A computer vision-based animal behavior micro-dynamics characterization and state analysis method and related device

Abstract

The application discloses a computer vision-based animal behavior microdynamics characterization and state analysis method and a related device, and relates to the technical field of animal behavior, which comprises the following steps: performing visual information extraction on a video image sequence; according to obtained structured visual information, extracting continuous target behavior segments belonging to the same macro behavior situation based on preset context anchoring rules; extracting multi-modal micro fluctuation characteristics of a macro behavior event, and constructing a micro fluctuation characteristic sequence of a unified time scale through multi-modal micro fluctuation characteristic alignment; adopting a mixed weight strategy to perform weighted combination on the micro fluctuation characteristic sequence, and calculating a dynamic instability index of a target animal; based on the index, performing unsupervised time state analysis on the macro behavior event by using a threshold value pre-fitted from a global macro behavior reference data set, and generating a quantitative result based on a microdynamics phase sequence of multiple ordered microdynamics phases and time nodes of the microdynamics phases.

Description

Technical Field

[0001] This application relates to the field of animal behavior technology, and in particular to a method and related apparatus for representing and analyzing the micro-dynamics of animal behavior based on computer vision. Background Technology

[0002] Animal behavior video analysis technology has been widely used to identify animal behaviors such as standing, lying down, feeding, drinking, moving, and social interaction. Existing methods typically rely on object detection, individual tracking, posture estimation, or action classification models to output discrete category labels for animal behaviors and perform basic macro-statistics on their frequency or duration.

[0003] However, this traditional discretization classification paradigm obscures the rich microscopic fluctuation information hidden within the entire behavioral spectrum (encompassing both dynamic and static aspects). In real-world scenarios, whether in static dwell or dynamic motion, the same macroscopic behavior contains varying degrees of subtle kinematic and morphological evolution.

[0004] Because existing systems lack quantitative analysis of the "fine-grained" fluctuations within macroscopic tags, they severely lose the ability to perceive early health warnings, environmental responsiveness, and deep-seated social stress in animals. Therefore, a high-dimensional analysis technology solution for animal behavior videos is urgently needed. Summary of the Invention

[0005] The purpose of this application is to provide a method and related device for representing and analyzing the micro-dynamics of animal behavior based on computer vision, which can solve the problem that existing technologies mainly rely on macro-level behavioral categories and duration statistics, and are difficult to analyze subtle dynamic differences within the same behavioral context.

[0006] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a computer vision-based method for representing and analyzing the micro-dynamics of animal behavior and its state, including: Acquire video image sequences of the target animal; Visual information is extracted from the video image sequence to obtain the structured visual information of the target animal at each sampling time. Based on the structured visual information and a preset context anchoring rule, continuous target behavior segments belonging to the same macro-behavioral context are extracted from the video image sequence to obtain macro-behavioral events. Multimodal micro-fluctuation features of the macro-behavioral events are extracted under a short time window, and a micro-fluctuation feature sequence with a unified time scale is constructed by aligning the multimodal micro-fluctuation features under different time resolutions. A mixed weighting strategy is used to weight and combine the micro-fluctuation feature sequences to calculate the dynamic instability index of the target animal at each sampling time. Based on the dynamic instability index, an unsupervised time state analysis is performed on the macroscopic behavior event using a global threshold prefitted from the global macroscopic behavior reference dataset to obtain a sequence of microscopic dynamic phase states with multiple ordered microscopic dynamic phase states; the global threshold is a common threshold fitted from the global macroscopic behavior reference dataset composed of dynamic instability sequences of multiple context-anchored behavior segments. Based on the microdynamic phase sequence and its time nodes, the microdynamic quantification results of animal behavior are generated; the microdynamic quantification results of animal behavior include the phase time ratio; continuous residence time; phase transition density per unit time; phase transition probability matrix; the ratio of the longest stable phase; and the combined ratio of intermediate unstable phase and highly unstable phase.

[0007] In a second aspect, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the computer vision-based method for representing and analyzing the micro-dynamics of animal behavior as described in any one of the above-mentioned methods.

[0008] Thirdly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of any of the above-described computer vision-based methods for representing and analyzing the micro-dynamics of animal behavior.

[0009] Fourthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of any of the above-described computer vision-based methods for representing and analyzing the micro-dynamics of animal behavior.

[0010] According to the specific embodiments provided in this application, this application has the following technical effects: This application provides a computer vision-based method and apparatus for representing and analyzing the micro-dynamics of animal behavior and its state. By anchoring target behavioral segments within the same macro-behavioral context, it extracts and integrates multimodal micro-fluctuation features, constructs dynamic instability indices, and achieves unsupervised micro-dynamic phase analysis, overcoming the limitation of traditional animal behavior analysis that can only output discrete macro-level category labels. This approach eliminates the need for manual frame-by-frame annotation of micro-states to quantitatively extract deep motion features, reducing the labor costs and subjective biases of micro-behavioral annotation while outputting high-dimensional digital behavioral phenotypes. This provides more refined quantitative support for scenarios such as early warning of animal health and welfare assessment, effectively bridging the analytical gap between macro-behavioral labels and micro-physiological state phenotypes. Compared to traditional methods that rely solely on macro-behavioral duration statistics, this application can capture subtle dynamic differences within the same macro-behavioral context, demonstrating stronger perception capabilities for abnormal states such as early stages of disease and mild environmental stress that have not yet triggered changes in macro-behavioral behavior. This better meets the needs of refined animal behavior analysis in practical applications. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is an application environment diagram of a computer vision-based method for representing and analyzing the micro-dynamics of animal behavior in one embodiment of this application. Figure 2 A flowchart illustrating a computer vision-based method for characterizing and analyzing the micro-dynamics of animal behavior and its state, provided in an embodiment of this application; Figure 3 This is a schematic diagram illustrating animal video acquisition, target parsing, and context-constrained segment extraction according to an embodiment of this application; Figure 4 This is a schematic diagram illustrating the multimodal micro-wave characteristics provided in an embodiment of this application; Figure 5 A flowchart of microscopic dynamic state analysis and state transition analysis provided in an embodiment of this application; Figure 6 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application.

[0013] Figure label: 102 - Terminal, 104 - Server. Detailed Implementation

[0014] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0015] Existing methods typically rely on object detection, individual tracking, posture estimation, or action classification models to output discrete category labels for animal behaviors and perform basic macro-statistics on their frequency or duration.

[0016] However, this traditional discretization paradigm obscures the rich microscopic fluctuations hidden within the entire behavioral spectrum (encompassing both dynamic and static aspects). In real-world scenarios, whether in static dwelling or dynamic movement, subtle kinematic and morphological evolutions exist to varying degrees within the same macroscopic behavior. For example, when group-raised animals (such as weaned piglets) are in a low-displacement static state, such as lying down or standing still, existing visual monitoring systems (such as the conventional YOLO+DeepSORT architecture) can only crudely label them as "resting" or "still," even discarding these low-displacement periods as "no-event" time. However, behavioral studies show that individuals experiencing stress, physical discomfort, or social oppression often exhibit high-frequency localized postural fine-tuning, abnormal deformation, or muscle tremors during their "resting" state. On the other hand, when animals are in dynamic situations such as walking, exploring or fighting, existing systems can only output trajectories or macroscopic action labels, but cannot quantify and capture microscopic feature differences that can directly reflect fatigue, injury risks or the intensity of social confrontation, such as gait symmetry micro-changes, force gradients or high-frequency spatiotemporal deformations.

[0017] Therefore, the purpose of this application is to provide a method and related device for representing and analyzing the micro-dynamics of animal behavior based on computer vision, which can solve the problem that the existing technology mainly relies on macro-level behavioral categories and duration statistics and is difficult to analyze subtle dynamic differences within the same behavioral context.

[0018] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, this application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] The computer vision-based method for representing and analyzing the micro-dynamics of animal behavior provided in this application can be applied to, for example... Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be set up independently, integrated into server 104, or placed in the cloud or on another server.

[0020] The terminal 102 can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, and smart in-vehicle devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted devices. The server 104 can be implemented using a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0021] In one exemplary embodiment, such as Figure 2 As shown, a computer vision-based method for representing the microscopic dynamics of animal behavior and analyzing its state is provided. This method is executed by a computer device, specifically a terminal or server, or both. In this embodiment, the method is applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 207. Wherein: Step 201: Obtain a video image sequence of the target animal; Step 202: Visual information is extracted from the video image sequence to obtain the structured visual information of the target animal at each sampling time. Step 203: Based on the structured visual information and a preset context anchoring rule, extract continuous target behavior segments belonging to the same macro-behavioral context from the video image sequence to obtain macro-behavioral events; Step 204: Extract the multimodal micro-fluctuation features of the macro-behavioral events under a short time window, and construct a micro-fluctuation feature sequence with a unified time scale by aligning the multimodal micro-fluctuation features under different time resolutions. Step 205: The micro-fluctuation feature sequence is weighted and combined using a hybrid weighting strategy to calculate the dynamic instability index of the target animal at each sampling time. Step 206: Based on the dynamic instability index, an unsupervised time state analysis is performed on the macroscopic behavior event using a global threshold pre-fitted from the global macroscopic behavior reference dataset to obtain a sequence of microscopic dynamic phase states of multiple ordered microscopic dynamic phase states; the global threshold is a common threshold fitted from the global macroscopic behavior reference dataset composed of dynamic instability sequences of multiple context-anchored behavior segments. Step 207: Based on the microdynamic phase sequence and its time nodes, generate the microdynamic quantification results of animal behavior; the microdynamic quantification results of animal behavior include the phase time ratio; continuous residence time; phase transition density per unit time; phase transition probability matrix; the ratio of the longest stable phase; and the combined ratio of intermediate unstable phase and highly unstable phase.

[0022] In one exemplary embodiment, when performing steps 201-207, the specific steps may be as follows: like Figure 3 As shown, a video image sequence of the target animal is acquired by an image acquisition device installed within the animal's activity space, and the video image sequence is input into a computer device. The target animal can be a pig, cow, sheep, dog, cat, rat, or other group-raised animals with observable behavioral activities. In this embodiment, the target animal is a weaned piglet.

[0023] The video image sequence is subjected to target detection, target tracking, instance segmentation, and key point pose recognition to obtain structured visual information of the target animal at each sampling time.

[0024] The structured visual information includes at least one or more of the following: the spatial location or centroid location of the target animal; the bounding box information of the target animal; the outline mask information of the target animal; the location information of the head, tail or other key parts of the target animal; and the functional area information of the target animal.

[0025] Based on the structured visual information, the macroscopic behavioral context of the target animal is determined. According to preset context anchoring rules, any continuous target behavioral segment requiring microscopic dynamic analysis is selected from the video image sequence and defined as a macroscopic behavioral event. This macroscopic behavioral event covers the entire behavioral spectrum (including but not limited to dynamic movement segments and static dwelling segments) and is jointly determined by at least two types of constraint information: macroscopic movement state determined based on changes in centroid displacement or movement velocity characteristics (the macroscopic movement state includes low-displacement dwelling states with displacement below a preset threshold, and dynamic movement states with displacement above or equal to a preset threshold); specific behavioral categories determined based on posture or action recognition results (e.g., lying down, standing, walking, running, exploring, or fighting); and regional context determined based on the spatial relationship between the animal's location and preset functional areas (e.g., feed trough area, rest area, activity area, or excretion area). Through this preset context anchoring method, accurate extraction of target behavioral segments under any specific context constraint can be achieved, avoiding disordered mixing of different macroscopic behavioral categories, thus providing a unified and rigorous contextual benchmark for subsequent refined analysis of microscopic features.

[0026] like Figure 4As shown, for the extracted macroscopic behavioral events, multimodal microscopic fluctuation features of the target animal are extracted within a short temporal window. To accurately capture action details in different dimensions, this embodiment employs a multi-temporal resolution extraction strategy and aligns the features acquired at different sampling rates along the temporal axis to construct a unified feature fragment sequence. The multimodal microscopic fluctuation features include at least the following groups: 1) Geometric and kinematic feature set: Based on the basic temporal resolution (e.g., second-level aggregation window), extract velocity, acceleration, rate of change of direction and their temporal statistics generated from key points or target bounding boxes; 2) Morphological deformation feature set: Based on the basic temporal resolution, extract features including contour area, bounding box shape ratio, curvature, contour relative difference degree and its temporal statistics; 3) Deep spatiotemporal semantic feature set: Based on high-frequency temporal resolution (e.g., inputting continuous high frame rate sequences higher than the basic temporal resolution, such as 16 frames / second level), multi-scale spatiotemporal gradient energy and fine-grained temporal variation features are extracted through dual-stream three-dimensional convolutional neural networks (such as deep spatiotemporal architectures like SlowFast); Subsequently, the deep spatiotemporal semantic features extracted based on high-frequency temporal resolution and the geometric and morphological features extracted based on basic temporal resolution are downsampled or aligned and pooled in the temporal dimension to generate a micro-fluctuation feature representation sequence with a unified time scale (such as continuous seconds).

[0027] Within each sampling time window, construct a system containing dynamic instabilities ( ), local boundary strength ( ), degree of sustained fluctuation ( ) and overall activity level ( The local dynamics descriptor vector.

[0028] A hybrid weighting strategy was used to weight and combine the multimodal micro-fluctuation characteristics to calculate the dynamic instability index of the target animal at each sampling time. .

[0029] To overcome the subjectivity of manual weight allocation in existing technologies and the susceptibility of pure mathematical variance to environmental engineering noise (such as tracking jitter and sudden changes in lighting), this embodiment employs a hybrid weighting strategy combining data-driven initialization and expert behavioral regularization to calculate the dynamic instability index. : ; in, Indicates the first One characteristic in time The absolute value of the standardized characteristic fluctuation at that location. For the total number of features, This is the final refined weight. For example... Figure 5 As shown, the refining weight The generation steps are as follows: 1) Data-driven baseline weight initialization: Extract a long-term (e.g., 72-hour) global resident reference dataset, construct a local covariance matrix, and perform principal component analysis (PCA). Extract the variance explained by the first principal component (…). Based on the absolute value of the feature loading along the first principal component, objective baseline weights are proportionally allocated to obtain the baseline weights. .

[0030] 2) Domain-Adaptive Behavioral Calibration: Introducing a domain-adaptive calibration vector corresponding to the feature set. Apply a penalty scaling factor of less than 1 to feature groups that are susceptible to key point occlusion or breathing noise interference (such as morphological deformation group and head-body coordination group); apply an amplification scaling factor of greater than 1 to feature groups that characterize deep behavioral patterns (such as depth spatiotemporal semantic group and shape stability group).

[0031] This hybrid weighting strategy ensures that The calculation maintains extremely high statistical objectivity while possessing extremely strong robustness against visual artifacts.

[0032] After obtaining the dynamic instability index, based on the dynamic instability index Unsupervised temporal state analysis is performed on the macroscopic behavioral events to obtain the microscopic dynamic phase sequence of the target animal. For example... Figure 5 As shown, the unsupervised temporal state parsing includes: Global thresholds are extracted based on a global macroscopic behavior reference dataset. and ; Using the global threshold, the continuous dynamic instability index is discretized into three ordered phases: a stable phase (SP, ...). ), moderately unstable phase (IIP, ) and highly unstable phases (HIP, ); Temporal consistency corrections, including hysteresis effects and minimum duration constraints, are applied to the initial phase sequence to filter out spurious transition points caused by visual transient noise.

[0033] Based on the multi-level microdynamic phase sequence and its time nodes (including but not limited to occurrence, transition, and termination time points), digital behavioral phenotypic quantification results are generated. The quantification results include, but are not limited to: the occurrence frequency, temporal proportion, and temporal evolution structure of each microdynamic phase in the target behavioral segment; the transition probability and density matrix between phases; and the above quantification indicators are output as digital indicators for assessing the latent physical discomfort, environmental responsiveness, social stress level, disease warning status, or overall welfare level of the target animal (covering individual or group-housed conditions).

[0034] Specifically, the transition time of any preset phase transition path is used as the time alignment point (for example, alignment analysis is performed on the SP→IIP and IIP→SP paths). Multimodal micro-fluctuation features within a preset time window before and after this time point are extracted to form a state transition alignment signature matrix.

[0035] By quantitatively profiling the coordinated changes in geometric kinematics, morphological deformation, and depth spatiotemporal features before and after state transitions, along with multidimensional topological shifts, the key contribution feature sequences corresponding to state transitions are output, which are used to quantify the feature changes of unsupervised phase boundaries.

[0036] In one exemplary embodiment, the refining weight In the generation step, the data-driven baseline weight initialization can also use independent component analysis (ICA), factor analysis, or latent representation reconstruction error based on autoencoder to objectively extract the feature covariance structure and initialize the baseline weights of each feature group.

[0037] In one exemplary embodiment, the deep spatiotemporal semantic group can also be based on dynamic skeleton topological features extracted by a spatiotemporal graph convolutional network (ST-GCN) or spatiotemporal attention flow features extracted by a video vision transformer (ViT).

[0038] In one exemplary embodiment, during phase decoding and boundary partitioning, the local dynamic instability index sequence can also be input into algorithms such as Hidden Markov Model (HMM), Gaussian Mixture Model (GMM), or density-based spatial clustering (DBSCAN) to dynamically generate the hidden state boundary of the microscopic dynamic phase through unsupervised clustering.

[0039] This application also provides a specific embodiment, using group-housed weaned piglets in a commercial-scale pig farm as the monitoring object, and based on a computer vision-based method for representing and analyzing the micro-dynamics of animal behavior, to provide a detailed description of the specific implementation process and effects: Hardware deployment: Deploy image acquisition devices (such as RGB cameras with a resolution of 1080P and a frame rate of 30fps) on the top or side of the pigsty. Video data is transmitted in real time or offline to computing nodes equipped with graphics processing units (GPUs).

[0040] Structured information extraction: A computer vision target detection and pose estimation network is used to extract the two-dimensional centroid coordinates, bounding box, contour mask, head / tail key points, and pose information of piglets frame by frame.

[0041] Contextual stationary event anchoring: Set a macroscopic static determination threshold. For example, when the target piglet's center of mass does not continuously shift and the posture recognition determines that it is in a "lying down" or "standing" state for more than a preset time (such as 1 minute), this time period is extracted as a stationary bout.

[0042] Multimodal feature construction: For this dwell event, micro-fluctuation features at the second level are extracted with a time window of 1 second. In this embodiment, the feature matrix includes shallow geometric kinematic features (such as keypoint displacement velocity), morphological deformation features (such as the curvature and area change rate of the contour mask), and multi-scale deep spatiotemporal gradient features extracted using the SlowFast two-stream network.

[0043] After extracting the multimodal feature sequences, a hybrid weighting strategy combining data-driven approaches and behavioral expert regularization is used to calculate the local dynamic instability index for each second-level window. This is to shield against sudden changes in lighting and tracking jitter noise. The specific implementation is as follows: Baseline weight initialization: Extract a long-term (72-hour continuous) global residency reference dataset of piglets. For a specific feature functional domain, construct a local covariance matrix and perform principal component analysis (PCA). Extract the variance explained by the first principal component (…). ), and obtain objective baseline weights based on characteristic loads. .

[0044] Expert regularization calibration: Pre-set a 10-dimensional feature neighborhood adaptive calibration vector In a preferred embodiment for weaned piglets, the following is set: Among them, the weights of feature groups that are susceptible to cluster occlusion and breathing noise interference (such as the head-body coordination feature group 6, with a multiplier of 0.8) are attenuated; and the weights of the deep spatiotemporal feature groups with strong noise-resistant structural regularity (such as the group 10, with a multiplier of 1.3) are amplified and compensated.

[0045] Indicator Calculation: The first Refined weights of each feature The calculation formula is: in This corresponds to the relevant feature function group. Finally, the dynamic instability index at that moment is output: in, This represents the absolute value of the standardized characteristic fluctuation.

[0046] Based on the distribution of the aforementioned 72-hour global reference dataset, a first threshold is pre-fitted and extracted. Second threshold For sequences calculated in real time Apply dual-threshold discretization: when When the phase is stable (SP), it is determined to be a stable phase. when At that time, it was determined to be a moderately unstable phase (IIP); when When the phase is in this state, it is determined to be a highly unstable phase (HIP).

[0047] Subsequently, a hysteresis effect and a minimum duration constraint (such as filtering out phases that do not last for 2 seconds) are introduced to complete the time consistency correction.

[0048] The phase transition density from SP to HIP and the proportion of SP residence time during this residence event were statistically analyzed. This spatiotemporal organizational structure feature was used as a digital reading. When an abnormally high phase transition density was detected in a piglet while it was "macroscopically lying down," an early health and welfare warning was output to the terminal indicating potential physical discomfort or exposure to high environmental and social stress.

[0049] In one exemplary embodiment, a computer device is provided, which may be a server or a terminal, and its internal structure diagram may be as follows. Figure 6As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The database stores the quantification results of animal behavioral microdynamics. The I / O interfaces are used for information exchange between the processor and external devices. The communication interface is used for communication with external terminals via a network connection. When executed by the processor, the computer program implements a computer vision-based method for representing and analyzing the microdynamics of animal behavior.

[0050] Figure 6 The structures shown are merely block diagrams of some structures related to the present application and do not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than shown in the figures, or combine certain components, or have different component arrangements. In an exemplary embodiment, a computer device is provided, including a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0051] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0052] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0053] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of related data must comply with relevant regulations and be authorized by the owner of the corresponding device.

[0054] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0055] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0056] In summary, this application has the following technical effects: 1) This application defines the target behavior segment by pre-defined rules of macroscopic motion state, posture category and functional area information, so that microscopic dynamic analysis can be carried out in any specified macroscopic behavior context that simultaneously covers dynamic motion and static dwell. This effectively eliminates the analysis bias caused by disordered mixing of different macroscopic behavior contexts and can provide a rigorous benchmark for microscopic quantitative analysis.

[0057] 2) This application significantly improves the ability to accurately express the subtle movements and posture fluctuations of animals in complex physical environments by fusing multimodal micro-features such as geometric kinematics, morphological deformation and deep spatiotemporal gradient of the target animal under a short time window, combined with a hybrid weighted anti-noise calibration strategy based on data-driven and domain expert priors.

[0058] 3) This application uses an unsupervised temporal state parsing and decoding mechanism to stably extract the multi-level micro-dynamic phase sequence hidden inside the macro-behavioral fragment without the need for costly and subjectively biased manual frame-by-frame micro-annotation.

[0059] 4) By outputting the time proportion, occurrence frequency, temporal evolution structure, and phase transition probability and density matrix of each micro-dynamic phase, this application successfully extends animal behavior analysis from traditional discrete macro-categorical statistics to continuous micro-dynamic state analysis, bridging the quantitative gap between macro-labels and micro-phenotypes in the field of behavior.

[0060] 5) This application can directly output high-dimensional digital behavioral phenotypic results, which can be used to comprehensively and non-invasively assess the potential physical discomfort, environmental responsiveness, social stress level, early disease warning status and overall welfare level of group-housed or individual target animals. It has extremely high application value in the fields of refined animal husbandry, pet health monitoring and experimental animal phenotypic analysis.

[0061] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0062] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for representing and analyzing the microscopic dynamics of animal behavior based on computer vision, characterized in that, include: Acquire video image sequences of the target animal; Visual information is extracted from the video image sequence to obtain the structured visual information of the target animal at each sampling time. Based on the structured visual information and a preset context anchoring rule, continuous target behavior segments belonging to the same macro-behavioral context are extracted from the video image sequence to obtain macro-behavioral events. Multimodal micro-fluctuation features of the macro-behavioral events are extracted under a short time window, and a micro-fluctuation feature sequence with a unified time scale is constructed by aligning the multimodal micro-fluctuation features under different time resolutions. A mixed weighting strategy is used to weight and combine the micro-fluctuation feature sequences to calculate the dynamic instability index of the target animal at each sampling time. Based on the dynamic instability index, an unsupervised time state analysis is performed on the macroscopic behavior event using a global threshold prefitted from the global macroscopic behavior reference dataset, resulting in a microscopic dynamic phase sequence of multiple ordered microscopic dynamic phases. The global threshold is a common threshold fitted from a global macroscopic behavior reference dataset consisting of dynamic instability sequences of multiple context-anchored behavior segments. Based on the microdynamic phase sequence and its time nodes, a quantitative result of animal behavior microdynamics is generated; the quantitative result of animal behavior microdynamics includes the phase time ratio and continuous residence time; Phase transition density per unit time; Phase transition probability matrix; percentage of longest stable phase; combined percentage of intermediate unstable phase and highly unstable phase.

2. The method for representing and analyzing the micro-dynamics of animal behavior based on computer vision according to claim 1, characterized in that, The structured visual information includes the spatial location or centroid location of the target animal, bounding box information, contour mask information, key part location information, and the functional area information in which it is located.

3. The method for representing and analyzing the micro-dynamics of animal behavior based on computer vision according to claim 2, characterized in that, The preset context anchoring rules include: macroscopic motion state determined based on changes in centroid displacement or motion velocity characteristics, specific behavior category determined based on posture or action recognition results, and regional context determined based on the spatial relationship between the animal's location and preset functional areas; the macroscopic motion state includes low displacement dwell state with displacement below a preset threshold and dynamic motion state with displacement above or equal to a preset threshold.

4. The method for representing and analyzing the microscopic dynamics of animal behavior based on computer vision according to claim 3, characterized in that, The multimodal micro-fluctuation features include geometric kinematic features, morphological deformation features, and deep spatiotemporal semantic features; the geometric kinematic features are extracted based on the basic temporal resolution and include velocity, acceleration, rate of change of direction, and their temporal statistics generated from key points or target bounding boxes. The morphological deformation features are extracted based on the basic temporal resolution, including contour area, bounding box shape ratio, curvature, contour relative difference degree and its temporal statistics; the depth spatiotemporal semantic features are obtained by a video spatiotemporal feature extraction network; the video spatiotemporal feature extraction network is a dual-path three-dimensional convolutional network; the depth spatiotemporal semantic features are aligned with the geometric kinematic features and morphological deformation features in the temporal dimension by downsampling or aligned pooling, generating a micro-fluctuation feature representation sequence with a unified time scale.

5. The method for representing and analyzing the micro-dynamics of animal behavior based on computer vision according to claim 4, characterized in that, A weighted combination strategy is used to weight and combine the micro-fluctuation feature sequences to calculate the dynamic instability index of the target animal at each sampling time, specifically including: According to the formula Calculate the dynamic instability index of the target animal at each sampling time; In the formula, Indicates the first One characteristic in time The absolute value of the standardized characteristic fluctuation at that location. For the total number of features, This is the final refined weight.

6. The method for representing and analyzing the micro-dynamics of animal behavior based on computer vision according to claim 5, characterized in that, Based on the aforementioned dynamic instability index, and using a pre-fitted threshold from a global macroscopic behavior reference dataset, unsupervised temporal state analysis is performed on the macroscopic behavior events to obtain a multi-level discrete microscopic dynamic phase sequence, specifically including: Global thresholds are extracted based on a global macroscopic behavior reference dataset through prefitting. The continuous dynamic instability index is discretized into three ordered phases using the global threshold: a stable phase, a moderately unstable phase, and a highly unstable phase. Based on at least one of hysteresis threshold, minimum dwell time, boundary strength, fluctuation degree and overall activity, time consistency correction is performed on the initial phase sequence to filter out false transition points caused by visual transient noise, thereby obtaining a multi-level discrete microdynamic phase sequence; the initial phase sequence is a phase sequence initially divided based on dynamic instability index.

7. The method for representing and analyzing the micro-dynamics of animal behavior based on computer vision according to claim 6, characterized in that, The quantification results of the animal behavior microdynamics are output as behavioral abnormality degree, phase organization characteristics or candidate numerical behavioral phenotypes for animal status assessment.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and capable of running on the processor, characterized in that the processor executes the computer program to implement a computer vision-based method for representing and analyzing the micro-dynamics of animal behavior according to any one of claims 1-7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements a computer vision-based method for representing and analyzing the micro-dynamics of animal behavior according to any one of claims 1-7.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements a computer vision-based method for representing and analyzing the micro-dynamics of animal behavior according to any one of claims 1-7.

Images