Livestock and poultry behavior pattern abnormity identification method based on distribution estimation algorithm and residual network

Abstract

The invention discloses a livestock and poultry behavior pattern abnormity identification method based on a distribution estimation algorithm and a residual network. The method comprises the following steps: S1, obtaining a livestock and poultry behavior uniform structure data set with a consistent structure; s2, constructing a livestock and poultry behavior probability distribution model and storing model parameters; s3, generating a behavior probability difference scoring matrix; s4, constructing a residual neural network model, completing model parameter initialization, and executing multi-layer residual operation to obtain a depth anomaly feature vector; and S5, marking a normal behavior mark or an abnormal behavior mark for each piece of livestock and poultry behavior data, generating a behavior track aggregation result in combination with the behavior abnormality judgment result and the corresponding timestamp information, and completing abnormal grade division on the behavior abnormality judgment result and the behavior track aggregation result according to a preset rule. According to the method, the collaborative change rule among the multi-modal data can be captured, so that the model can detect the significant probability deviation at the initial stage of stress or the early stage of sudden abnormality, and the method has higher prospective abnormality early warning capability.

Description

Technical Field

[0001] The present invention relates to the field of livestock and poultry technology, and in particular to a method for identifying abnormal livestock and poultry behavior patterns based on a distribution estimation algorithm and a residual network. Background Art

[0002] With the widespread application of Internet of Things and artificial intelligence technologies in the agricultural field, livestock and poultry behavior recognition, as an important part of intelligent farming, has become an important means to improve farming efficiency, warn of disease spread and ensure animal welfare. At present, mainstream livestock and poultry behavior recognition technologies are mostly based on video image analysis, motion sensor data acquisition or audio signal processing single modality to perform behavior detection and analysis. Although the recognition of some key behaviors has been initially achieved, there are still significant technical bottlenecks when facing the complexity of behavioral patterns, environmental noise interference, multimodal data fusion and fine discrimination of behavioral anomalies.

[0003] The existing technologies have the following main defects: First, most behavior recognition methods rely on single-source data such as images or acceleration, and lack the ability to jointly model auxiliary information such as infrared thermal imaging or audio data, resulting in recognition results that are highly sensitive to the quality of individual modal data and insufficient robustness. Second, in terms of abnormal behavior detection, a large number of studies are still stuck in static threshold judgment or simple classification based on fixed feature sets, and are unable to dynamically adapt to statistical deviations in the process of behavioral evolution, especially in the early identification of low-frequency mutational abnormal behaviors. In addition, the time synchronization, scale differences and information redundancy problems of multimodal data have limited processing methods in existing systems, resulting in behavior modeling often failing to truly reflect the relationship between temporal and spatial changes, and the behavior judgment results lack timeliness and continuity.

[0004] In terms of processing recognition results, existing technologies are mainly based on binary label output, lacking a hierarchical classification mechanism for the degree of abnormal behavior, making it difficult to support subsequent graded responses and intelligent intervention decisions. In terms of spatial information application, most current studies only use location labels for simple behavior clustering or heat map display, and have not yet formed trajectory reasoning capabilities linked to time and behavior type, which limits the in-depth exploration of the mechanism of abnormal behavior.

[0005] In summary, the current technologies for identifying abnormal behavior patterns of livestock and poultry have obvious deficiencies in modal fusion capabilities, abnormal pattern characterization accuracy, dynamic discrimination mechanism, and trajectory analysis capabilities. There is an urgent need for a new recognition method that can integrate multi-source data, model behavior evolution probability, and have dynamic discrimination and trajectory backtracking capabilities to better adapt to the needs of refined monitoring of livestock and poultry health status and behavior patterns in modern intelligent breeding environments. Summary of the Invention

[0006] One purpose of the present invention is to propose a method for identifying abnormalities in livestock and poultry behavior patterns based on a distribution estimation algorithm and a residual network. The present invention can capture the coordinated change patterns between multimodal data, so that the model can detect significant probability deviations in the early stages of stress or sudden abnormalities, thereby having a stronger forward-looking abnormality warning capability.

[0007] According to an embodiment of the present invention, a method for identifying abnormal livestock and poultry behavior patterns based on a distribution estimation algorithm and a residual network includes the following steps:

[0008] S1. Acquire a multi-source raw dataset of livestock and poultry behaviors covering individual and group behaviors in a livestock and poultry breeding environment in real time, and perform preprocessing to obtain a unified structured dataset of livestock and poultry behaviors with consistent structure;

[0009] S2. Extracting a set of statistical feature vectors based on a unified structured dataset of livestock and poultry behaviors, training the set of statistical feature vectors using a distribution estimation algorithm, constructing a livestock and poultry behavior probability distribution model, and saving the model parameters;

[0010] S3. During the target monitoring period, input the real-time updated data of the unified structured livestock behavior dataset into the livestock behavior probability distribution model, calculate the probability difference score of each data item relative to the model, and generate a behavior probability difference score matrix;

[0011] S4. Build a residual neural network model and initialize the model parameters. Use the behavior probability difference score matrix as input data for the residual neural network model and perform multi-layer residual operations to obtain the deep anomaly feature vector.

[0012] S5. Perform behavior classification based on deep anomaly feature vectors, output behavior anomaly determination results, and label each piece of livestock and poultry behavior data with a normal behavior tag or an abnormal behavior tag. Combine the behavior anomaly determination results with the corresponding timestamp information, perform time series backtracking analysis and spatial position mapping, generate behavior trajectory aggregation results, and complete the anomaly level classification based on preset rules by combining the behavior anomaly determination results and the behavior trajectory aggregation results.

[0013] Optionally, the S1 includes the following steps:

[0014] S11. In the livestock and poultry breeding environment, obtain multi-source original data of livestock and poultry behavior and construct a multi-source original data set of livestock and poultry behavior. Each data item in the multi-source original data set of livestock and poultry behavior consists of five parts: acquisition timestamp t i , image frame data I i , infrared thermal imaging data M i , motion acceleration data A i and audio feature vector V i , image frame data I iIt is a three-dimensional tensor structure, which is represented by an image composed of height, width and channel dimensions. The infrared thermal imaging data M i Record temperature distribution information; motion acceleration data A i Represents the acceleration of the object in each axis, the audio feature vector V i It is an audio signal feature expression composed of frequency and amplitude. The livestock and poultry behavior multi-source original dataset contains N livestock and poultry behavior data items;

[0015] S12. Eliminate data defects from the multi-source original dataset of livestock and poultry behavior, filter out data items containing missing frames, time breakpoints, missing values, or outliers, and retain only valid data items that meet integrity and legality constraints. Unify the size of all image frame data to the specified standard size H0×W0, and unify the dimension of audio feature vectors to a fixed length L. v And the motion acceleration data is sampled at a uniform rate f A Resample and establish a reference timeline T based on the acquisition timestamp, aligning all modal data to a unified time scale. The reference timeline contains K time steps, each of which corresponds to a timestamp t k , at each time step, data from different modalities are synchronously aggregated to generate a set of time-aligned multimodal data sequences;

[0016] S13. Perform noise filtering on the time-aligned multimodal data sequence set, use edge enhancement algorithm to improve image clarity on image frame data, apply sliding window filter to audio feature vector to eliminate high-frequency disturbance, and perform median filtering on motion acceleration data to smooth mutation signals, and obtain noise-suppressed image frame data I′ respectively. k , infrared thermal imaging data M′ k , motion acceleration data A′ k and the audio feature vector V′ k , and obtain a unified structured dataset D of livestock and poultry behaviors with consistent structure final .

[0017] Optionally, S2 includes the following steps:

[0018] S21. Based on the unified structured dataset of livestock and poultry behavior D final , from the image frame data I′ k , infrared thermal imaging data M′ k , motion acceleration data A′ k and the audio feature vector V′ k Extract the modal perception statistical features and construct the modal perception statistical feature vector f k ;

[0019] S22. Set the modal perception statistical feature vector Input the improved conditional distribution estimation algorithm module, introduce the temporal context dependency and modal mutual information coupling relationship in the modeling process of the improved conditional distribution estimation algorithm module, and construct the conditional probability density model P(f k ∣f k-1 ):

[0020]

[0021] Among them, f k-1 is the modal perception statistical eigenvector of the continuous time step k-1, I(f k ) represents the information redundancy measure between different modalities, which is calculated based on the mutual information criterion and reflects the degree of coordinated changes between image, temperature, action and audio. α and β are control parameters, which measure the importance of behavioral continuity and modal coupling respectively. Z is the normalization factor.

[0022] S23. By maximizing the training samples in the conditional probability density model P(f k ∣f k-1 ) to obtain the optimal model parameter set Θ cde .

[0023] Optionally, the modal perception statistical feature vector f k Contains the behavior description information and change trend measurement of each modality at the current time step, including the image frame brightness mean μ I,k , image frame brightness standard deviation σ I,k , brightness change rate between image frames ΔI k ;

[0024] Brightness change rate ΔI k Indicates the degree of change in the frame difference between the current frame and the previous frame, and is used to depict the intensity of the behavior at the image level.

[0025] The features extracted from infrared thermal imaging data include the mean temperature μ M,k , temperature standard deviation σ M,k and the amplitude of body temperature mutation ΔM k , where the body temperature mutation amplitude ΔM k It represents the absolute difference between the mean body temperature of the current time step and the mean body temperature of the previous time step, and is used to reflect behavioral abnormalities such as stress, disease, or local overheating.

[0026] The features extracted from motion acceleration data include the mean acceleration μ A,k , acceleration standard deviation σ A,k and the motion variation ΔA k , where the action variation ΔA kIt represents the Euclidean distance between the acceleration vector of the current time step and the acceleration vector of the previous time step, reflecting the degree of mutation of livestock and poultry movement and the instability of gait;

[0027] The features extracted from the audio feature vector include the audio frequency mean μ V,k , audio energy standard deviation σ V,k And the sound change amplitude ΔV k , where the sound variation amplitude ΔV k Indicates the absolute difference between the current audio frequency mean and the previous time step audio frequency mean, used to identify abnormal audio behaviors such as sudden changes in calls, calls for help, and screams.

[0028] Optionally, S3 includes the following steps:

[0029] S31. During the target monitoring period, collect the current time step t k The unified structure data item d of livestock and poultry behavior k , generate modal perception statistical feature vector f k , the modal perception statistical feature vector f k and the modal perception statistical feature vector f of the previous time step k-1 Input conditional probability density model P(f k ∣f k-1 ), calculate the log-likelihood score of the current behavior:

[0030]

[0031] in, represents the behavior log-likelihood score, α, β, and Z are the behavior continuity control parameter, modal coupling control parameter, and normalization factor determined by model training, respectively;

[0032] S32. Using the average of the behavioral log-likelihood scores obtained during the training phase and standard deviation As a benchmark, calculate the current time step t k Behavioral probability difference score S k , behavioral probability difference score S k Indicates the degree of deviation of the current behavior in the distribution model, the behavior probability difference score S k It is obtained by dividing the difference between the current behavior log likelihood score and the baseline mean by the score standard deviation, which is used to quantify the degree to which the current behavior deviates from the normal pattern;

[0033] S33. In the sliding window W containing the current time step k ={S k-w+1 ,...,S k}, calculate the local average of the window score sequence Standard deviation σW , skewness γ W and kurtosis κ W , define the dynamic behavior anomaly threshold, dynamic behavior anomaly threshold τ k It is a linear combination of the local mean score, standard deviation, skewness and kurtosis;

[0034] S34. Set the current time step t k Behavioral probability difference score S k and dynamic behavior abnormality threshold τ k For comparison, if the behavior probability difference score S k Exceeding the threshold τ k , then the current time step t k The behavior is marked as abnormal behavior, and the set of all time steps marked as abnormal behavior constitutes the preliminary abnormal behavior mark set A init ;

[0035] S35. Preliminary abnormal behavior tag set A init The corresponding modal perception statistical feature vectors are clustered for density analysis, and the abnormal behaviors are divided into multiple abnormal behavior clusters according to the similarity of their modal perception features to form a cluster set C. Each cluster c in the cluster set j Characterize a class of abnormalities with similar patterns of behavioral changes;

[0036] S36. If any abnormal behavior cluster c j The number of abnormal behaviors contained in exceeds the preset cluster number threshold θ group , then the global scoring threshold reduction mechanism is triggered, and the global scoring threshold reduction mechanism reduces the current dynamic behavior abnormality threshold τ k Compare with the weighted combination of the local average score value and the skewness to update a new global score threshold lower limit

[0037] S37. Output the behavior probability difference score matrix S containing the following three pieces of information final : Behavior probability difference score S at each time step k , global scoring threshold adjusted by clustering feedback And the cluster identifier ClusterID to which this time step belongs in the abnormal behavior cluster analysis k .

[0038] Optionally, the S4 further includes the following enhancement steps:

[0039] S41. Embed a multi-scale temporal perception module in the residual neural network model to evaluate the probability difference of the input behavior matrix S final Perform multi-scale convolution processing and construct a scale set Each scale R in the scale set (s) Represents the residual representation extracted under the condition that the convolution kernel receptive field is s;

[0040] S42. Represent the residuals at each scale as R (s) Input the behavior attention guidance module to generate the corresponding behavior attention weight α (s) :

[0041]

[0042] in, is the residual representation of time step k under scale s, C k Score features for raw behavior, W r 、W c , w, b are the training parameters in the attention mechanism, is the behavioral attention weight of the s-th scale at the k-th time step;

[0043] S43. Perform weighted fusion on all scale residual representations combined with behavioral attention weights to generate cross-scale residual fusion feature representation Used to characterize the dynamic characteristics of abnormal behaviors of livestock and poultry at different granularities;

[0044] S44. Introduce the abnormal sparse perception module and introduce the abnormal reconstruction loss function in the training stage Optimize the residual neural network model's ability to discriminate rare anomalies:

[0045]

[0046] Among them, z k is the depth anomaly feature vector at the k-th time step, To reconstruct the target vector, is the flag function for whether the kth time step is abnormal, ‖z k ‖1 is the regularization term for the abnormal feature sparsity, λ s is the sparsity adjustment coefficient;

[0047] S45. The residual network output of the behavioral attention guidance module, cross-scale residual fusion features and abnormal sparse perception module is used as the final abnormality discrimination input to form an enhanced deep abnormality feature sequence Z enhanced .

[0048] Optionally, the S5 includes the following steps:

[0049] S51. Enhanced deep anomaly feature sequence Input behavior classification and discrimination module, using soft classification method to calculate each time step t k Corresponding abnormal behavior probability

[0050]

[0051] in, is the deep abnormal feature vector extracted by the residual neural network model at the kth time step, which contains the multimodal behavior fusion representation of livestock and poultry at the current moment, W o is the classification weight vector, which is used to transform the deep anomaly feature vector Mapped to scalar space, b o is the bias term, exp(·) is the natural exponential function, and together with 1+exp(·) in the denominator, it forms the logistic function;

[0052] S52. Set the classification threshold θ based on the probability of abnormal behavior p , when the probability of abnormal behavior When , the time walk behavior is marked as abnormal behavior, and the abnormal behavior mark set A is generated. flag , all unmarked time steps correspond to behaviors that default to normal behaviors;

[0053] S53. Combine each abnormal time step t k ∈A flag The collection timestamp and the spatial positioning information of the original livestock and poultry individuals or groups are used to perform time series retrospective analysis, construct the behavioral evolution trajectory of livestock and poultry individuals or groups within a certain time window before and after the abnormality occurs, and form a set of abnormal behavior trajectories;

[0054] S54. Based on the abnormal behavior trajectory set and the spatial position data L corresponding to each abnormal behavior time step k ,construct a set of spatial position mappings and perform statistical analysis on the spatial density of abnormal points to reveal the spatiotemporal aggregation characteristics of abnormal behaviors;

[0055] S55. The abnormal behavior determination result is integrated with the abnormal behavior trajectory set, and the abnormal behavior of each time step is graded according to the preset abnormal level classification rules.

[0056] Optionally, the abnormality level classification rules refer to the following standards:

[0057] Level 1 anomaly: probability of abnormal behavior Continuous anomalies occur ≥3 times within the corresponding trajectory range, and the anomalies form a high-density cluster in space, and the number of anomalies in the local area within 5 minutes is ≥5;

[0058] Secondary anomaly: probability of abnormal behavior There are 2 to 3 abnormal points within the trajectory range, and the spatial distribution is locally concentrated or periodic;

[0059] Level 3 abnormality: probability of abnormal behavior There are single abnormal points or random distribution in the trajectory, and no clustering characteristics are formed.

[0060] The beneficial effects of the present invention are:

[0061] (1) The present invention introduces an improved conditional distribution estimation algorithm that integrates the temporal context and the modal mutual information coupling mechanism to construct a conditional probability density model for behavioral evolution, which significantly enhances the model's ability to characterize the continuity of behavioral evolution. The introduction of behavioral continuity control parameters and modal coupling control parameters in conditional distribution modeling can capture the coordinated change patterns between multimodal data, so that the model can detect significant probability deviations in the early stages of stress or sudden abnormalities, thereby having a stronger forward-looking abnormality warning capability.

[0062] (2) The present invention designs a multi-scale behavior perception residual neural network structure, embeds a multi-scale temporal perception module and a behavior attention guidance mechanism in the residual network, and can effectively model the performance differences of livestock and poultry behaviors at different time granularities. The residual signals of behaviors at different receptive fields are extracted through multi-scale convolution, and the weights of the residual features of each scale are adaptively adjusted using the attention mechanism, so that the model can maintain high stability and sensitivity when facing behavioral patterns with frequent switching or superimposed interference.

[0063] (3) The present invention further proposes an abnormal sparse perception enhancement mechanism, which improves the model's ability to recognize rare abnormal behaviors of livestock and poultry by introducing sparsity constraints and abnormal reconstruction loss functions during the training process. Existing livestock and poultry behavior recognition models generally tend to learn high-frequency behavior patterns, resulting in low detection accuracy for low-frequency behaviors. Based on the residual feature sequence, a behavior anomaly probability output and behavior trajectory clustering feedback mechanism is constructed. Combined with the reconstruction error and sparsity adjustment of abnormal behavior, fine modeling of weak signals and high-variability behaviors is achieved, effectively alleviating the sample imbalance problem. BRIEF DESCRIPTION OF THE DRAWINGS

[0064] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0065] Figure 1 This is a flow chart of a method for identifying abnormal livestock and poultry behavior patterns based on a distribution estimation algorithm and a residual network proposed by the present invention. DETAILED DESCRIPTION

[0066] The present invention will now be described in further detail with reference to the accompanying drawings, which are simplified schematic diagrams that illustrate the basic structure of the present invention in a schematic manner.

[0067] refer to Figure 1 A method for identifying abnormal behavior patterns of livestock and poultry based on a distribution estimation algorithm and a residual network includes the following steps:

[0068] S1. Acquire a multi-source raw dataset of livestock and poultry behaviors covering individual and group behaviors in a livestock and poultry breeding environment in real time, and perform preprocessing to obtain a unified structured dataset of livestock and poultry behaviors with consistent structure;

[0069] S2. Extracting a set of statistical feature vectors based on a unified structured dataset of livestock and poultry behaviors, training the set of statistical feature vectors using a distribution estimation algorithm, constructing a livestock and poultry behavior probability distribution model, and saving the model parameters;

[0070] S3. During the target monitoring period, input the real-time updated data of the unified structured livestock behavior dataset into the livestock behavior probability distribution model, calculate the probability difference score of each data item relative to the model, and generate a behavior probability difference score matrix;

[0071] S4. Build a residual neural network model and initialize the model parameters. Use the behavior probability difference score matrix as input data for the residual neural network model and perform multi-layer residual operations to obtain the deep anomaly feature vector.

[0072] S5. Perform behavior classification based on deep anomaly feature vectors, output behavior anomaly determination results, and label each piece of livestock and poultry behavior data with a normal behavior tag or an abnormal behavior tag. Combine the behavior anomaly determination results with the corresponding timestamp information, perform time series backtracking analysis and spatial position mapping, generate behavior trajectory aggregation results, and complete the anomaly level classification based on preset rules by combining the behavior anomaly determination results and the behavior trajectory aggregation results.

[0073] In this embodiment, S1 includes the following steps:

[0074] S11. In the livestock and poultry breeding environment, obtain multi-source original data of livestock and poultry behavior and construct a multi-source original data set of livestock and poultry behavior. Each data item in the multi-source original data set of livestock and poultry behavior consists of five parts: acquisition timestamp t i , image frame data I i , infrared thermal imaging data M i , motion acceleration data A i and audio feature vector V i , image frame data I i It is a three-dimensional tensor structure, which is represented by an image composed of height, width and channel dimensions. The infrared thermal imaging data M i Record temperature distribution information; motion acceleration data A i Represents the acceleration of the object in each axis, the audio feature vector V iIt is an audio signal feature expression composed of frequency and amplitude. The livestock and poultry behavior multi-source original dataset contains N livestock and poultry behavior data items;

[0075] S12. Eliminate data defects from the multi-source original dataset of livestock and poultry behavior, filter out data items containing missing frames, time breakpoints, missing values, or outliers, and retain only valid data items that meet integrity and legality constraints. Unify the size of all image frame data to the specified standard size H0×W0, and unify the dimension of audio feature vectors to a fixed length L. v And the motion acceleration data is sampled at a uniform rate f A Resample and establish a reference timeline T based on the acquisition timestamp, aligning all modal data to a unified time scale. The reference timeline contains K time steps, each of which corresponds to a timestamp t k , at each time step, data from different modalities are synchronously aggregated to generate a set of time-aligned multimodal data sequences;

[0076] S13. Perform noise filtering on the time-aligned multimodal data sequence set, use edge enhancement algorithm to improve image clarity on image frame data, apply sliding window filter to audio feature vector to eliminate high-frequency disturbance, and perform median filtering on motion acceleration data to smooth mutation signals, and obtain noise-suppressed image frame data I′ respectively. k , infrared thermal imaging data M′ k , motion acceleration data A′ k and the audio feature vector V′ k , and obtain a unified structured dataset D of livestock and poultry behaviors with consistent structure final .

[0077] In this embodiment, S2 includes the following steps:

[0078] S21. Based on the unified structured dataset of livestock and poultry behavior D final , from the image frame data I′ k , infrared thermal imaging data M′ k , motion acceleration data A′ k and the audio feature vector V′ k Extract the modal perception statistical features and construct the modal perception statistical feature vector f k ;

[0079] S22. Set the modal perception statistical feature vector Input the improved conditional distribution estimation algorithm module, introduce the temporal context dependency and modal mutual information coupling relationship in the modeling process of the improved conditional distribution estimation algorithm module, and construct the conditional probability density model P(f k ∣f k-1 ):

[0080]

[0081] Among them, f k-1 is the modal perception statistical eigenvector of the continuous time step k-1, I(f k ) represents the information redundancy measure between different modalities, which is calculated based on the mutual information criterion and reflects the degree of coordinated changes between image, temperature, action and audio. α and β are control parameters, which measure the importance of behavioral continuity and modal coupling respectively. Z is the normalization factor.

[0082] S23. By maximizing the training samples in the conditional probability density model P(f k ∣f k-1 ) to obtain the optimal model parameter set Θ cde .

[0083] In this embodiment, the modal perception statistical feature vector f k Contains the behavior description information and change trend measurement of each modality at the current time step, including the image frame brightness mean μ I,k , image frame brightness standard deviation σ I,k , brightness change rate between image frames ΔI k ;

[0084] Brightness change rate ΔI k Indicates the degree of change in the frame difference between the current frame and the previous frame, and is used to depict the intensity of the behavior at the image level.

[0085] The features extracted from infrared thermal imaging data include the mean temperature μ M,k , temperature standard deviation σ M,k and the amplitude of body temperature mutation ΔM k , where the body temperature mutation amplitude ΔM k It represents the absolute difference between the mean body temperature of the current time step and the mean body temperature of the previous time step, and is used to reflect behavioral abnormalities such as stress, disease, or local overheating.

[0086] The features extracted from motion acceleration data include the mean acceleration μ A,k , acceleration standard deviation σ A,k and the motion variation ΔA k , where the action variation ΔA k It represents the Euclidean distance between the acceleration vector of the current time step and the acceleration vector of the previous time step, reflecting the degree of mutation of livestock and poultry movement and the instability of gait;

[0087] The features extracted from the audio feature vector include the audio frequency mean μ V,k , audio energy standard deviation σ V,kAnd the sound change amplitude ΔV k , where the sound variation amplitude ΔV k Indicates the absolute difference between the current audio frequency mean and the previous time step audio frequency mean, used to identify abnormal audio behaviors such as sudden changes in calls, calls for help, and screams.

[0088] In this embodiment, S3 includes the following steps:

[0089] S31. During the target monitoring period, collect the current time step t k The unified structure data item d of livestock and poultry behavior k , generate modal perception statistical feature vector f k , the modal perception statistical feature vector f k and the modal perception statistical feature vector f of the previous time step k-1 Input conditional probability density model P(f k ∣f k-1 ), calculate the log-likelihood score of the current behavior:

[0090]

[0091] in, represents the behavior log-likelihood score, α, β, and Z are the behavior continuity control parameter, modal coupling control parameter, and normalization factor determined by model training, respectively;

[0092] S32. Using the average of the behavioral log-likelihood scores obtained during the training phase and standard deviation As a benchmark, calculate the current time step t k Behavioral probability difference score S k , behavioral probability difference score S k Indicates the degree of deviation of the current behavior in the distribution model, the behavior probability difference score S k It is obtained by dividing the difference between the current behavior log likelihood score and the baseline mean by the score standard deviation, which is used to quantify the degree to which the current behavior deviates from the normal pattern;

[0093] S33. In the sliding window W containing the current time step k ={S k-w+1 ,...,S k}, calculate the local average of the window score sequence Standard deviation σ W , skewness γ W and kurtosis κ W , define the dynamic behavior anomaly threshold, dynamic behavior anomaly threshold τ k It is a linear combination of the local mean score, standard deviation, skewness and kurtosis;

[0094] S34. Set the current time step t k Behavioral probability difference score S k and dynamic behavior abnormality threshold τ k For comparison, if the behavior probability difference score S k Exceeding the threshold τ k , then the current time step t k The behavior is marked as abnormal behavior, and the set of all time steps marked as abnormal behavior constitutes the preliminary abnormal behavior mark set A init ;

[0095] S35. Preliminary abnormal behavior tag set A init The corresponding modal perception statistical feature vectors are clustered for density analysis, and the abnormal behaviors are divided into multiple abnormal behavior clusters according to the similarity of their modal perception features to form a cluster set C. Each cluster c in the cluster set j Characterize a class of abnormalities with similar patterns of behavioral changes;

[0096] S36. If any abnormal behavior cluster c j The number of abnormal behaviors contained in exceeds the preset cluster number threshold θ group , then the global scoring threshold reduction mechanism is triggered, and the global scoring threshold reduction mechanism reduces the current dynamic behavior abnormality threshold τ k Compare with the weighted combination of the local average score value and the skewness to update a new global score threshold lower limit

[0097] S37. Output the behavior probability difference score matrix S containing the following three pieces of information final : Behavior probability difference score S at each time step k , global scoring threshold adjusted by clustering feedback And the cluster identifier ClusterID to which this time step belongs in the abnormal behavior cluster analysis k .

[0098] In this embodiment, S4 further includes the following enhancement steps:

[0099] S41. Embed a multi-scale temporal perception module in the residual neural network model to evaluate the probability difference of the input behavior matrix S final Perform multi-scale convolution processing and construct a scale set Each scale R in the scale set (s) Represents the residual representation extracted under the condition that the convolution kernel receptive field is s;

[0100] S42. Represent the residuals at each scale as R (s) Input the behavior attention guidance module to generate the corresponding behavior attention weight α(s) :

[0101]

[0102] in, is the residual representation of time step k under scale s, C k Score features for raw behavior, W r 、W c , w, b are the training parameters in the attention mechanism, is the behavioral attention weight of the s-th scale at the k-th time step;

[0103] S43. Perform weighted fusion on all scale residual representations combined with behavioral attention weights to generate cross-scale residual fusion feature representation Used to characterize the dynamic characteristics of abnormal behaviors of livestock and poultry at different granularities;

[0104] S44. Introduce the abnormal sparse perception module and introduce the abnormal reconstruction loss function in the training stage Optimize the residual neural network model's ability to discriminate rare anomalies:

[0105]

[0106] Among them, z k is the depth anomaly feature vector at the kth time step, To reconstruct the target vector, is the flag function for whether the kth time step is abnormal, ‖z k ‖1 is the regularization term for abnormal feature sparsity, λ s is the sparsity adjustment coefficient;

[0107] S45. The residual network output of the behavioral attention guidance module, cross-scale residual fusion features and abnormal sparse perception module is used as the final abnormality discrimination input to form an enhanced deep abnormality feature sequence Z enhanced .

[0108] In this embodiment, S5 includes the following steps:

[0109] S51. Enhanced deep anomaly feature sequence Input behavior classification and discrimination module, using soft classification method to calculate each time step t k Corresponding abnormal behavior probability

[0110]

[0111] in, is the deep abnormal feature vector extracted by the residual neural network model at the kth time step, which contains the multimodal behavior fusion representation of livestock and poultry at the current moment, W o is the classification weight vector, which is used to transform the deep anomaly feature vector Mapped to scalar space, b o is the bias term, exp(·) is the natural exponential function, and together with 1+exp(·) in the denominator, it forms the logistic function;

[0112] S52. Set the classification threshold θ based on the probability of abnormal behavior p , when the probability of abnormal behavior When , the time walk behavior is marked as abnormal behavior, and the abnormal behavior mark set A is generated. flag , all unmarked time steps correspond to behaviors that default to normal behaviors;

[0113] S53. Combine each abnormal time step t k ∈A flag The collection timestamp and the spatial positioning information of the original livestock and poultry individuals or groups are used to perform time series retrospective analysis, construct the behavioral evolution trajectory of livestock and poultry individuals or groups within a certain time window before and after the abnormality occurs, and form a set of abnormal behavior trajectories;

[0114] S54. Based on the abnormal behavior trajectory set and the spatial position data L corresponding to each abnormal behavior time step k ,construct a set of spatial position mappings and perform statistical analysis on the spatial density of abnormal points to reveal the spatiotemporal aggregation characteristics of abnormal behaviors;

[0115] S55. The abnormal behavior determination result is integrated with the abnormal behavior trajectory set, and the abnormal behavior of each time step is graded according to the preset abnormal level classification rules.

[0116] In this implementation, the abnormality classification rules refer to the following standards:

[0117] Level 1 anomaly: probability of abnormal behavior Continuous anomalies occur ≥3 times within the corresponding trajectory range, and the anomalies form a high-density cluster in space, and the number of anomalies in the local area within 5 minutes is ≥5;

[0118] Secondary anomaly: probability of abnormal behavior There are 2 to 3 abnormal points within the trajectory range, and the spatial distribution is locally concentrated or periodic;

[0119] Level 3 abnormality: probability of abnormal behavior There are single abnormal points or random distribution in the trajectory, and no clustering characteristics are formed.

[0120] Example 1:

[0121] In November 2024, a large broiler chicken breeding base in City A began to deploy an intelligent recognition system based on the present invention during the peak disease incidence period in autumn. The base raises more than 130,000 broilers per day and adopts a three-dimensional cage structure. 24-hour video surveillance and infrared thermal imaging equipment have long been popular. However, in daily operation and maintenance, the following problems are still frequently encountered: delayed response to abnormal behavior, missed low-frequency events, and failure to effectively warn in the early stages of disease transmission. Traditional methods based on image processing and fixed threshold classification have an accuracy and recall rate of less than 65%, and the cost of manual intervention remains high.

[0122] Starting from the day of deployment, the system described in the present invention conducted a 14-day behavioral pattern monitoring experiment in the A3 breeding workshop. A total of 12 multimodal acquisition nodes were deployed in the field. Each node collected image frames, infrared thermal imaging data, motion acceleration and audio features, and collected approximately 430,000 complete data records every day. After system preprocessing, a unified structured data set was constructed and the time axis was synchronized to form a high-quality multimodal input sequence.

[0123] In the pre-experimental stage, the research team introduced historical data from September (a total of about 240 hours of behavioral records) as model training samples, which included 93 categories and about 4,200 confirmed abnormal behavior records of "anxious gathering", "movement shaking", and "continuous low-frequency calls" to jointly train the conditional probability density model and the residual neural network model in the present invention. During the modeling process, the feature vector of each time step data was extracted through the distribution estimation algorithm, and a behavioral evolution model with time context and modal mutual information as the core was constructed to achieve fitting of the normal behavior distribution. In the model evaluation stage, the recognition accuracy of the training set reached 94.8%, and the accuracy of the validation set was 91.3%, which was significantly better than the 87.2% and 85.6% of the benchmark Gaussian mixture model under the same conditions.

[0124] At 3:26 a.m. on the fourth day of official operation, the system automatically identified for the first time that a group of broiler chickens in the northeast area of Building A3 had a series of sudden changes in vocal frequency, local temperature increases, and drastic changes in image frames within 6 minutes. The corresponding behavior score was higher than the dynamic threshold for 5 consecutive time steps. After cluster analysis, it was classified into the "stress and overheating" abnormal cluster. After the system issued a third-level abnormality prompt, the breeder's on-site inspection found that the waterer in the area was faulty, the water pipe was blocked, and the chickens were gathered and anxious. In this incident, the system issued an advance warning for a total of 13 minutes, successfully avoiding large-scale deaths caused by heat stress.

[0125] To fully verify the practical value of the system, the research team selected another workshop, Building B1, on the same farm as a control group using traditional methods. They used a behavior recognition system based on a threshold model and unimodal image processing. Statistics showed that over a period of 10 consecutive days:

[0126] Table 1 Comparative data of the method of the present invention and the traditional method

[0127] index Method of the present invention (Building A3) Traditional method (Building B1) Total number of abnormal identifications 37 times 24 times Number of correct identifications 34 times 17 times Missed detection rate 8.1％ 29.2％ False alarm rate 5.4％ 21.8％ Average response time lead 11 minutes and 38 seconds 3 minutes 05 seconds Accuracy of rare anomaly recognition (such as stepping, shaking) 91.7％ 63.4％

[0128] Furthermore, for the low-frequency abnormal behavior category, the researchers recorded a total of 56 related events in the subsequent test sample set, of which the method of the present invention correctly identified 51 times, while the traditional method only identified 31 times, which clearly verified the superiority of the sparse anomaly modeling mechanism of the present invention.

[0129] Furthermore, on the morning of the tenth day after the system was deployed, monitoring detected a flock of chickens in Building A3 between 9:17 and 9:29, with significant spatial clustering, movement acceleration exceeding 15% below the historical average, and decreased sound energy. The abnormal behavior probability score reached 0.93, marking the system as a Level 1 anomaly. Tracking and cluster heat map analysis revealed the location of a dead chicken that had not been promptly addressed, leading to the flock's panic. Following the incident, the on-site hygiene issues were corrected, effectively preventing the spread of the disease.

[0130] Through system deployment and comparative tests in real breeding environments, the livestock and poultry behavior pattern abnormality identification method proposed in this invention is superior to existing technologies in terms of recognition accuracy, response speed, and rare anomaly detection capabilities. Especially when faced with complex multimodal data and dynamic behavioral changes, it can significantly improve the intelligence level and early warning capabilities of livestock and poultry health management, and provide solid algorithmic support for large-scale smart breeding scenarios.

[0131] The present invention constructs a conditional probability density model for behavioral evolution by introducing an improved conditional distribution estimation algorithm that integrates time context and modal mutual information coupling mechanism, which significantly enhances the model's ability to characterize the continuity of behavioral evolution. The introduction of behavioral continuity control parameters and modal coupling control parameters in conditional distribution modeling can capture the coordinated change patterns between multimodal data, so that the model can detect significant probability deviations in the early stages of stress or sudden anomalies, thereby having stronger forward-looking anomaly warning capabilities.

[0132] The present invention designs a multi-scale behavior-perception residual neural network structure, embeds a multi-scale temporal perception module and a behavior attention guidance mechanism in the residual network, and can effectively model the performance differences of livestock and poultry behaviors at different time granularities. The residual signals of behaviors under different receptive fields are extracted through multi-scale convolution, and the weights of residual features at each scale are adaptively adjusted using the attention mechanism, so that the model maintains high stability and sensitivity when facing behavioral patterns with frequent switching or superimposed interference.

[0133] The present invention further proposes an abnormal sparse perception enhancement mechanism, which improves the model's ability to recognize rare abnormal behaviors by introducing sparsity constraints and abnormal reconstruction loss functions during the training process. Existing livestock and poultry behavior recognition models generally tend to learn high-frequency behavior patterns, resulting in low detection accuracy for low-frequency behaviors. Based on the residual feature sequence, a behavior anomaly probability output and behavior trajectory clustering feedback mechanism is constructed. Combined with the reconstruction error and sparsity adjustment of abnormal behaviors, fine modeling of weak signals and high-variance behaviors is achieved, effectively alleviating the problem of sample imbalance.

[0134] The above description is only a preferred specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any technician familiar with the technical field, within the technical scope disclosed by the present invention, who makes equivalent replacements or changes based on the technical solution and inventive concept of the present invention, should be covered by the scope of protection of the present invention.

Claims

1. A method for identifying abnormal behavior patterns of livestock and poultry based on distribution estimation algorithm and residual network, characterized in that: The steps include: S1. Acquire a multi-source raw dataset of livestock and poultry behaviors covering individual and group behaviors in a livestock and poultry breeding environment in real time, and perform preprocessing to obtain a unified structured dataset of livestock and poultry behaviors with consistent structure; S2. Extracting a set of statistical feature vectors based on a unified structured dataset of livestock and poultry behaviors, training the set of statistical feature vectors using a distribution estimation algorithm, constructing a livestock and poultry behavior probability distribution model, and saving the model parameters; S3. During the target monitoring period, input the real-time updated data of the unified structured livestock behavior dataset into the livestock behavior probability distribution model, calculate the probability difference score of each data item relative to the model, and generate a behavior probability difference score matrix; S4. Build a residual neural network model and initialize the model parameters. Use the behavior probability difference score matrix as input data for the residual neural network model and perform multi-layer residual operations to obtain the deep anomaly feature vector. S5. Perform behavior classification based on deep anomaly feature vectors, output behavior anomaly determination results, and label each piece of livestock and poultry behavior data with a normal behavior tag or an abnormal behavior tag. Combine the behavior anomaly determination results with the corresponding timestamp information, perform time series backtracking analysis and spatial position mapping, generate behavior trajectory aggregation results, and complete the anomaly level classification based on preset rules by combining the behavior anomaly determination results and the behavior trajectory aggregation results.

2. The method for identifying abnormal behavior patterns of livestock and poultry based on a distribution estimation algorithm and a residual network according to claim 1, characterized in that: Said S1 comprises the following steps: S11. In the livestock and poultry breeding environment, obtain multi-source original data of livestock and poultry behavior and construct a multi-source original data set of livestock and poultry behavior. Each data item in the multi-source original data set of livestock and poultry behavior consists of five parts: acquisition timestamp t i , image frame data I i , infrared thermal imaging data M i , motion acceleration data A i and audio feature vector V i , image frame data I i It is a three-dimensional tensor structure, which is represented by an image composed of height, width and channel dimensions. The infrared thermal imaging data M i Record temperature distribution information; motion acceleration data A i Represents the acceleration of the object in each axis, the audio feature vector V i It is an audio signal feature expression composed of frequency and amplitude. The livestock and poultry behavior multi-source original dataset contains N livestock and poultry behavior data items; S12. Eliminate data defects from the multi-source original dataset of livestock and poultry behavior, filter out data items containing missing frames, time breakpoints, missing values, or outliers, and retain only valid data items that meet integrity and legality constraints. Unify the size of all image frame data to a specified standard size and the dimension of the audio feature vector to a fixed length L. v And the motion acceleration data is sampled at a uniform rate f A Resample and establish a reference timeline T based on the acquisition timestamp, aligning all modal data to a unified time scale. The reference timeline contains K time steps, each of which corresponds to a timestamp t k , at each time step, data from different modalities are synchronously aggregated to generate a set of time-aligned multimodal data sequences; S13. Perform noise filtering on the time-aligned multimodal data sequence set, use edge enhancement algorithm to improve image clarity on image frame data, apply sliding window filter to audio feature vector to eliminate high-frequency disturbance, and perform median filtering on motion acceleration data to smooth mutation signals, and obtain noise-suppressed image frame data I′ respectively. k , infrared thermal imaging data M′ k , motion acceleration data A′ k and the audio feature vector V′ k , and obtain a unified structured dataset D of livestock and poultry behaviors with consistent structure final .

3. The method for identifying abnormal behavior patterns of livestock and poultry based on a distribution estimation algorithm and a residual network according to claim 2, characterized in that: The S2 comprises the following steps: S21. Based on the unified structured dataset of livestock and poultry behavior D final , from the image frame data I′ k , infrared thermal imaging data M′ k , motion acceleration data A′ k and the audio feature vector V′ k Extract the modal perception statistical features and construct the modal perception statistical feature vector f k ; S22. Set the modal perception statistical feature vector Input the improved conditional distribution estimation algorithm module, introduce the temporal context dependency and modal mutual information coupling relationship in the modeling process of the improved conditional distribution estimation algorithm module, and construct the conditional probability density model P(f k ∣f k-1 ): Among them, f k-1 is the modal perception statistical eigenvector of the continuous time step k-1, I(f k ) represents the information redundancy measure between different modalities, which is calculated based on the mutual information criterion and reflects the degree of coordinated changes between image, temperature, action and audio. α and β are control parameters, which measure the importance of behavioral continuity and modal coupling respectively. Z is the normalization factor. S23. By maximizing the training samples in the conditional probability density model P(f k ∣f k-1 ) to obtain the optimal model parameter set Θ cde .

4. The method for identifying abnormal behavior patterns of livestock and poultry based on a distribution estimation algorithm and a residual network according to claim 3, characterized in that: The modal perception statistical feature vector f k Contains the behavior description information and change trend measurement of each modality at the current time step, including the image frame brightness mean μ I,k , image frame brightness standard deviation σ I,k , brightness change rate between image frames ΔI k ; Brightness change rate ΔI k Indicates the degree of change in the frame difference between the current frame and the previous frame, and is used to depict the intensity of the behavior at the image level. The features extracted from infrared thermal imaging data include the mean temperature μ M,k , temperature standard deviation σ M,k and the amplitude of body temperature mutation ΔM k , where the body temperature mutation amplitude ΔM k It represents the absolute difference between the mean body temperature of the current time step and the mean body temperature of the previous time step, and is used to reflect behavioral abnormalities such as stress, disease, or local overheating. The features extracted from motion acceleration data include the mean acceleration μ A,k , acceleration standard deviation σ A,k and the motion variation ΔA k , where the action variation ΔA k It represents the Euclidean distance between the acceleration vector of the current time step and the acceleration vector of the previous time step, reflecting the degree of mutation of livestock and poultry movement and the instability of gait; The features extracted from the audio feature vector include the audio frequency mean μ V,k , audio energy standard deviation σ V,k And the sound change amplitude ΔV k , where the sound variation amplitude ΔV k Indicates the absolute difference between the current audio frequency mean and the previous time step audio frequency mean, used to identify abnormal audio behaviors such as sudden changes in calls, calls for help, and screams.

5. The method for identifying abnormal behavior patterns of livestock and poultry based on a distribution estimation algorithm and a residual network according to claim 3, characterized in that: The S3 includes the following steps: S31. During the target monitoring period, collect the current time step t k The unified structure data item d of livestock and poultry behavior k , generate modal perception statistical feature vector f k , the modal perception statistical feature vector f k and the modal perception statistical feature vector f of the previous time step k-1 Input conditional probability density model P(f k ∣f k-1 ), calculate the log-likelihood score of the current behavior: in, represents the behavior log-likelihood score, α, β, and Z are the behavior continuity control parameter, modal coupling control parameter, and normalization factor determined by model training, respectively; S32. Using the average of the behavioral log-likelihood scores obtained during the training phase and standard deviation As a benchmark, calculate the current time step t k Behavioral probability difference score S k , behavioral probability difference score S k Indicates the degree of deviation of the current behavior in the distribution model, the behavior probability difference score S k It is obtained by dividing the difference between the current behavior log likelihood score and the baseline mean by the score standard deviation, which is used to quantify the degree to which the current behavior deviates from the normal pattern; S33. In the sliding window W containing the current time step k ={S k-w+1 ,...,S k }, calculate the local average of the window score sequence Standard deviation σ W , skewness γ W and kurtosis κ W , define the dynamic behavior anomaly threshold, dynamic behavior anomaly threshold τ k It is a linear combination of the local mean score, standard deviation, skewness and kurtosis; S34. Set the current time step t k Behavioral probability difference score S k and dynamic behavior abnormality threshold τ k For comparison, if the behavior probability difference score S k Exceeding the threshold τ k , then the current time step t k The behavior is marked as abnormal behavior, and the set of all time steps marked as abnormal behavior constitutes the preliminary abnormal behavior mark set A init ; S35. Preliminary abnormal behavior tag set A init The corresponding modal perception statistical feature vectors are clustered for density analysis, and the abnormal behaviors are divided into multiple abnormal behavior clusters according to the similarity of their modal perception features to form a cluster set C. Each cluster c in the cluster set j Characterize a class of abnormalities with similar patterns of behavioral changes; S36. If any abnormal behavior cluster c j The number of abnormal behaviors contained in exceeds the preset cluster number threshold θ group , then the global scoring threshold reduction mechanism is triggered, and the global scoring threshold reduction mechanism reduces the current dynamic behavior abnormality threshold τ k Compare with the weighted combination of the local average score value and the skewness to update a new global score threshold lower limit S37. Output the behavior probability difference score matrix S containing the following three pieces of information final : Behavior probability difference score S at each time step k , global scoring threshold adjusted by clustering feedback And the cluster identifier ClusterID to which this time step belongs in the abnormal behavior cluster analysis k .

6. The method for identifying abnormal behavior patterns of livestock and poultry based on a distribution estimation algorithm and a residual network according to claim 5, characterized in that: The S4 further includes the following enhancement steps: S41. Embed a multi-scale temporal perception module in the residual neural network model to evaluate the probability difference of the input behavior matrix S final Perform multi-scale convolution processing and construct a scale set Each scale R in the scale set (s) Represents the residual representation extracted under the condition that the convolution kernel receptive field is s; S42. Represent the residuals at each scale as R (s) Input the behavior attention guidance module to generate the corresponding behavior attention weight α (s) : in, is the residual representation of time step k under scale s, C k Score features for raw behavior, W r 、W c , w, b are the training parameters in the attention mechanism, is the behavioral attention weight of the s-th scale at the k-th time step, tanh is the hyperbolic tangent function, which means that in the multi-scale temporal perception module, Under the condition of scale s at time step k ′ The residual feature representation, C k′ is the time step k ′ The following behavioral scoring characteristics; S43. Perform weighted fusion on all scale residual representations combined with behavioral attention weights to generate cross-scale residual fusion feature representation Used to characterize the dynamic characteristics of abnormal behaviors of livestock and poultry at different granularities; S44. Introduce the abnormal sparse perception module and introduce the abnormal reconstruction loss function in the training stage Optimize the residual neural network model's ability to discriminate rare anomalies: Among them, z k is the depth anomaly feature vector at the k-th time step, To reconstruct the target vector, is the flag function for whether the kth time step is abnormal, ‖z k ‖1 is the regularization term for the abnormal feature sparsity, λ s is the sparsity adjustment coefficient; S45. The residual network output of the behavioral attention guidance module, cross-scale residual fusion features and abnormal sparse perception module is used as the final abnormality discrimination input to form an enhanced deep abnormality feature sequence Z enhanced .

7. The method for identifying abnormal behavior patterns of livestock and poultry based on a distribution estimation algorithm and a residual network according to claim 6, characterized in that: The S5 comprises the following steps: S51. Enhanced deep anomaly feature sequence Input behavior classification and discrimination module, using soft classification method to calculate each time step t k Corresponding abnormal behavior probability in, is the deep abnormal feature vector extracted by the residual neural network model at the kth time step, which contains the multimodal behavior fusion representation of livestock and poultry at the current moment, W o is the classification weight vector, which is used to transform the deep anomaly feature vector Mapped to scalar space, b o is the bias term, exp(·) is the natural exponential function, and together with 1+exp(·) in the denominator, it forms the logistic function; S52. Set the classification threshold θ based on the probability of abnormal behavior p , when the probability of abnormal behavior When , the time walk behavior is marked as abnormal behavior, and the abnormal behavior mark set A is generated. flag , all unmarked time steps correspond to behaviors that default to normal behaviors; S53. Combine each abnormal time step t k ∈A flag The collection timestamp and the spatial positioning information of the original livestock and poultry individuals or groups are used to perform time series retrospective analysis, construct the behavioral evolution trajectory of livestock and poultry individuals or groups within a certain time window before and after the abnormality occurs, and form a set of abnormal behavior trajectories; S54. Based on the abnormal behavior trajectory set and the spatial position data L corresponding to each abnormal behavior time step k ,construct a set of spatial position mappings and perform statistical analysis on the spatial density of abnormal points to reveal the spatiotemporal aggregation characteristics of abnormal behaviors; S55. The abnormal behavior determination result is integrated with the abnormal behavior trajectory set, and the abnormal behavior of each time step is graded according to the preset abnormal level classification rules.

8. The method for identifying abnormal behavior patterns of livestock and poultry based on a distribution estimation algorithm and a residual network according to claim 7, characterized in that: The abnormality classification rules refer to the following standards: Level 1 anomaly: probability of abnormal behavior Continuous anomalies occur ≥3 times within the corresponding trajectory range, and the anomalies form a high-density cluster in space, and the number of anomalies in the local area within 5 minutes is ≥5; Secondary anomaly: probability of abnormal behavior There are 2 to 3 abnormal points within the trajectory range, and the spatial distribution is locally concentrated or periodic; Level 3 abnormality: probability of abnormal behavior There are single abnormal points or random distribution in the trajectory, and no clustering characteristics are formed.

Images