarm," but a "monitoring result" that includes possible causes, enabling them to take targeted measures (such as adjusting environmental control or conducting disease screening). This significantly advances the intervention window, fundamentally changing the traditional, lagging, and passive management model that relies on manual inspections and individual sampling. It significantly improves the predictability, accuracy, and proactive defense capabilities of large-scale poultry flock health management, playing a substantial role in reducing the risk of disease spread, minimizing economic losses, and protecting animal welfare.

[0007] Furthermore, the electrical impedance imaging sensor array includes multiple electrodes evenly arranged circumferentially along the inner wall of the poultry house; controlling the electrical impedance imaging sensor array deployed within the poultry house to synchronously acquire the electrical impedance imaging measurement voltage data includes: According to the preset excitation mode, safe AC excitation current is injected into the selected pair of electrodes in sequence; The boundary voltages between the remaining multiple electrode pairs in the electrical impedance imaging sensor array are measured simultaneously to obtain a voltage dataset for image reconstruction, thereby completing one full data acquisition cycle.

[0008] It should be noted that this method establishes a stable and reliable physical and data foundation for the entire monitoring method by clearly defining the configuration of the impedance imaging sensor array uniformly arranged along the circumference of the poultry house's inner wall, and the standard data acquisition process of "sequentially injecting excitation and synchronously measuring voltage." This design first ensures that the excitation current can form a stable and controllable electric field distribution across the cross-section of the poultry house, making subsequent measurements repeatable and comparable. The "synchronous measurement" operation ensures that the boundary voltage data acquired under one excitation corresponds to the poultry flock state at the same moment, effectively avoiding data timing misalignment caused by flock activity, which is crucial for capturing dynamic changes and using them for image reconstruction. The resulting complete and spatiotemporally synchronized "voltage dataset" is the necessary raw input for subsequent high-precision impedance distribution image reconstruction. The standardized data collection paradigm established in this step ensures the quality and consistency of the original data from the source, enabling the impedance distribution image reconstructed based on this data to more realistically and stably reflect the spatial distribution and activity status within the poultry flock. This provides a reliable data starting point for all advanced intelligent judgments such as subsequent feature extraction, pattern comparison, and correlation analysis. It is a prerequisite and technical guarantee for the entire system to achieve continuous, stable, and automated operation and generate accurate early warning results.

[0009] Furthermore, the reconstructing of the impedance distribution image of the poultry house cross-section based on the impedance imaging measurement voltage data includes: Establish a finite element calculation model corresponding to the cross-sectional geometry of the poultry house; Calculate the sensitivity matrix based on the finite element calculation model; The statistical value of the voltage data measured by electrical impedance imaging during historical healthy periods was used as the reference voltage; Based on the current impedance imaging measurement voltage data, the reference voltage, and the sensitivity matrix, a dynamic image reconstruction algorithm is used to obtain the conductivity change distribution map relative to the reference state at the current moment, which serves as the impedance distribution image.

[0010] It should be noted that this method, by establishing a finite element calculation model that strictly corresponds to the actual cross-sectional geometry of the poultry house, first provides a precise mathematical foundation for image reconstruction that conforms to the physical scene, ensuring the accuracy of solving both forward and inverse problems. Furthermore, by calculating the sensitivity matrix, the influence of internal conductivity changes on boundary measurements is quantified, establishing a reliable transformation bridge for inferring the internal state from the measured voltage. Crucially, the method innovatively introduces statistical values ​​of measurement data from "historical healthy periods" as a dynamic reference voltage benchmark. This effectively eliminates background signal interference caused by static factors such as the fixed structure and equipment layout of the poultry house, allowing the reconstruction target to focus on the relative conductivity changes caused by changes in the poultry flock's own state. Based on this, by combining real-time measurement data and the sensitivity matrix at the current moment, and using an image reconstruction algorithm specifically designed for dynamic targets, a stable and accurate impedance distribution image, characterized by relative changes, reflecting the spatial distribution and activity state of the poultry flock within the cross-section of the poultry house, can be generated. As the source of subsequent feature extraction, the accuracy and stability of this image directly determine the sensitivity and reliability of the entire monitoring system in sensing abnormal states within the flock, thus providing an irreplaceable core data representation for early detection of deviations in flock behavior patterns and achieving accurate early warning.

[0011] Furthermore, the determination of whether the health status of the poultry flock is abnormal includes: Calculate the deviation between the poultry multimodal feature vector at the current moment and the health status baseline pattern for the corresponding time period; If the deviation exceeds the dynamic threshold and the duration of the abnormal state exceeds the preset time window, the health status of the poultry flock is determined to be abnormal.

[0012] It should be noted that this method achieves an objective and digital assessment of poultry health status by quantitatively comparing the real-time extracted multimodal feature vectors of poultry with the corresponding time-period health status benchmark pattern and calculating the deviation, replacing the traditional manual judgment relying on subjective experience. Specifically, by setting a "dynamic threshold" instead of a fixed threshold, the system can adapt to the natural fluctuation range of normal poultry activity at different times (such as feeding and resting periods), improving the flexibility and accuracy of the judgment. More importantly, the introduction of a "duration exceeding a preset time window" criterion effectively filters out accidental and transient behavioral abnormalities caused by brief frights, temporary disturbances, etc., ensuring that the system only issues alarms for stable and continuous abnormal patterns. This dual criterion combining "deviation" and "persistence" greatly reduces the false alarm rate, making the final "abnormal poultry health status" judgment highly reliable. This ensures that managers can respond promptly to real and potential health risks, avoiding decreased trust and wasted resources due to frequent or inaccurate alarms, and enhancing the practicality and authority of the entire monitoring system.

[0013] Furthermore, the generation of poultry health status monitoring results by combining the correlation analysis results between the impedance spatiotemporal characteristics, the environmental characteristics, the acoustic characteristics, and the visual behavioral characteristics includes: The specific pattern of the impedance spatiotemporal characteristics on which the anomaly is determined is analyzed; Retrieve the environmental features associated with the abnormal spatial region indicated by the specific pattern, and determine whether there are environmental parameters that exceed the preset safety range; If present, a monitoring result indicating that environmental stress is the primary cause of the abnormality will be generated; If not, the acoustic features and visual behavioral features are combined. If the acoustic features or visual behavioral features indicate the presence of abnormal individual behavior, a monitoring result indicating potential disease risk is generated.

[0014] It should be noted that, based on the determination of abnormal poultry flock conditions, the introduction of a progressive attribution analysis logic based on multimodal feature association significantly enhances the decision support value of monitoring results. This method first analyzes the core electrical characteristics of the abnormality and the specific patterns of impedance spatiotemporal features, thereby pinpointing the nature and spatial location of the abnormality. Subsequently, instead of drawing conclusions in isolation, it prioritizes checking the environmental characteristics of the abnormal area. If environmental parameters are found to be excessive, the root cause of the abnormality is directly attributed to immediately intervened "environmental stress," enabling managers to adjust environmental control equipment immediately and quickly resolve the problem at its source. If environmental parameters are normal, it means the abnormality is more likely to originate from the flock itself; in this case, acoustic and visual behavioral characteristics are further integrated for corroboration. When these externally perceived characteristics also indicate the presence of abnormal individual behavior, the results can be clearly pointed to "potential disease risk," thereby triggering different biosafety and veterinary inspection response procedures. This "inside-out, environment-first-individual" correlation analysis and attribution mechanism means that the system outputs not just generalized "abnormal alerts," but also specific "diagnostic conclusions." This fundamentally elevates monitoring from the "perception" level to the "cognition" level, greatly enhancing the pertinence and timeliness of management responses. It avoids resource misallocation or delays caused by misjudgments and provides direct intelligent decision-making basis for achieving precise and efficient poultry health management.

[0015] Furthermore, the combination of the acoustic features and the visual behavioral features includes: When the environmental characteristics do not exceed the safety range, perform spectrum analysis and event detection on the audio data to identify specific abnormal acoustic events and their frequencies; The video data is then subjected to target detection and pose estimation models to identify specific abnormal individual behaviors and their numbers. If the frequency of the abnormal acoustic events exceeds a first frequency threshold, or the number of abnormal individual behaviors exceeds a first quantity threshold, then a monitoring result indicating potential disease risk is generated. If the frequency of the abnormal acoustic events exceeds a second frequency threshold, and the number of abnormal individual behaviors exceeds a second quantity threshold, a monitoring result indicating a high-confidence disease risk is generated, and the warning level is raised.

[0016] It should be noted that, after excluding environmental factors as the primary cause, this method employs a hierarchical decision-making mechanism based on the fusion of acoustic and visual dual-modal evidence to achieve accurate assessment of potential disease risks. This method identifies abnormal acoustic events in groups through in-depth analysis of audio data, while simultaneously capturing abnormal behavioral manifestations at the individual level through intelligent parsing of video data. This allows for the collection of evidence of disease signs from two independent yet complementary dimensions: "abnormal group sound" and "abnormal individual behavior." Its core innovation lies in setting differentiated judgment rules: when abnormal evidence in a single dimension (such as the frequency of abnormal sounds or the number of abnormal behaviors) exceeds a basic threshold, the system cautiously determines that there is a "potential disease risk," ensuring monitoring sensitivity and avoiding missed detections. Conversely, when abnormal evidence in both acoustic and visual dimensions simultaneously reaches higher thresholds, it is judged as a "high-confidence disease risk," and the warning level is upgraded. This significantly improves the specificity and reliability of the judgment through cross-validation of evidence, effectively reducing false alarms that may be caused by false alarms from a single sensor. This intelligent decision-making logic of multi-evidence fusion and graded response enables the system to output disease risk conclusions with different confidence levels based on the sufficiency and consistency of the evidence. This guides managers to take more targeted verification and intervention measures that match the risk level, optimizes the allocation of management resources, and improves the accuracy of early warning of epidemics.

[0017] Furthermore, the method also includes: Based on the abnormal type and spatial location of the environmental parameters that exceed the preset safety range, an adjustment command for the environmental control equipment is generated; The environmental control equipment adjustment command is sent to the environmental control system of the poultry house to automatically perform at least one of the following operations: ventilation, cooling, humidification, or light adjustment.

[0018] It should be noted that after the system determines that the abnormality is caused by environmental stress, this method does not stop at generating monitoring results, but further transforms this intelligent diagnostic conclusion into directly executable and precise environmental control commands, which are then automatically sent to the execution system. This design achieves a millisecond-level closed loop from "perception and analysis" to "decision execution." Its core value lies in the fact that when poultry flocks begin to exhibit early signs of stress, such as detectable aggregation and abnormal activity, due to environmental factors such as localized overheating and poor ventilation, the system can automatically and immediately initiate ventilation, cooling, and other regulatory measures in the corresponding areas based on accurate judgment of the abnormal parameter type and spatial location. This proactively and quickly corrects the unbalanced indoor environment before management personnel intervene or before the stress causes large-scale health damage. This not only quickly breaks the vicious cycle of declining animal welfare and productivity caused by environmental imbalances, minimizing potential losses, but also ensures the timeliness, accuracy, and consistency of intervention measures by replacing manual judgment and operation. It fundamentally improves the level of intelligence and stable control of large-scale poultry house environmental management, and is a key automatic protection mechanism to ensure that poultry flocks are in a healthy and comfortable environment in the long term.

[0019] Furthermore, the extraction of impedance spatiotemporal features characterizing the spatial distribution and dynamic activities of bird flocks from the sequence impedance distribution image includes: Perform time-frequency domain transformation on the sequence impedance distribution image within a preset time window, and calculate the impedance fluctuation energy in the preset frequency band; Areas where the impedance fluctuation energy is lower than the statistical energy benchmark of historical healthy periods and the area of ​​the spatially continuous region exceeds a preset proportion of the total projected area of ​​the current flock are identified as static gathering areas. The region where the impedance fluctuation energy is higher than the statistical energy benchmark and the overall regional center of gravity movement speed is lower than the preset movement speed benchmark value is identified as a high-activity low-movement region. The spatiotemporal distribution characteristics of the static aggregation area and the high-activity, low-mobility area are used as impedance spatiotemporal characteristics for judging abnormal group behavior.

[0020] It should be noted that this method, by introducing time-frequency domain analysis, extracts two core physical quantities from dynamic impedance distribution image sequences: "impedance fluctuation energy" and "regional center of gravity movement velocity." Based on this, it innovatively defines two types of characteristic regions with clear pathological or behavioral orientations: "static aggregation areas" and "high-activity, low-movement areas." This method goes beyond the traditional extraction of simple geometric or statistical features (such as area and center of gravity) from images. Instead, it starts from the essential nature of "energy" and "ordered movement" that inevitably accompany biological activity and can be captured by electrical signals, to deeply characterize the state of poultry flocks. By associating "static aggregation areas" with low fluctuation energy and large spatial proportion, it effectively identifies negative states such as huddling and inactivity caused by severe discomfort or disease in poultry flocks. Conversely, by combining "high-activity, low-movement areas" with high fluctuation energy and low mobility, it can capture the aimless and anxious activities of poultry flocks within a small area caused by restlessness, pain, or other reasons. The extraction of these two types of features enables the system to directly separate and quantify specific behavioral patterns that are highly correlated with health risks from complex impedance distribution dynamics. This provides highly discriminative and interpretable input features for subsequent anomaly judgment, thereby significantly improving the system's detection sensitivity and diagnostic accuracy for early, subtle, and patterned group behavioral anomalies.

[0021] This specification provides one or more embodiments of a poultry health status monitoring device, comprising: At least one processor and bus; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to: Control the electrical impedance imaging sensor array and multimodal acquisition device group deployed in the poultry house to synchronously collect monitoring data, and add timestamps and corresponding spatial location information to the monitoring data to obtain a spatiotemporally aligned raw multimodal data stream. The monitoring data includes at least electrical impedance imaging measurement voltage data, environmental parameter data, audio data and video data. Based on the original multimodal data stream, a multimodal feature vector of poultry is extracted. The multimodal feature vector of poultry includes: extracting spatiotemporal impedance features based on the electrical impedance imaging voltage measurement data, extracting environmental features based on the environmental parameter data, and extracting acoustic features and visual behavioral features based on the audio data and video data. The spatiotemporal impedance features are the impedance distribution image of the poultry house cross section reconstructed based on the electrical impedance imaging voltage measurement data, and the relevant features characterizing the spatial distribution and dynamic activities of the poultry flock are extracted from the sequence impedance distribution image. The poultry multimodal feature vectors are analyzed and compared with a pre-established health status benchmark pattern to determine whether the poultry flock's health status is abnormal. When an anomaly is detected, the monitoring results of poultry health status are generated by combining the correlation analysis results between the impedance spatiotemporal characteristics, the environmental characteristics, the acoustic characteristics and the visual behavioral characteristics.

[0022] This specification provides one or more embodiments of a non-volatile computer storage medium storing computer-executable instructions, which, when executed by a computer, can perform the following: Control the electrical impedance imaging sensor array and multimodal acquisition device group deployed in the poultry house to synchronously collect monitoring data, and add timestamps and corresponding spatial location information to the monitoring data to obtain a spatiotemporally aligned raw multimodal data stream. The monitoring data includes at least electrical impedance imaging measurement voltage data, environmental parameter data, audio data and video data. Based on the original multimodal data stream, a multimodal feature vector of poultry is extracted. The multimodal feature vector of poultry includes: extracting spatiotemporal impedance features based on the electrical impedance imaging voltage measurement data, extracting environmental features based on the environmental parameter data, and extracting acoustic features and visual behavioral features based on the audio data and video data. The spatiotemporal impedance features are the impedance distribution image of the poultry house cross section reconstructed based on the electrical impedance imaging voltage measurement data, and the relevant features characterizing the spatial distribution and dynamic activities of the poultry flock are extracted from the sequence impedance distribution image. The poultry multimodal feature vectors are analyzed and compared with a pre-established health status benchmark pattern to determine whether the poultry flock's health status is abnormal. When an anomaly is detected, the monitoring results of poultry health status are generated by combining the correlation analysis results between the impedance spatiotemporal characteristics, the environmental characteristics, the acoustic characteristics and the visual behavioral characteristics.

[0023] The above-described at least one technical solution adopted in the embodiments of this specification can achieve the following beneficial effects: This method simultaneously acquires and spatiotemporally aligns multimodal data, including impedance imaging, environmental data, audio, and video. First, it constructs a non-invasive, penetrating impedance distribution image and its dynamic characteristics that reflect the internal spatial distribution and collective activity state of poultry flocks. This overcomes the limitations of traditional visual monitoring, such as susceptibility to occlusion and difficulty in quantifying internal states, laying a data foundation for early anomaly detection at the flock level. Furthermore, by intelligently comparing the extracted multimodal features with pre-established health benchmark patterns, the system can keenly identify subtle deviations in flock behavior patterns, issuing early warnings before individual clinical manifestations appear. Crucially, upon detecting an anomaly, the method does not make isolated judgments but further integrates and analyzes the correlation between impedance anomaly patterns and environmental parameters, audio-visual features, thereby effectively distinguishing whether the root cause of the anomaly lies in environmental stress or potential disease risk. This allows poultry farmers to receive not just a single "abnormal alarm," but a "monitoring result" that includes possible causes, enabling them to take targeted measures (such as adjusting environmental control or conducting disease screening). This significantly advances the intervention window, fundamentally changing the traditional, lagging, and passive management model that relies on manual inspections and individual sampling. It significantly improves the predictability, accuracy, and proactive defense capabilities of large-scale poultry flock health management, playing a substantial role in reducing the risk of disease spread, minimizing economic losses, and protecting animal welfare. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments or prior art of this specification, the drawings used in the description of the embodiments or prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings: Figure 1 A flowchart illustrating a method for monitoring the health status of poultry provided in one or more embodiments of this specification; Figure 2 This is a structural schematic diagram of a poultry health status monitoring device provided for one or more embodiments of this specification. Detailed Implementation

[0025] This specification provides a method, device, and medium for monitoring the health status of poultry.

[0026] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0027] Figure 1 This diagram illustrates a process for monitoring poultry health status according to one or more embodiments of this specification. The process can be executed by a data processing system deployed within a poultry house. Certain input parameters or intermediate results in the process can be manually adjusted to help improve accuracy.

[0028] The method flow steps of the embodiments in this specification are as follows: S101, control the electrical impedance imaging sensor array and multimodal acquisition device group deployed in the poultry house to synchronously acquire monitoring data, and add timestamps and corresponding spatial location information to the monitoring data to obtain a spatiotemporally aligned raw multimodal data stream. The monitoring data includes at least electrical impedance imaging measurement voltage data, environmental parameter data, audio data and video data.

[0029] In the embodiments described in this specification, a set of electrodes is uniformly embedded circumferentially at a suitable height from the ground along the perimeter of the inner wall of the poultry house, forming an impedance imaging sensor array. A multimodal acquisition device group, including microphones, network cameras, and sensors for temperature, humidity, ammonia, and carbon dioxide, is distributed throughout the house. The data processing system sends a synchronization clock signal to all acquisition units, controlling them to begin acquisition at the same time.

[0030] For electrical impedance tomography (EIT) voltage data acquisition, the system selects one pair of electrodes in the array at a time in a preset sequence (such as adjacent mode) and injects a safe micro-AC current. Simultaneously, it rapidly measures the boundary voltages between all other electrode pairs. After completing the measurement of all excitation pair combinations, a complete voltage dataset is obtained.

[0031] For multimodal data acquisition, the camera is triggered to capture images, the microphone to record audio clips, and various environmental sensors to read current parameter values ​​simultaneously.

[0032] The system adds a unique timestamp to all data collected at this moment (voltage datasets, images, audio clips, and environmental parameter values). Simultaneously, spatial location tags are added to the data based on the preset installation coordinates of each device within the poultry house. All data with spatiotemporal tags is packaged into a single data packet, forming a spatiotemporally aligned raw multimodal data stream.

[0033] S102, based on the original multimodal data stream, extract poultry multimodal feature vectors. The poultry multimodal feature vectors include: extracting impedance spatiotemporal features based on the impedance imaging measurement voltage data, extracting environmental features based on the environmental parameter data, and extracting acoustic features and visual behavior features based on the audio data and video data. The impedance spatiotemporal features are the impedance distribution image of the poultry house cross section reconstructed based on the impedance imaging measurement voltage data, and the relevant features characterizing the spatial distribution and dynamic activities of the poultry flock are extracted from the sequence impedance distribution image.

[0034] In the embodiments described in this specification, a pre-established finite element model conforming to the geometry of the poultry house cross-section and a calculated sensitivity matrix are used. The current voltage dataset obtained in S101 is read, and the average voltage data from historical healthy periods stored in the system is used as a reference voltage. A dynamic image reconstruction algorithm (such as the time difference method) is employed to solve for the change of the current data relative to the reference state, and the impedance distribution image of the poultry house cross-section (essentially a relative conductivity change distribution map) is calculated.

[0035] The impedance distribution images of a sequence within a continuous time window are analyzed. By calculation, parameters such as the area of ​​high conductivity regions, the position of the center of gravity, the movement speed, and the image entropy are extracted to form the spatiotemporal characteristics of impedance, which are used to quantify the spatial distribution and dynamic activities of the poultry flock.

[0036] The environmental parameter data collected by S101 is preprocessed (e.g., filtered), and its values ​​or calculated statistics (e.g., gradients) are directly used as environmental features. Spectral analysis is performed on the audio data collected by S101 to identify specific frequency band energy or specific sound events (e.g., coughing, startling), and their frequencies are statistically analyzed as acoustic features. For the video data collected by S101, a target detection model is used to identify poultry, and group movement, distribution density, or abnormal behaviors (e.g., drooping wings, huddling) are analyzed using posture estimation or optical flow methods, and quantified as visual behavioral features. The impedance spatiotemporal features, environmental features, acoustic features, and visual behavioral features extracted at the same time are combined to construct the poultry multimodal feature vector for that time moment.

[0037] Electrical impedance tomography (EIT) can reconstruct the conductivity distribution across a cross-section within a poultry house. When poultry breathe, changes in lung air content lead to regular alterations in conductivity within the thoracic cavity. By analyzing image sequences, the collective respiratory rate and patterns of poultry throughout the entire area can be monitored non-contactly. This can provide early warning of respiratory diseases (such as Newcastle disease and infectious bronchitis) or environmental stress (such as rapid breathing caused by excessive ammonia concentration). Furthermore, poultry activity, movement, and aggregation cause changes in tissue location and density, thus altering the EIT images. This allows for inference of flock density distribution, activity hotspots, and resting states.

[0038] Medical electrical impedance imaging focuses on a single organism (such as a patient) to examine the tissue characteristics of specific organs (such as the lungs or brain) for the diagnosis of individual diseases. In contrast, poultry house electrical impedance imaging examines a cross-section of a living group. The goal is no longer to see the lungs of a single chicken, but rather to view the changes in conductivity distribution throughout the space as a macroscopic signal reflecting the collective behavior and state of the flock.

[0039] In medicine, image changes correspond to individual physiological activities (such as changes in lung conductivity caused by respiration). In poultry houses, image changes are interpreted as indicators of flock behavior: periodic changes are interpreted as synchronized respiratory rates across the flock, enabling non-contact monitoring of flock respiratory health. Non-periodic changes are interpreted as flock movement, aggregation, or stillness, allowing analysis of density distribution and activity patterns.

[0040] Meanwhile, medical electrical impedance tomography (EIT) is typically a standalone, dedicated diagnostic device. Poultry house EIT is designed as a core sensing unit within a large-scale, multimodal farm Internet of Things (IoT). It works synchronously with microphones, cameras, and gas sensors to achieve data fusion. Its purpose transcends mere "diagnosis," elevating it to the level of intelligent livestock management: early disease warning (e.g., detecting Newcastle disease through abnormal breathing), assessing animal welfare (e.g., judging environmental stress through activity levels), and optimizing production management (e.g., understanding density distribution).

[0041] Furthermore, electrical impedance tomography (EIT) is a versatile physical imaging method, the most mature and commercially successful in the medical field, including: 1. Lung monitoring, real-time display of lung ventilation / perfusion distribution, and rapid detection of atelectasis, pneumothorax, and pulmonary edema. 2. Breast / superficial tumor screening, used for screening lesions in superficial organs such as the breast, thyroid, and lymph nodes, as a supplement to mammography / MRI. 3. Brain function monitoring, monitoring of cerebral ischemia, hemorrhage, edema, and hypoxic-ischemic encephalopathy (HIE). 4. Cardiac electrical activity / perfusion, abdominal organ function, burn depth assessment, and intraoperative tissue viability monitoring, etc.

[0042] Electrical impedance imaging is also widely used in industrial, testing, and environmental protection fields, including: 1. Industrial non-destructive testing, such as detecting internal voids and cracks in concrete, detecting leaks in walls and dams, detecting internal defects in composite materials and carbon fibers, and detecting internal corrosion and blockages in pipes and tanks; 2. Geological engineering, such as groundwater level monitoring, soil moisture distribution, landslide and geological structure detection, and imaging of polluted areas (heavy metals, pollutants); 3. Biological and agricultural fields, such as internal quality testing of fruits and vegetables (sugar content, maturity, hollowness, rot), imaging of plant roots, and detection of internal defects and foreign objects in food; 4. Environmental protection and process monitoring, such as two-phase flow and bubble distribution in wastewater treatment, material distribution in chemical reactors, and visualization of fluidized beds and gas-liquid two-phase flow; 5. Safety inspection, such as detection of hazardous materials and hidden objects, and imaging of the internal structure of enclosed containers.

[0043] Electrical impedance imaging is essentially using "electricity" to draw a diagram of the internal structure of an object. The core principle is that different materials have different impedances → external voltage is measured → the internal impedance distribution is deduced → and an image is drawn.

[0044] Different substances / tissues have different electrical impedances. For example: air has very high impedance, water has low impedance, metals have extremely low impedance, and fat, muscle, and tumors all have different impedances. The impedance distribution will change as long as the internal composition changes.

[0045] This application embodiment uses a multi-electrode array arranged on the poultry house floor / fence to collect impedance distribution maps in real time, reconstruct animal positions, activity levels, postures, and distribution densities, and compares them with normal models to identify whether poultry activity levels suddenly drop significantly, whether they lie down for a long time, whether they huddle together abnormally when separated from the flock, whether poultry have respiratory micro-movements, and whether they remain stationary for a long time (death), thus determining the health status of the poultry group.

[0046] S103, the poultry multimodal feature vectors are analyzed, and the health status of the flock is judged to be abnormal by comparing them with the pre-established health status benchmark model.

[0047] In the embodiments of this specification, the system invokes a pre-established health status benchmark model. This model is typically established during the system learning phase using machine learning methods such as cluster analysis on multimodal feature vectors of healthy poultry collected over a long period, describing the characteristic range of normal status at different time periods (e.g., morning feeding, afternoon rest). The multimodal feature vector of poultry generated in S102 at the current time is compared with the health status benchmark model corresponding to the current time period. The deviation between the two is calculated using a specific algorithm.

[0048] Set a dynamic deviation threshold and duration window. If the deviation of the current feature continues to exceed the threshold for a period of time, the flock's health status is determined to be abnormal.

[0049] S104, when an anomaly is determined, the monitoring results of poultry health status are generated by combining the correlation analysis results between the impedance spatiotemporal characteristics, the environmental characteristics, the acoustic characteristics and the visual behavioral characteristics.

[0050] In the embodiments of this specification, when an anomaly is determined in S103, the system first analyzes which impedance spatiotemporal characteristics (such as the appearance of a large area of ​​stillness) caused the deviation. The system retrieves environmental characteristics that match the spatial region indicated by the abnormal impedance spatiotemporal characteristics to determine whether parameters such as temperature or harmful gases exceed the standard in that region.

[0051] If the associated environmental parameters exceed the standard, a monitoring result indicating "environmental stress" as the primary cause is generated. If the environmental parameters are normal, acoustic and visual behavioral characteristics under the same time and space are further examined. If significant abnormal sounds or abnormal individual behavior are detected, a monitoring result indicating "potential disease risk" is generated.

[0052] The monitoring results (including anomaly type, possible cause, location, time, etc.) are output to the monitoring screen or management terminal and can trigger corresponding early warning notifications.

[0053] It should be noted that this method, by simultaneously acquiring and spatiotemporally aligning multimodal data such as impedance imaging, environmental data, audio, and video, first constructs an impedance distribution image and its dynamic characteristics that non-invasively and penetratingly reflects the internal spatial distribution and collective activity state of poultry flocks. This overcomes the limitations of traditional visual monitoring, which is easily obstructed and has difficulty in quantifying internal states, laying a data foundation for early anomaly detection at the flock level. Furthermore, by intelligently comparing the extracted multimodal features with pre-established health benchmark patterns, the system can keenly identify subtle deviations in flock behavior patterns and issue early warnings before individual clinical manifestations appear. Crucially, when an anomaly is detected, the method does not make isolated judgments but further integrates and analyzes the correlation between impedance anomaly patterns and environmental parameters, audio-visual features, thereby effectively distinguishing whether the root cause of the anomaly lies in environmental stress or potential disease risk. This allows poultry farmers to receive not just a single "abnormal alarm," but a "monitoring result" that includes possible causes, enabling them to take targeted measures (such as adjusting environmental control or conducting disease screening). This significantly advances the intervention window, fundamentally changing the traditional, lagging, and passive management model that relies on manual inspections and individual sampling. It significantly improves the predictability, accuracy, and proactive defense capabilities of large-scale poultry flock health management, playing a substantial role in reducing the risk of disease spread, minimizing economic losses, and protecting animal welfare.

[0054] Furthermore, the electrical impedance imaging sensor array includes multiple electrodes evenly arranged circumferentially along the inner wall of the poultry house; the method flow steps of this embodiment of the specification are as follows during the process of controlling the electrical impedance imaging sensor array deployed in the poultry house to synchronously acquire the electrical impedance imaging measurement voltage data: S201, according to the preset excitation mode, inject safe AC excitation current into the selected pair of electrodes in sequence.

[0055] In the embodiments described in this specification, the system selects one electrode pair from multiple electrodes of the electrical impedance imaging sensor array as the "drive-receive" pair for this excitation, based on a preset excitation mode (such as an adjacent excitation mode). The system can inject an AC excitation current with a safely set frequency and amplitude into the selected electrode pair using a constant current source. This current is weak and will not harm the flock; its frequency is typically selected in a specific frequency band that has good penetration into biological tissues and avoids major power frequency interference.

[0056] S202, synchronously measure the boundary voltage between the remaining multiple electrode pairs in the electrical impedance imaging sensor array to obtain a voltage dataset for image reconstruction, thereby completing one full data acquisition cycle.

[0057] In the embodiments described in this specification, at the same moment as the current is injected in step S201, the system simultaneously measures the potential difference, i.e., the boundary voltage, between multiple electrode pairs in the sensor array other than the currently excited electrode pair through a multi-channel synchronous acquisition circuit. All boundary voltage values ​​obtained in this measurement are recorded together with the electrode pair numbering information of this excitation.

[0058] The completion of steps S201 and S202 marks the end of a single measurement under a pair of electrode excitations. The system repeats steps S201 and S202, changing the excitation electrode pairs according to the preset excitation modes, until all preset excitation combinations have been completed once. Finally, all boundary voltages acquired throughout the cycle are summarized to form a complete voltage dataset for subsequent image reconstruction, marking the completion of one full data acquisition cycle.

[0059] It should be noted that this method establishes a stable and reliable physical and data foundation for the entire monitoring method by clearly defining the configuration of the impedance imaging sensor array uniformly arranged along the circumference of the poultry house's inner wall, and the standard data acquisition process of "sequentially injecting excitation and synchronously measuring voltage." This design first ensures that the excitation current can form a stable and controllable electric field distribution across the cross-section of the poultry house, making subsequent measurements repeatable and comparable. The "synchronous measurement" operation ensures that the boundary voltage data acquired under one excitation corresponds to the poultry flock state at the same moment, effectively avoiding data timing misalignment caused by flock activity, which is crucial for capturing dynamic changes and using them for image reconstruction. The resulting complete and spatiotemporally synchronized "voltage dataset" is the necessary raw input for subsequent high-precision impedance distribution image reconstruction. The standardized data collection paradigm established in this step ensures the quality and consistency of the original data from the source, enabling the impedance distribution image reconstructed based on this data to more realistically and stably reflect the spatial distribution and activity status within the poultry flock. This provides a reliable data starting point for all advanced intelligent judgments such as subsequent feature extraction, pattern comparison, and correlation analysis. It is a prerequisite and technical guarantee for the entire system to achieve continuous, stable, and automated operation and generate accurate early warning results.

[0060] Furthermore, in the process of reconstructing the impedance distribution image of the poultry house cross-section based on the impedance imaging measurement voltage data, the method flow steps of this embodiment are as follows: S301, Establish a finite element calculation model corresponding to the cross-sectional geometry of the poultry house.

[0061] In the embodiments of this specification, a corresponding two-dimensional (or three-dimensional) geometric model is established in a computer based on the actual cross-sectional geometry, dimensions, and layout of the main internal fixed facilities (such as feed lines, water lines, and supports) of the target poultry house. Using finite element analysis software, this geometric model is discretized into a collection of numerous small, interconnected basic units (such as triangular or quadrilateral units), forming a finite element calculation model of the poultry house cross-section. This model defines the solution domain for image reconstruction.

[0062] S302, Calculate the sensitivity matrix based on the finite element calculation model.

[0063] In this embodiment of the specification, based on the finite element calculation model established in S301, a uniform background conductivity is set as the initial distribution. Electromagnetic field simulation calculations determine the electric field distribution of each element within the model when excitation is applied to any pair of electrodes. Based on field distribution theory and the chain rule, the influence coefficient of a small change in the conductivity of each element on all possible measurable changes in boundary voltage is calculated. All coefficients are arranged into a matrix, which yields the sensitivity matrix. This matrix establishes a linear approximate relationship between internal conductivity perturbations and boundary voltage changes.

[0064] S303 uses the statistical value of the voltage data from historical healthy periods of electrical impedance imaging as the reference voltage.

[0065] In the embodiments described in this specification, the system retrieves and stores a large amount of electrical impedance tomography (EIT) voltage data collected over a pre-set period during which the poultry flock is confirmed to be healthy. Statistical analysis (such as calculating the average value) is performed on these historical health voltage data to obtain a stable voltage baseline value, which serves as a reference voltage. This reference voltage represents the typical background signal measured by the electrode array under normal conditions in the poultry house.

[0066] S304. Based on the current impedance imaging measurement voltage data, the reference voltage, and the sensitivity matrix, a dynamic image reconstruction algorithm is used to solve for the conductivity change distribution map relative to the reference state at the current moment, which is used as the impedance distribution image.

[0067] In the embodiments described in this specification, the voltage data measured by impedance imaging at the current moment is acquired. The difference between the current voltage data and the reference voltage obtained in S303 is used to obtain the voltage change. Using the sensitivity matrix calculated in S302 as the system matrix, and taking the aforementioned voltage change as a known quantity, a linear inverse problem is constructed.

[0068] This inverse problem can be solved using dynamic image reconstruction algorithms specifically designed for handling dynamic targets (such as temporal difference imaging algorithms). The goal of the algorithm is to find an internal conductivity variation distribution such that the voltage change predicted by the sensitivity matrix from this distribution best matches the actual measured voltage change.

[0069] The result obtained by the algorithm is a two-dimensional (or three-dimensional) conductivity variation distribution map. This map visually shows the relative change of conductivity at each point within the cross-section of the poultry house relative to the historical health reference state. This map is defined as an impedance distribution image, where bright (high conductivity) areas usually correspond to areas where poultry bodies are clustered.

[0070] It should be noted that this method, by establishing a finite element calculation model that strictly corresponds to the actual cross-sectional geometry of the poultry house, first provides a precise mathematical foundation for image reconstruction that conforms to the physical scene, ensuring the accuracy of solving both forward and inverse problems. Furthermore, by calculating the sensitivity matrix, the influence of internal conductivity changes on boundary measurements is quantified, establishing a reliable transformation bridge for inferring the internal state from the measured voltage. Crucially, the method innovatively introduces statistical values ​​of measurement data from "historical healthy periods" as a dynamic reference voltage benchmark. This effectively eliminates background signal interference caused by static factors such as the fixed structure and equipment layout of the poultry house, allowing the reconstruction target to focus on the relative conductivity changes caused by changes in the poultry flock's own state. Based on this, by combining real-time measurement data and the sensitivity matrix at the current moment, and using an image reconstruction algorithm specifically designed for dynamic targets, a stable and accurate impedance distribution image, characterized by relative changes, reflecting the spatial distribution and activity state of the poultry flock within the cross-section of the poultry house, can be generated. As the source of subsequent feature extraction, the accuracy and stability of this image directly determine the sensitivity and reliability of the entire monitoring system in sensing abnormal states within the flock, thus providing an irreplaceable core data representation for early detection of deviations in flock behavior patterns and achieving accurate early warning.

[0071] Furthermore, in the process of determining whether the health status of the poultry flock is abnormal, the method flow steps of this embodiment are as follows: S401, calculate the deviation between the poultry multimodal feature vector at the current moment and the health status benchmark mode for the corresponding time period.

[0072] In the embodiments described in this specification, the system acquires the multimodal feature vector of poultry at the current moment, generated by the upstream feature extraction module. Simultaneously, based on the current time (e.g., morning feeding time), it retrieves a pre-trained health status baseline model from the database. This baseline model defines the typical range or distribution of each feature value under normal conditions during that time period.

[0073] The system employs a pre-defined similarity or distance metric algorithm to perform a dimensional or multi-dimensional joint comparison between the current feature vector and the baseline pattern. This calculation outputs a comprehensive scalar value, the deviation, which quantifies the overall degree of difference between the current state and the normal baseline. A larger deviation value indicates a higher probability of an anomaly.

[0074] S402, if the deviation exceeds the dynamic threshold and the duration of the abnormal state exceeds the preset time window, then the health status of the poultry flock is determined to be abnormal.

[0075] In the embodiments described in this specification, the system compares the deviation calculated in S401 with a dynamic threshold. This dynamic threshold is adaptively adjusted according to different time periods, seasons, or the age of the flock, rather than being a fixed value.

[0076] The system not only checks the deviation at the current moment, but also whether the deviation is continuous. It retrieves the continuous deviation calculation results within a recent period (i.e., a preset time window, such as the past 30 minutes) and determines whether they all exceed the dynamic threshold.

[0077] The system will only determine that the flock's health is abnormal if both conditions are met simultaneously: "the current deviation exceeds the dynamic threshold" and "the abnormal state persists within a preset time window." If the deviation is only momentary and then recovers, it will be considered an occasional fluctuation and no alarm will be triggered.

[0078] It should be noted that this method achieves an objective and digital assessment of poultry health status by quantitatively comparing the real-time extracted multimodal feature vectors of poultry with the corresponding time-period health status benchmark pattern and calculating the deviation, replacing the traditional manual judgment relying on subjective experience. Specifically, by setting a "dynamic threshold" instead of a fixed threshold, the system can adapt to the natural fluctuation range of normal poultry activity at different times (such as feeding and resting periods), improving the flexibility and accuracy of the judgment. More importantly, the introduction of a "duration exceeding a preset time window" criterion effectively filters out accidental and transient behavioral abnormalities caused by brief frights, temporary disturbances, etc., ensuring that the system only issues alarms for stable and continuous abnormal patterns. This dual criterion combining "deviation" and "persistence" greatly reduces the false alarm rate, making the final "abnormal poultry health status" judgment highly reliable. This ensures that managers can respond promptly to real and potential health risks, avoiding decreased trust and wasted resources due to frequent or inaccurate alarms, and enhancing the practicality and authority of the entire monitoring system.

[0079] Furthermore, in the process of generating monitoring results for poultry health status by combining the correlation analysis results between the impedance spatiotemporal characteristics, the environmental characteristics, the acoustic characteristics, and the visual behavioral characteristics, the method flow steps of this embodiment are as follows: S501, the specific pattern of the impedance spatiotemporal characteristics on which the anomaly is determined.

[0080] In the embodiments described in this specification, the system retrieves the spatiotemporal characteristics of impedance at the current moment. These characteristics are analyzed to identify the specific patterns they exhibit. For example, it determines whether the anomaly is primarily manifested as a large-scale "static clustering area" or a small-scale, high-frequency "high-activity, low-movement area," and precisely locates the anomalous spatial region within the cross-section of the poultry house where these anomalous patterns are situated.

[0081] S502, retrieve the environmental features associated with the abnormal spatial region indicated by the specific pattern, and determine whether there are environmental parameters that exceed the preset safety range.

[0082] In the embodiments of this specification, based on the abnormal spatial area determined in S501, the system retrieves environmental characteristic data (such as temperature and ammonia concentration) collected by environmental sensors installed in or representing that area during the same time period from a database or real-time stream. The retrieved environmental parameter values ​​are then compared one by one with the preset safety ranges for various parameters in the system (such as upper temperature limits and upper limits for harmful gas concentrations). It is determined whether at least one environmental parameter exceeds the preset safety range.

[0083] S503, if present, generates monitoring results indicating that environmental stress is the primary cause of the abnormality.

[0084] In the embodiments of this specification, if the judgment result of S502 is "existence" of an environmental parameter exceeding the standard, the system infers that the current abnormal behavior of the poultry flock is mainly caused by external environmental discomfort. The system generates a clear monitoring result, in which the core conclusion indicates "environmental stress", and the result usually includes the abnormal type of the environmental parameter exceeding the standard, the degree of exceeding the standard, and the spatial location of the abnormality.

[0085] S504, if not, then combine the acoustic features with the visual behavioral features. If the acoustic features or visual behavioral features indicate the presence of abnormal individual behavior, then generate a monitoring result indicating potential disease risk.

[0086] In this embodiment of the specification, if the judgment result of S502 is "no" environmental parameters exceeding the standard, the system then switches to analyzing the internal state of the flock itself. The system retrieves acoustic and visual behavioral characteristics that occurred at the same time and space as the anomaly described in S501. It checks whether abnormal sound events (such as frequent coughing) are marked in the acoustic characteristics, or whether abnormal individual behaviors (such as drooping wings or standing still) are marked in the visual behavioral characteristics.

[0087] If at least one of the aforementioned acoustic or visual behavioral characteristics clearly indicates the presence of abnormal individual behavior, the system generates another clear monitoring result, with the core conclusion indicating "potential disease risk".

[0088] It should be noted that, based on the determination of abnormal poultry flock conditions, the introduction of a progressive attribution analysis logic based on multimodal feature association significantly enhances the decision support value of monitoring results. This method first analyzes the core electrical characteristics of the abnormality and the specific patterns of impedance spatiotemporal features, thereby pinpointing the nature and spatial location of the abnormality. Subsequently, instead of drawing conclusions in isolation, it prioritizes checking the environmental characteristics of the abnormal area. If environmental parameters are found to be excessive, the root cause of the abnormality is directly attributed to immediately intervened "environmental stress," enabling managers to adjust environmental control equipment immediately and quickly resolve the problem at its source. If environmental parameters are normal, it means the abnormality is more likely to originate from the flock itself; in this case, acoustic and visual behavioral characteristics are further integrated for corroboration. When these externally perceived characteristics also indicate the presence of abnormal individual behavior, the results can be clearly pointed to "potential disease risk," thereby triggering different biosafety and veterinary inspection response procedures. This "inside-out, environment-first-individual" correlation analysis and attribution mechanism means that the system outputs not just generalized "abnormal alerts," but also specific "diagnostic conclusions." This fundamentally elevates monitoring from the "perception" level to the "cognition" level, greatly enhancing the pertinence and timeliness of management responses. It avoids resource misallocation or delays caused by misjudgments and provides direct intelligent decision-making basis for achieving precise and efficient poultry health management.

[0089] Furthermore, in the process of combining the acoustic features with the visual behavioral features, the method flow steps of this embodiment are as follows: S601, when the environmental characteristics do not exceed the safety range, perform spectrum analysis and event detection on the audio data to identify specific abnormal acoustic events and their frequencies.

[0090] In the embodiments described in this specification, the system acquires raw audio data synchronously collected at the current anomaly determination time. Spectral analysis is performed on the audio data, decomposing it into different frequency components. Based on a preset acoustic feature model (such as a voiceprint template for coughs, wheezing, or whimpering sounds related to a specific disease), the occurrence of specific abnormal acoustic events is identified in the spectrum. The frequency of all abnormal acoustic events identified within an analysis time window is counted.

[0091] The acoustic feature model can be obtained by training a machine learning model (such as a convolutional neural network (CNN) or a support vector machine (SVM)) on a large amount of labeled poultry house audio data, which can effectively distinguish specific abnormal acoustic events from background noise. Spectral analysis can extract features using short-time Fourier transform (STFT) or Mel-frequency cepstral coefficients (MFCC). Event detection not only counts the frequency, but also records the event type, average intensity, and temporal distribution pattern as richer acoustic features.

[0092] S602, Apply a target detection and pose estimation model to the video data to identify specific abnormal individual behaviors and their number.

[0093] In the embodiments described in this specification, the system acquires raw video data synchronously collected at the current anomaly determination time. A target detection model is applied to the video frame sequence to locate individual birds, and then a pose estimation model is applied to analyze the pose of key body parts of each individual. Based on predefined rules or a trained behavior classification model, the system determines whether an individual exhibits specific abnormal behavior (such as drooping wings, head lowered, or unsteady gait) according to the pose sequence. The number of individuals exhibiting abnormal behavior within the current field of view is counted.

[0094] The target detection model can employ networks such as YOLO or Faster R-CNN; the pose estimation model can use OpenPose or a top-down approach based on deep learning. The behavior classification model can be a Temporal Convolutional Network (TCN) or a Long Short-Term Memory Network (LSTM) to analyze continuous sequences of pose keypoints to identify complex abnormal behavior patterns such as "persistent wing drooping for more than N seconds" or "periodic occurrence of gait instability." The system can assign different weights to different abnormal behaviors.

[0095] S603, if the frequency of the abnormal acoustic events exceeds a first frequency threshold, or the number of abnormal individual behaviors exceeds a first quantity threshold, then a monitoring result indicating potential disease risk is generated.

[0096] In the embodiments of this specification, the frequency of abnormal acoustic events obtained in S601 is compared with a first frequency threshold, and the number of abnormal individual behaviors obtained in S602 is compared with a first quantity threshold. An "OR" logic is used. As long as either the frequency exceeds the first frequency threshold or the quantity exceeds the first quantity threshold, the system generates a monitoring result indicating a potential disease risk. This result signifies that signs of disease have been detected and require attention.

[0097] The first frequency threshold, the first quantity threshold, the second frequency threshold, and the second quantity threshold can all be learned from historical health data or preset by expert experience based on poultry breed, age, and breeding stage. The "OR" logic ensures high monitoring sensitivity, aiming to detect potential risks as early as possible; while the "AND" logic, by requiring acoustic and visual evidence to simultaneously meet higher standards, ensures the specificity of high-confidence judgments and effectively reduces the false alarm rate. The two levels of thresholds together constitute a progressive risk confirmation mechanism.

[0098] S604, if the frequency of the abnormal acoustic events exceeds the second frequency threshold and the number of abnormal individual behaviors exceeds the second quantity threshold, then a monitoring result indicating a high-confidence disease risk is generated, and the warning level is raised.

[0099] In the embodiments of this specification, the frequency obtained in S601 is compared with a higher second frequency threshold, and the quantity obtained in S602 is compared with a higher second quantity threshold.

[0100] The system employs AND logic. Only when both the frequency and the quantity exceed the second frequency threshold are met simultaneously will the system generate a monitoring result indicating a high-confidence disease risk.

[0101] At the same time, the system automatically upgrades the warning level (such as marking it as a red high-risk alert on the monitoring interface, or pushing a more urgent notification).

[0102] The analyses in S601 and S602 can be executed in parallel to improve processing efficiency. The analysis time window, the strategy for escalating the alert level (e.g., from yellow alert to red alert), and the corresponding notification methods (e.g., interface pop-ups, SMS, APP push notifications) can all be configured by the user according to actual management needs. When a "high confidence disease risk" result is generated, the system can automatically associate and retrieve audio and video clips from the abnormal time period and area for management personnel to review.

[0103] It should be noted that, after excluding environmental factors as the primary cause, this method employs a hierarchical decision-making mechanism based on the fusion of acoustic and visual dual-modal evidence to achieve accurate assessment of potential disease risks. This method identifies abnormal acoustic events in groups through in-depth analysis of audio data, while simultaneously capturing abnormal behavioral manifestations at the individual level through intelligent parsing of video data. This allows for the collection of evidence of disease signs from two independent yet complementary dimensions: "abnormal group sound" and "abnormal individual behavior." Its core innovation lies in setting differentiated judgment rules: when abnormal evidence in a single dimension (such as the frequency of abnormal sounds or the number of abnormal behaviors) exceeds a basic threshold, the system cautiously determines that there is a "potential disease risk," ensuring monitoring sensitivity and avoiding missed detections. Conversely, when abnormal evidence in both acoustic and visual dimensions simultaneously reaches higher thresholds, it is judged as a "high-confidence disease risk," and the warning level is upgraded. This significantly improves the specificity and reliability of the judgment through cross-validation of evidence, effectively reducing false alarms that may be caused by false alarms from a single sensor. This intelligent decision-making logic of multi-evidence fusion and graded response enables the system to output disease risk conclusions with different confidence levels based on the sufficiency and consistency of the evidence. This guides managers to take more targeted verification and intervention measures that match the risk level, optimizes the allocation of management resources, and improves the accuracy of early warning of epidemics.

[0104] Furthermore, after the system determines that the anomaly is caused by environmental stress, it does not stop at generating monitoring results, but further transforms this intelligent diagnostic conclusion into directly executable and precise environmental control commands, and automatically sends them to the execution system. The method flow steps of this embodiment are as follows: S701, based on the abnormal type and abnormal spatial location of the environmental parameters that exceed the preset safety range, generate an environmental control device adjustment command.

[0105] In the embodiments of this specification, the system receives input from step S503, and identifies the known "abnormal type of environmental parameters exceeding the preset safety range" (such as excessively high temperature or excessive ammonia concentration) and the "abnormal spatial location" where it occurs (such as the southeast corner of the poultry house).

[0106] The system uses built-in expert rules or control strategy mapping tables to convert specific abnormal parameters, abnormal types, and locations into adjustment instructions for one or more environmental control devices. For example, for "excessive temperature in the southeast corner," the generated instruction might be "turn on the backup fan in the southeast area, fan speed level three"; for "excessive ammonia concentration in the central area," the instruction might be "increase the opening of the central ventilation window to 70%."

[0107] The above adjustment logic is encapsulated into a standardized environmental control device adjustment instruction data packet that can be recognized by downstream hardware systems.

[0108] S702, the environmental control device adjustment command is sent to the environmental control system of the poultry house to automatically perform at least one of the following operations: ventilation, cooling, humidification or light adjustment.

[0109] In the embodiments described in this specification, the system transmits the environmental control device adjustment command data packets generated by S701 to the poultry house's environmental control system in real time via a standard industrial communication protocol. This system is typically a programmable logic controller or a smart gateway.

[0110] After receiving an instruction, the environmental control system of the poultry house drives the corresponding actuators (such as variable frequency fans, wet curtain water pumps, ventilation window motors, and lighting controllers) to perform actions, thereby automatically executing one or more combined operations of ventilation, cooling, humidification, or light adjustment required by the instruction.

[0111] It's important to note that this design achieves a millisecond-level closed loop from "perception and analysis" to "decision execution." Its core value lies in its ability to automatically and instantly activate ventilation and cooling measures in corresponding areas when poultry exhibit early signs of stress such as agglomeration and abnormal activity due to environmental factors like localized overheating or poor ventilation. This is based on precise judgment of the abnormal parameters, types, and spatial locations of the abnormalities. This proactively and quickly corrects the imbalanced environment within the poultry house before management intervention or before the stress causes widespread health damage. This not only rapidly breaks the vicious cycle of declining animal welfare and productivity caused by environmental imbalances, minimizing potential losses, but also ensures the timeliness, accuracy, and consistency of intervention responses by replacing manual judgment and operation. It fundamentally improves the intelligence level and stable control capabilities of large-scale poultry house environmental management, serving as a key automatic protection mechanism to ensure poultry flocks maintain a healthy and comfortable environment in the long term.

[0112] Furthermore, in the process of extracting the spatiotemporal impedance features characterizing the spatial distribution and dynamic activities of bird flocks from the sequence impedance distribution image, the method flow steps of this embodiment are as follows: S801 performs time-frequency domain transformation on the sequence impedance distribution image within a preset time window and calculates the impedance fluctuation energy in the preset frequency band.

[0113] In the embodiments described in this specification, the system acquires and reconstructs a series of impedance distribution images continuously collected within a preset time window. For the signal representing the impedance value of each spatial unit (pixel or finite element mesh) in the image sequence as a function of time, a time-frequency domain transformation (such as a short-time Fourier transform) is performed, converting it from the time domain to a frequency domain representation. In the transformed frequency domain representation, the fluctuation intensity of each spatial unit within a preset frequency band (typically corresponding to the characteristic frequency range of normal physiological activities in poultry flocks) is calculated, obtaining the impedance fluctuation energy value of that unit. Finally, an energy value is calculated for each unit in the entire image region, forming an "energy distribution map".

[0114] S802, the area where the impedance fluctuation energy is lower than the statistical energy benchmark of the historical healthy period and the area of ​​the spatially continuous region exceeds a preset proportion of the total projected area of ​​the current poultry flock is identified as a static gathering area.

[0115] In the embodiments described in this specification, the system calls a pre-stored historical health period statistical energy benchmark (an energy threshold) and calculates the sum of high conductivity regions in the current image as the total projected area of ​​the current flock. In the energy distribution map obtained in S801, all spatial cells with impedance fluctuation energy lower than the historical benchmark are identified.

[0116] The spatially connected low-energy units identified above are aggregated into continuous regions. The area of ​​each continuous region is calculated. Continuous regions whose area exceeds a preset proportion of the total projected area of ​​the current flock are ultimately identified as static aggregation areas. This region represents the parts of the flock that remain still and huddled together for extended periods.

[0117] S803, the region where the impedance fluctuation energy is higher than the statistical energy benchmark and the overall regional center of gravity movement speed is lower than the preset movement speed benchmark value is identified as a high-activity low-movement zone.

[0118] In this embodiment of the specification, in the energy distribution map obtained in S801, all spatial units with impedance fluctuation energy higher than the historical benchmark are identified and aggregated into candidate regions. For each candidate region, the overall regional center of gravity movement speed is calculated by analyzing the changes in the center of gravity coordinates of that region in the image sequence. Regions with center of gravity movement speeds lower than a preset movement speed benchmark value are ultimately identified as high-activity, low-movement areas. This region represents a part of the flock where there is frequent activity within a small area but a lack of overall directional movement.

[0119] S804, the spatiotemporal distribution characteristics of the static aggregation area and the high-activity low-mobility area are used as impedance spatiotemporal characteristics for judging abnormal group behavior.

[0120] In the embodiments described in this specification, the system summarizes the information of all static clustering areas identified by S802 and all high-activity, low-mobility areas identified by S803.

[0121] The spatiotemporal distribution characteristics of these regions are extracted and formatted, including parameters such as their respective numbers, anomalous spatial locations, areas, energy levels, and movement speeds. These formatted feature sets are then used as impedance spatiotemporal features to determine abnormal group behavior and output to the downstream health status analysis module.

[0122] It should be noted that this method, by introducing time-frequency domain analysis, extracts two core physical quantities from dynamic impedance distribution image sequences: "impedance fluctuation energy" and "regional center of gravity movement velocity." Based on this, it innovatively defines two types of characteristic regions with clear pathological or behavioral orientations: "static aggregation areas" and "high-activity, low-movement areas." This method goes beyond the traditional extraction of simple geometric or statistical features (such as area and center of gravity) from images. Instead, it starts from the essential nature of "energy" and "ordered movement" that inevitably accompany biological activity and can be captured by electrical signals, to deeply characterize the state of poultry flocks. By associating "static aggregation areas" with low fluctuation energy and large spatial proportion, it effectively identifies negative states such as huddling and inactivity caused by severe discomfort or disease in poultry flocks. Conversely, by combining "high-activity, low-movement areas" with high fluctuation energy and low mobility, it can capture the aimless and anxious activities of poultry flocks within a small area caused by restlessness, pain, or other reasons. The extraction of these two types of features enables the system to directly separate and quantify specific behavioral patterns that are highly correlated with health risks from complex impedance distribution dynamics. This provides highly discriminative and interpretable input features for subsequent anomaly judgment, thereby significantly improving the system's detection sensitivity and diagnostic accuracy for early, subtle, and patterned group behavioral anomalies.

[0123] It's important to note that all of the above tasks can be controlled by an intelligent agent, or simply put, a workflow for that agent. This workflow tells the agent when to query the knowledge base, when to perform risk control operations, and so on.

[0124] Figure 2 This is a schematic diagram of the structure of a poultry health status monitoring device provided in one or more embodiments of this specification. The device is executed by a data processing system deployed in a poultry house and includes: At least one processor and bus; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to: Control the electrical impedance imaging sensor array and multimodal acquisition device group deployed in the poultry house to synchronously collect monitoring data, and add timestamps and corresponding spatial location information to the monitoring data to obtain a spatiotemporally aligned raw multimodal data stream. The monitoring data includes at least electrical impedance imaging measurement voltage data, environmental parameter data, audio data and video data. Based on the original multimodal data stream, a multimodal feature vector of poultry is extracted. The multimodal feature vector of poultry includes: extracting spatiotemporal impedance features based on the electrical impedance imaging voltage measurement data, extracting environmental features based on the environmental parameter data, and extracting acoustic features and visual behavioral features based on the audio data and video data. The spatiotemporal impedance features are the impedance distribution image of the poultry house cross section reconstructed based on the electrical impedance imaging voltage measurement data, and the relevant features characterizing the spatial distribution and dynamic activities of the poultry flock are extracted from the sequence impedance distribution image. The poultry multimodal feature vectors are analyzed and compared with a pre-established health status benchmark pattern to determine whether the poultry flock's health status is abnormal. When an anomaly is detected, the monitoring results of poultry health status are generated by combining the correlation analysis results between the impedance spatiotemporal characteristics, the environmental characteristics, the acoustic characteristics and the visual behavioral characteristics.

[0125] This specification provides one or more embodiments of a non-volatile computer storage medium, which is executed by a data processing system deployed in a poultry house. The medium stores computer-executable instructions that, when executed by a computer, can achieve the following: Control the electrical impedance imaging sensor array and multimodal acquisition device group deployed in the poultry house to synchronously collect monitoring data, and add timestamps and corresponding spatial location information to the monitoring data to obtain a spatiotemporally aligned raw multimodal data stream. The monitoring data includes at least electrical impedance imaging measurement voltage data, environmental parameter data, audio data and video data. Based on the original multimodal data stream, a multimodal feature vector of poultry is extracted. The multimodal feature vector of poultry includes: extracting spatiotemporal impedance features based on the electrical impedance imaging voltage measurement data, extracting environmental features based on the environmental parameter data, and extracting acoustic features and visual behavioral features based on the audio data and video data. The spatiotemporal impedance features are the impedance distribution image of the poultry house cross section reconstructed based on the electrical impedance imaging voltage measurement data, and the relevant features characterizing the spatial distribution and dynamic activities of the poultry flock are extracted from the sequence impedance distribution image. The poultry multimodal feature vectors are analyzed and compared with a pre-established health status benchmark pattern to determine whether the poultry flock's health status is abnormal. When an anomaly is detected, the monitoring results of poultry health status are generated by combining the correlation analysis results between the impedance spatiotemporal characteristics, the environmental characteristics, the acoustic characteristics and the visual behavioral characteristics.

[0126] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0127] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the apparatus embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0128] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software 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.

[0129] In the embodiments provided in this application, it should be understood that the disclosed apparatus / network devices and methods can be implemented in other ways. For example, the apparatus / network device embodiments described above are merely illustrative. For instance, the division of modules or units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0130] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0131] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The aforementioned units can be implemented in hardware or software.

[0132] If the integrated module / unit is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, computer-readable media do not include electrical carrier signals and telecommunication signals.

[0133] The above-described embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A method of monitoring the health status of poultry, characterized in that, The method includes: Control the electrical impedance imaging sensor array and multimodal acquisition device group deployed in the poultry house to synchronously collect monitoring data, and add timestamps and corresponding spatial location information to the monitoring data to obtain a spatiotemporally aligned raw multimodal data stream. The monitoring data includes at least electrical impedance imaging measurement voltage data, environmental parameter data, audio data and video data. Based on the original multimodal data stream, a multimodal feature vector of poultry is extracted. The multimodal feature vector of poultry includes: extracting spatiotemporal impedance features based on the electrical impedance imaging voltage measurement data, extracting environmental features based on the environmental parameter data, and extracting acoustic features and visual behavioral features based on the audio data and video data. The spatiotemporal impedance features are the impedance distribution image of the poultry house cross section reconstructed based on the electrical impedance imaging voltage measurement data, and the relevant features characterizing the spatial distribution and dynamic activities of the poultry flock are extracted from the sequence impedance distribution image. The poultry multimodal feature vectors are analyzed and compared with a pre-established health status benchmark pattern to determine whether the poultry flock's health status is abnormal. When an anomaly is detected, the monitoring results of poultry health status are generated by combining the correlation analysis results between the impedance spatiotemporal characteristics, the environmental characteristics, the acoustic characteristics and the visual behavioral characteristics.

2. The method of claim 1, wherein, The electrical impedance imaging sensor array includes multiple electrodes evenly arranged circumferentially along the inner wall of the poultry house; controlling the electrical impedance imaging sensor array deployed within the poultry house to synchronously acquire the electrical impedance imaging measurement voltage data includes: According to the preset excitation mode, safe AC excitation current is injected sequentially into a pair of electrodes on the circumferential direction of the selected poultry house inner wall; The boundary voltages between the remaining multiple electrode pairs in the electrical impedance imaging sensor array are measured simultaneously to obtain a voltage dataset for image reconstruction, thereby completing one full data acquisition cycle.

3. The method of claim 1, wherein, The process of reconstructing the impedance distribution image of the poultry house cross-section based on the voltage data measured by the electrical impedance imaging includes: Establish a finite element calculation model corresponding to the cross-sectional geometry of the poultry house; Calculate the poultry house sensitivity matrix based on the finite element calculation model; The statistical value of the voltage data measured by electrical impedance imaging during historical healthy periods was used as the reference voltage; Based on the current impedance imaging voltage data, the reference voltage, and the poultry house sensitivity matrix, a dynamic image reconstruction algorithm is used to solve for the poultry house conductivity change distribution map relative to the reference state at the current moment, which is used as the impedance distribution image.

4. The method of claim 1, wherein, The determination of whether the health status of the poultry flock is abnormal includes: Calculate the deviation between the poultry multimodal feature vector at the current moment and the health status baseline pattern for the corresponding time period; If the deviation exceeds the dynamic threshold and the duration of the abnormal state exceeds the preset time window, the health status of the poultry flock is determined to be abnormal.

5. The method of claim 1, wherein, The correlation analysis results between the impedance spatiotemporal characteristics, environmental characteristics, acoustic characteristics, and visual behavioral characteristics are combined to generate monitoring results of poultry health status, including: The preset specific pattern of the impedance spatiotemporal characteristics on which the anomaly is determined is used for analysis; Retrieve the environmental features associated with the abnormal spatial region indicated by the specific pattern, and determine whether there are environmental parameters that exceed the preset safety range; If present, a monitoring result indicating that environmental stress is the primary cause of the abnormality will be generated; If not, the acoustic features and visual behavioral features are combined. If the acoustic features or visual behavioral features indicate the presence of abnormal individual behavior, a monitoring result indicating potential disease risk is generated.

6. The method according to claim 5, characterized in that, The combination of the acoustic features and the visual behavioral features includes: When the environmental characteristics do not exceed the safety range, perform spectrum analysis and event detection on the audio data to identify specific abnormal acoustic events and their frequencies; The video data is then subjected to target detection and pose estimation models to identify specific abnormal individual behaviors and their numbers. If the frequency of the abnormal acoustic events exceeds a first frequency threshold, or the number of abnormal individual behaviors exceeds a first quantity threshold, then a monitoring result indicating potential disease risk is generated. If the frequency of the abnormal acoustic events exceeds a second frequency threshold, and the number of abnormal individual behaviors exceeds a second quantity threshold, a monitoring result indicating a high-confidence disease risk is generated, and the warning level is raised.

7. The method according to claim 5, characterized in that, The method further includes: Based on the abnormal type and spatial location of the environmental parameters that exceed the preset safety range, an adjustment command for the environmental control equipment is generated; The environmental control equipment adjustment command is sent to the environmental control system of the poultry house to automatically perform at least one of the following operations: ventilation, cooling, humidification, or light adjustment.

8. The method according to claim 1, characterized in that, The extraction of impedance spatiotemporal features characterizing the spatial distribution and dynamic activities of bird flocks from sequence impedance distribution images includes: The impedance distribution image of the sequence within a preset time window is transformed in the time-frequency domain to calculate the impedance fluctuation energy in the preset frequency band. Areas where the impedance fluctuation energy is lower than the statistical energy benchmark of historical healthy periods and the area of ​​the spatially continuous region exceeds a preset proportion of the total projected area of ​​the current flock are identified as static gathering areas. The region where the impedance fluctuation energy is higher than the statistical energy benchmark and the overall regional center of gravity movement speed is lower than the preset movement speed benchmark value is identified as a high-activity low-movement region. The spatiotemporal distribution characteristics of the static aggregation area and the high-activity, low-mobility area are used as impedance spatiotemporal characteristics for judging abnormal group behavior.

9. A poultry health status monitoring device, characterized in that, include: At least one processor and bus; as well as, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to: Control the electrical impedance imaging sensor array and multimodal acquisition device group deployed in the poultry house to synchronously collect monitoring data, and add timestamps and corresponding spatial location information to the monitoring data to obtain a spatiotemporally aligned raw multimodal data stream. The monitoring data includes at least electrical impedance imaging measurement voltage data, environmental parameter data, audio data and video data. Based on the original multimodal data stream, a multimodal feature vector of poultry is extracted. The multimodal feature vector of poultry includes: extracting spatiotemporal impedance features based on the electrical impedance imaging voltage measurement data, extracting environmental features based on the environmental parameter data, and extracting acoustic features and visual behavioral features based on the audio data and video data. The spatiotemporal impedance features are the impedance distribution image of the poultry house cross section reconstructed based on the electrical impedance imaging voltage measurement data, and the relevant features characterizing the spatial distribution and dynamic activities of the poultry flock are extracted from the sequence impedance distribution image. The poultry multimodal feature vectors are analyzed and compared with a pre-established health status benchmark pattern to determine whether the poultry flock's health status is abnormal. When an anomaly is detected, the monitoring results of poultry health status are generated by combining the correlation analysis results between the impedance spatiotemporal characteristics, the environmental characteristics, the acoustic characteristics and the visual behavioral characteristics.

10. A non-volatile computer storage medium, characterized in that, It stores computer-executable instructions, which, when executed by a computer, can achieve the following: Control the electrical impedance imaging sensor array and multimodal acquisition device group deployed in the poultry house to synchronously collect monitoring data, and add timestamps and corresponding spatial location information to the monitoring data to obtain a spatiotemporally aligned raw multimodal data stream. The monitoring data includes at least electrical impedance imaging measurement voltage data, environmental parameter data, audio data and video data. Based on the original multimodal data stream, a multimodal feature vector of poultry is extracted. The multimodal feature vector of poultry includes: extracting spatiotemporal impedance features based on the electrical impedance imaging voltage measurement data, extracting environmental features based on the environmental parameter data, and extracting acoustic features and visual behavioral features based on the audio data and video data. The spatiotemporal impedance features are the impedance distribution image of the poultry house cross section reconstructed based on the electrical impedance imaging voltage measurement data, and the relevant features characterizing the spatial distribution and dynamic activities of the poultry flock are extracted from the sequence impedance distribution image. The poultry multimodal feature vectors are analyzed and compared with a pre-established health status benchmark pattern to determine whether the poultry flock's health status is abnormal. When an anomaly is detected, the monitoring results of poultry health status are generated by combining the correlation analysis results between the impedance spatiotemporal characteristics, the environmental characteristics, the acoustic characteristics and the visual behavioral characteristics.

Images