A method and system for intelligent risk assessment of brucellosis for large-scale dairy farms

Abstract

The application relates to the technical field of artificial intelligence. A method and system for intelligent risk assessment of brucellosis in a large-scale dairy farm are provided, wherein the method comprises the following steps: collecting multi-source heterogeneous proxy data related to biological safety regulation execution through Internet of Things terminals and visual sensing devices deployed in key areas of the dairy farm; performing computer vision behavior recognition and process mining on the multi-source heterogeneous proxy data to obtain dynamic quantitative indexes of biological safety compliance; performing multi-source data fusion on the dynamic quantitative indexes of biological safety compliance to obtain a biological safety compliance index BCI; and injecting the biological safety compliance index BCI as a time-varying risk weight into a preset brucellosis transmission model to generate a dynamic risk assessment result fused with human risk, so that the technical effects of improving dynamic capture capability of human risk, enhancing multi-source data fusion accuracy and improving real-time risk assessment are achieved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an intelligent risk assessment method and system for brucellosis in large-scale dairy farms. Background Technology

[0002] With the deep application of artificial intelligence technology in the field of agricultural biosafety, the risk assessment of brucellosis in large-scale dairy farms increasingly relies on intelligent monitoring and dynamic quantitative methods. As a zoonotic disease, the transmission risk of brucellosis is highly dependent on the quality of implementation of biosafety procedures, such as the standardization of personnel disinfection operations and the compliance of protective equipment wearing.

[0003] Traditional techniques employ manual observation and recording, where regulatory personnel record the execution of biosafety procedures on-site. However, this method suffers from strong subjectivity, limited coverage, and poor real-time performance, making it difficult to comprehensively capture dynamic risks. Single-sensor monitoring technologies, such as RFID (Radio Frequency Identification) or simple visual inspection devices, collect localized data. However, these technologies suffer from isolated data sources and a lack of behavioral semantic understanding, failing to construct complete operational trajectory sequences. Furthermore, static weighted risk assessment models use historical average compliance rates as fixed weights input into brucellosis transmission models, while simple data splicing and fusion methods directly weight and average multi-source data to generate compliance indicators. The former ignores the time-varying characteristics of human actions, resulting in delayed risk assessment and insufficient accuracy, while the latter fails to address spatiotemporal heterogeneity and feature interaction issues, causing key risk signals to be drowned out by noise. Summary of the Invention

[0004] Therefore, it is necessary to provide a method and system for intelligent risk assessment of brucellosis in large-scale dairy farms to address the aforementioned technical problems, so as to improve the ability to dynamically capture human-caused risks, enhance the accuracy of multi-source data fusion, and improve the real-time performance of risk assessment.

[0005] Firstly, this application provides a method for intelligent risk assessment of brucellosis in large-scale dairy farms, the method comprising:

[0006] By deploying IoT terminals and visual sensing devices in key areas of dairy farms, multi-source heterogeneous proxy data related to the execution of biosafety procedures is collected.

[0007] Computer vision behavior recognition and process mining were performed on multi-source heterogeneous agent data to obtain dynamic quantitative indicators of biosafety compliance.

[0008] The Biosafety Compliance Index (BCI) was obtained by fusing multi-source data on dynamic quantitative indicators of biosafety compliance.

[0009] The Biosafety Compliance Index (BCI) is used as a time-varying risk weight and injected into a pre-defined brucellosis transmission model to generate a dynamic risk assessment result that incorporates human-caused risks.

[0010] In one embodiment, computer vision behavior recognition and process mining are performed on multi-source heterogeneous proxy data to obtain dynamic quantitative indicators of biosafety compliance, including:

[0011] Spatiotemporal behavioral maps were constructed from video stream data in multi-source heterogeneous proxy data to obtain behavioral trajectory sequences of key biosafety operations.

[0012] An attention-based action compliance assessment is performed on the behavioral trajectory sequence to generate a multi-dimensional visual compliance feature vector.

[0013] Perform process variation structure analysis on log data in multi-source heterogeneous proxy data to extract spatiotemporal offset features of process nodes;

[0014] By fusing multi-dimensional visual compliance feature vectors with spatiotemporal offset features of process nodes across modalities, a dynamic quantitative index of biosafety compliance is obtained.

[0015] In one embodiment, multi-dimensional visual compliance feature vectors and spatiotemporal offset features of process nodes are fused across modalities to obtain dynamic quantitative indicators of biosafety compliance, including:

[0016] Spatiotemporal attention enhancement is performed on multi-dimensional visual compliance feature vectors to generate attention-weighted visual feature tensors;

[0017] Topological structure embedding is performed on the spatiotemporal offset features of process nodes to obtain a topology-enhanced process feature matrix;

[0018] Heterogeneous feature interaction learning is performed on the attention-weighted visual feature tensor and the topology-enhanced process feature matrix to generate cross-modal fusion features;

[0019] Dynamic compliance quantification mapping is performed on cross-modal fusion features to obtain a dynamic quantitative index of biosafety compliance. The expression for the dynamic quantitative index of biosafety compliance is as follows:

[0020]

[0021] in, This represents a dynamic quantitative indicator of biosafety compliance. Represents the importance weights between modes. Represents the element-level Hadamard product. Represents the tensor outer product. Indicates the feature fusion operator, This represents the feature selection function. Represents the feature transformation function. This represents the dynamic quantization mapping function. Indicates the number of modal interaction channels. Indicates the first Attention-weighted visual feature tensor of the channel Indicates the first The topology enhancement process feature matrix of the channel. This represents the visual feature tensor after spatiotemporal attention enhancement. This represents the process feature matrix after topology embedding. Indicates the channel index.

[0022] In one embodiment, a spatiotemporal behavioral map is constructed from video stream data in multi-source heterogeneous proxy data to obtain a behavioral trajectory sequence of key biosafety operations, including:

[0023] By detecting key operation nodes in video stream data, a set of key biosafety action segments is obtained;

[0024] Cross-frame spatiotemporal correlation analysis was performed on a set of key biosafety action segments to obtain the trajectory of action continuity;

[0025] Behavioral semantic encoding is performed on the continuous trajectory of actions to generate a sequence of behavioral trajectories for critical biosafety operations. The expression for the sequence of behavioral trajectories is as follows:

[0026]

[0027] in, Represents a sequence of behavioral trajectories. Indicates the first A segment of key biosafety actions, Represents a set of adjacent segments within a spatiotemporal neighborhood. This indicates a feature concatenation operation. Indicates the aggregation of neighborhood features. This represents a Long Short-Term Memory (LSTM) network encoder. Represents the sequence aggregation operator, Indicates the first A neighborhood behavior fragment, Indicates the neighborhood fragment index, This indicates the total number of critical biosafety action segments. Indicates the index of the behavior segment currently being processed.

[0028] In one embodiment, the Biosafety Compliance Index (BCI) is used as a time-varying risk weight and injected into a pre-defined brucellosis transmission model to generate a dynamic risk assessment result that incorporates human-caused risks, including:

[0029] Risk transmission path mapping is performed on the Biosafety Compliance Index (BCI) to generate a spatiotemporal dynamic weight distribution;

[0030] By coupling the spatiotemporal dynamic weight distribution with the static environmental risk factors in the brucellosis transmission model through multipath risk, a coupled risk field is obtained.

[0031] Propagation dynamics simulation of the coupled risk field yields dynamic risk assessment results that incorporate human-caused risks.

[0032] In one embodiment, the spatiotemporal dynamic weight distribution is coupled with the static environmental risk factors in the brucellosis transmission model through multipath risk to obtain a coupled risk field, including:

[0033] Path sensitivity analysis is performed on the spatiotemporal dynamic weight distribution to obtain a multi-path dynamic weight vector set;

[0034] Path-adaptive mapping is performed on static environmental risk factors to generate a path-aligned static risk base.

[0035] Nonlinear path coupling is performed between the multi-path dynamic weight vector set and the path-aligned static risk basis to obtain the coupled risk field.

[0036] In one embodiment, multi-source data fusion is performed on the dynamic quantitative indicators of biosafety compliance to obtain the Biosafety Compliance Index (BCI), including:

[0037] Time-domain alignment processing is performed on heterogeneous data streams in the dynamic quantitative indicators of biosafety compliance to generate a synchronized compliance feature set;

[0038] The synchronized compliance feature set is weighted based on the importance of features using an attention mechanism to obtain a dynamically weighted feature matrix;

[0039] The Biosafety Compliance Index (BCI) is obtained by nonlinear feature compression of the dynamically weighted feature matrix.

[0040] Secondly, this application also provides an intelligent risk assessment system for brucellosis in large-scale dairy farms, the system comprising:

[0041] The multi-source agent data acquisition module is used to collect multi-source heterogeneous agent data related to the execution of biosafety procedures through IoT terminals and visual sensing devices deployed in key areas of dairy farms.

[0042] The behavior process mining module is used to perform computer vision behavior recognition and process mining on multi-source heterogeneous agent data to obtain dynamic quantitative indicators of biosafety compliance.

[0043] The BCI fusion generation module is used to fuse multi-source data on the dynamic quantitative indicators of biosafety compliance to obtain the biosafety compliance index (BCI).

[0044] The dynamic risk assessment module uses the Biosafety Compliance Index (BCI) as a time-varying risk weight, injects it into a pre-defined brucellosis transmission model, and generates a dynamic risk assessment result that incorporates human-caused risks.

[0045] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods in the first aspect of this application.

[0046] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the steps of any of the methods in the first aspect of this application.

[0047] This application provides a method and system for intelligent risk assessment of brucellosis in large-scale dairy farms. The method includes: collecting multi-source heterogeneous agent data related to the execution of biosafety procedures through IoT terminals and visual sensing devices deployed in key areas of the dairy farm, providing multi-dimensional foundational support for subsequent correlation analysis of human behavior and environmental risks; then performing computer vision behavior recognition and process mining on the multi-source heterogeneous agent data, extracting action compliance and process spatiotemporal offset features from video streams and log data, and generating dynamic quantitative indicators of biosafety compliance, which can more sensitively capture dynamic risk fluctuations in personnel operations, thereby improving the ability to dynamically capture human risks; and then performing multi-source data fusion on the dynamic quantitative indicators of biosafety compliance, generating a biosafety compliance index (BCI) through temporal alignment, attention weighting, and nonlinear compression, effectively integrating key risk signals in heterogeneous data, reducing noise interference, and thus enhancing the accuracy of multi-source data fusion.

[0048] By incorporating the Biosafety Compliance Index (BCI) as a time-varying risk weight into a pre-defined brucellosis transmission model, the risk weight can be adjusted in real time according to the dynamic changes in personnel compliance, rather than relying on static historical data. This improves the real-time nature of risk assessment and helps large-scale dairy farms to more accurately perceive and manage the risk of brucellosis transmission. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying 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.

[0050] Figure 1 A flowchart of an intelligent risk assessment method for brucellosis in large-scale dairy farms, as described in one embodiment of the present invention;

[0051] Figure 2 This is a flowchart illustrating the process of fusing multi-source data to obtain the Biosafety Compliance Index (BCI) from dynamic quantitative indicators of biosafety compliance in one embodiment of the present invention.

[0052] Figure 3 This is a structural diagram of an intelligent risk assessment system for brucellosis in large-scale dairy farms, according to one embodiment of the present invention. Detailed Implementation

[0053] To make the above-mentioned objects, features, and advantages of this application more apparent and understandable, the specific embodiments of this application will be described in detail below with reference to the accompanying drawings. Many specific details are set forth in the following description to provide a thorough understanding of this application. However, this application can be implemented in many other ways different from those described herein, and those skilled in the art can make similar modifications without departing from the spirit of the application. Therefore, this application is not limited to the specific embodiments disclosed below.

[0054] First, the application scenarios of the embodiments of this application will be described. In the embodiments of this application, a method and system for intelligent risk assessment of brucellosis in large-scale dairy farms is provided, applicable to, but not limited to, daily biosecurity risk management and control, routine monitoring of disease prevention and control in dairy farms, and disease risk classification and early warning scenarios in the livestock industry.

[0055] In illustrative purposes, the intelligent risk assessment method and system for brucellosis in large-scale dairy farms provided in this application embodiment can also be applied to other large-scale livestock and poultry breeding bases for zoonotic disease risk assessment, cross-field joint prevention and control of diseases in livestock breeding parks, and biosafety compliance verification of livestock and poultry breeding. This is only an example and does not limit the specific application scenarios.

[0056] In this application, the full name of the Biosafety Compliance Index (BCI) is the Biological Safety Compliance Index, which is a comprehensive index formed by the fusion of multi-source data on the dynamic quantitative indicators of biosafety compliance by the biosafety assessment terminal.

[0057] like Figure 1 As shown, this application provides a method for intelligent risk assessment of brucellosis in large-scale dairy farms, the method comprising:

[0058] S101: Collect multi-source heterogeneous proxy data related to the execution of biosafety procedures through IoT terminals and visual sensing devices deployed in key areas of dairy farms.

[0059] For example, based on the core needs of biosafety management in large-scale dairy farms, the biosafety assessment terminal defines the specific scope of key areas within the farm, ensuring that this scope fully covers the core operational scenarios for implementing biosafety procedures. Following pre-defined equipment deployment specifications, the biosafety assessment terminal installs IoT terminals and visual sensing devices in these key areas, ensuring that the data collection range of both types of devices fully covers biosafety-related operations and environmental conditions within the key areas, thus guaranteeing the integrity and effectiveness of data collection.

[0060] The biosafety assessment terminal coordinates with IoT terminals to collect non-visual data such as environmental parameters and equipment operating status related to the implementation of biosafety procedures. Simultaneously, it coordinates with visual sensing devices to collect visual data such as the behavior of personnel performing biosafety operations and the wearing of protective equipment in key areas. The biosafety assessment terminal performs correlation filtering on all data collected by both types of devices, retaining only data directly related to the implementation of biosafety procedures. Then, it initially integrates the filtered data from different sources and of different types to obtain multi-source heterogeneous proxy data related to the implementation of biosafety procedures.

[0061] S102: Perform computer vision behavior recognition and process mining on multi-source heterogeneous agent data to obtain dynamic quantitative indicators of biosafety compliance.

[0062] For example, the biosafety assessment terminal performs computer vision behavior recognition processing on video stream data in multi-source heterogeneous proxy data, separates biosafety-related action segments through key operation node detection, constructs a complete behavior trajectory through cross-frame spatiotemporal correlation analysis, and then extracts visual features related to action compliance.

[0063] The biosafety assessment terminal performs process mining on log data from multi-source heterogeneous proxy data, analyzing the process execution information contained in the log data, identifying the spatiotemporal offset of process nodes, and extracting corresponding process features. The biosafety assessment terminal then integrates the extracted visual features with the process features across modalities, eliminating adaptation differences between heterogeneous features and strengthening the expression of key risk-related features. Subsequently, the integrated features undergo dynamic compliance quantification processing, generating indicators that reflect the real-time status of biosafety procedure execution through feature mapping and compliance judgment, resulting in a dynamic quantitative indicator of biosafety compliance.

[0064] S103: Multi-source data fusion is performed on the dynamic quantitative indicators of biosafety compliance to obtain the Biosafety Compliance Index (BCI).

[0065] For example, the biosafety assessment terminal performs time-domain alignment on feature data from different sources in the dynamic quantitative indicators of biosafety compliance, unifying the time dimension based on the time series of biosafety procedure execution and eliminating time bias. The biosafety assessment terminal then weights the time-domain aligned dynamic quantitative indicators of biosafety compliance according to feature importance, assigning weight coefficients based on the degree of impact of each indicator on compliance, thereby strengthening the expression of key risk indicators.

[0066] The biosafety assessment terminal performs non-linear compression on the weighted feature data, eliminating redundant information and integrating it into a feature set with a unified dimension. The terminal then performs quantification mapping and normalization on the integrated feature set, transforming the multi-dimensional data into a single index that comprehensively reflects the overall level of biosafety protocol implementation, resulting in the Biosafety Compliance Index (BCI).

[0067] S104: The Biosafety Compliance Index (BCI) is used as a time-varying risk weight and injected into a pre-defined brucellosis transmission model to generate a dynamic risk assessment result that incorporates human-caused risks.

[0068] For example, the biosafety assessment terminal performs risk transmission path matching on the Biosafety Compliance Index (BCI), clarifies the human risk impact path in the brucellosis transmission process corresponding to the BCI, and establishes the application dimensions of time-varying risk weights. The biosafety assessment terminal retrieves a preset brucellosis transmission model and couples the established time-varying risk weights with the static environmental risk factors in the brucellosis transmission model in multiple dimensions, eliminating the adaptation differences between dynamic weights and static factors.

[0069] Based on coupled risk parameters, the biosafety assessment terminal drives a brucellosis transmission model to simulate transmission dynamics, reconstructing the disease transmission pattern under the combined effects of human and environmental risks. The terminal quantifies and dynamically characterizes the simulation results, integrating them to form assessment conclusions that reflect real-time risk status, generating dynamic risk assessment results that incorporate human risk.

[0070] One embodiment of this application provides a method for intelligent risk assessment of brucellosis in large-scale dairy farms, comprising: collecting multi-source heterogeneous agent data related to the execution of biosafety procedures through IoT terminals and visual sensing devices deployed in key areas of the dairy farm, providing multi-dimensional basic support for subsequent correlation analysis of human behavior and environmental risks; then performing computer vision behavior recognition and process mining on the multi-source heterogeneous agent data, extracting action compliance and process spatiotemporal offset features from video streams and log data, generating a dynamic quantitative index of biosafety compliance, which can more sensitively capture dynamic risk fluctuations in personnel operations, thereby improving the ability to dynamically capture human risks; then performing multi-source data fusion on the dynamic quantitative index of biosafety compliance, generating a biosafety compliance index (BCI) through temporal alignment, attention weighting, and nonlinear compression, effectively integrating key risk signals in heterogeneous data, reducing noise interference, and thus enhancing the accuracy of multi-source data fusion.

[0071] By incorporating the Biosafety Compliance Index (BCI) as a time-varying risk weight into a pre-defined brucellosis transmission model, the risk weight can be adjusted in real time according to the dynamic changes in personnel compliance, rather than relying on static historical data. This improves the real-time nature of risk assessment and helps large-scale dairy farms to more accurately perceive and manage the risk of brucellosis transmission.

[0072] In one embodiment, computer vision behavior recognition and process mining are performed on multi-source heterogeneous proxy data to obtain dynamic quantitative indicators of biosafety compliance, including:

[0073] (1) Spatiotemporal behavior maps were constructed from video stream data in multi-source heterogeneous proxy data to obtain the behavior trajectory sequence of key biosafety operations.

[0074] For example, the biosafety assessment terminal extracts video stream data from multi-source heterogeneous proxy data, performs image preprocessing on the video stream data to remove environmental interference and invalid pixel information, and optimizes data quality; then it calls the preset biosafety key operation node detection algorithm to identify the core operation nodes in the video stream that correspond to the biosafety procedures, and filters out continuous video segments containing the core operation nodes; subsequently, it performs frame-by-frame feature extraction on the filtered video segments to capture the spatial morphological features and temporal evolution patterns of the operation actions in each frame.

[0075] The biosafety assessment terminal establishes feature matching relationships and logical connection paths for operational actions between different video frames through a cross-frame spatiotemporal correlation algorithm, forming a continuous and complete action flow link. It adopts a behavioral semantic coding model to structurally transform the action flow link, encoding unstructured video action information into semantically related sequence data, completing the construction of a spatiotemporal behavioral map of key biosafety operations, and obtaining the behavioral trajectory sequence of key biosafety operations.

[0076] Among them, the cross-frame spatiotemporal association algorithm includes the adjacent video frame action feature matching algorithm, the non-continuous frame action logical association path construction algorithm, the action spatiotemporal consistency verification algorithm, and the cross-frame action evolution pattern mining algorithm; the behavior semantic coding model includes the action behavior label mapping model, the unstructured action video information structured transformation model, the action sequence semantic association coding model, and the behavior compliance semantic judgment model; the action flow link includes the action connection link of different operation nodes, the action temporal evolution link of the same operation, and the cross-frame action feature association matching link; the behavior trajectory sequence includes the semantic coding sequence of key operation actions, the action spatiotemporal feature association sequence, the temporal evolution sequence of operation nodes, and the behavior compliance state change sequence.

[0077] (2) Perform an attention-based action compliance assessment on the behavior trajectory sequence and generate a multi-dimensional visual compliance feature vector.

[0078] For example, after the biosafety assessment terminal acquires the behavioral trajectory sequence of key biosafety operations, it expands the feature dimensions of the behavioral trajectory sequence, supplements the relevant derived features of the action execution, and enriches the feature expression dimensions of the sequence; then, through the attention weight calculation algorithm, it analyzes the priority of the impact of each action segment in the behavioral trajectory sequence on biosafety compliance, and determines the key assessment action segments with high impact weight; then, it retrieves the preset biosafety operation compliance standard library, which contains the standardized execution process, action detail requirements and judgment criteria for various key biosafety operations.

[0079] The biosafety assessment terminal will focus on comparing and analyzing the action segments with the corresponding regulatory requirements in the compliance standard library step by step and in detail to identify the types, locations, and degrees of compliance deviations during the action execution process. Based on the deviation analysis results, it will extract compliance features covering multiple dimensions such as action standardization, execution completeness, timing rationality, and operational standard conformity. Then, it will quantify the compliance features of each dimension and assign corresponding feature values. Finally, it will orderly integrate the multi-dimensional quantitative features according to preset feature combination rules to generate a multi-dimensional visual compliance feature vector.

[0080] (3) Perform process variation structure analysis on log data in multi-source heterogeneous proxy data and extract spatiotemporal offset features of process nodes.

[0081] For example, the biosafety assessment terminal extracts log data from multi-source heterogeneous proxy data, performs format standardization processing on the log data, converts heterogeneous log data generated by different acquisition devices into a unified data format, and eliminates the analysis obstacles caused by format differences; it extracts core fields from the standardized log data, and filters out key data related to the execution of biosafety procedures, such as node identifiers, execution time, execution subject, operation content, and node connection information; it retrieves a preset biosafety standard procedure template, and clarifies the preset node sequence, execution requirements of each node, logical relationships between nodes, and standard execution time.

[0082] The biosafety assessment terminal compares the actual process execution information in the log data with the biosafety standard process template node by node, identifying differences between the actual process and the standard process in terms of node sequence, execution time, node connection, and operation content. For the identified differences, it performs process variation structure analysis to determine the type of variation, its scope, and its impact on the execution of the biosafety process. At the same time, it tracks the specific manifestations of process nodes deviating from the preset execution time period in the time dimension and from the preset execution area in the spatial dimension. Based on the process variation structure analysis results and the spatiotemporal deviation of nodes, it extracts feature information that can characterize process execution deviations and extracts the spatiotemporal offset features of process nodes.

[0083] (4) The multi-dimensional visual compliance feature vector and the spatiotemporal offset feature of the process node are fused across modal features to obtain a dynamic quantitative index of biosafety compliance.

[0084] For example, the biosafety assessment terminal calibrates the feature dimensions of the multi-dimensional visual compliance feature vector, adjusting the dimensional specifications and numerical range of the features to ensure the consistency and rationality of the feature dimensions; it optimizes the data structure of the spatiotemporal offset features of process nodes, adjusting the organization and expression format of the features to adapt them to the multi-dimensional visual compliance feature vector; it establishes a mapping relationship between the multi-dimensional visual compliance feature vector and the spatiotemporal offset features of process nodes through a heterogeneous feature adaptation algorithm, eliminating the heterogeneity differences between the two types of features in terms of data structure and expression; and it initiates a feature interaction learning model to strengthen the correlation and fusion between the two types of features through bidirectional feature transfer and information complementarity, thereby mining the potential correlation information between visual compliance features and process offset features.

[0085] The biosafety assessment terminal integrates the feature information after interactive learning to form cross-modal fusion features that reflect the implementation status of biosafety procedures. Then, redundant information is removed from the cross-modal fusion features, retaining the core effective features to improve the relevance and effectiveness of the features. Through a dynamic compliance quantification algorithm, quantification rules and weight allocation schemes that meet the needs of biosafety assessment are set to transform the cross-modal fusion features into quantitative values ​​that can characterize the real-time compliance level. The quantitative values ​​are standardized to unify the value range and expression form, resulting in a dynamic quantitative index of biosafety compliance.

[0086] Among them, the feature interaction learning model is a feature fusion model applied to cross-modal feature fusion scenarios for biosafety compliance assessment. It has a built-in heterogeneous feature mapping unit and a bidirectional feature transfer mechanism, which can realize bidirectional information complementarity and correlation enhancement between multi-dimensional visual compliance feature vectors and spatiotemporal offset features of process nodes, thereby mining potential correlation information between the two types of heterogeneous features.

[0087] In one embodiment, multi-dimensional visual compliance feature vectors and spatiotemporal offset features of process nodes are fused across modalities to obtain dynamic quantitative indicators of biosafety compliance, including:

[0088] (1) Spatiotemporal attention enhancement is performed on multidimensional visual compliance feature vectors to generate attention-weighted visual feature tensors.

[0089] For example, the biosafety assessment terminal acquires multi-dimensional visual compliance feature vectors, performs denoising and normalization preprocessing on these vectors to eliminate data redundancy and dimensional differences; it then performs spatiotemporal feature decomposition on the preprocessed multi-dimensional visual compliance feature vectors to separate the sequential changes of visual features in the time dimension and the regional distribution information in the spatial dimension; it introduces a spatiotemporal attention mechanism, using an attention weight calculation algorithm to analyze the degree of influence of visual features in different time intervals and spatial regions on biosafety compliance, assigning higher attention weights to features with high influence; it then reconstructs the weighted temporal and spatial features into tensors, integrating the spatiotemporal correlation information to form a high-dimensional tensor structure containing spatiotemporal attention weights; finally, it performs dimensional verification on the reconstructed tensors to ensure the consistency and rationality of the tensor dimensions, generating an attention-weighted visual feature tensor.

[0090] Spatiotemporal attention enhancement includes core steps such as multi-dimensional visual compliance feature vector preprocessing, spatiotemporal dimension feature decomposition, attention weight calculation, feature weighting, tensor reconstruction, and dimension verification, which are used to strengthen the visual compliance feature expression of key spatiotemporal regions.

[0091] (2) The spatiotemporal offset features of process nodes are embedded with topology structure to obtain the topology-enhanced process feature matrix.

[0092] For example, the biosafety assessment terminal acquires the spatiotemporal offset features of process nodes, cleans the spatiotemporal offset features of process nodes to remove invalid and abnormal offset records; performs topological structure analysis on the cleaned spatiotemporal offset features of process nodes to identify the predecessor and successor relationships, parallel execution paths and dependency constraint logic between process nodes; through a topology embedding algorithm, the parsed topology structure information is mapped to a high-dimensional feature space and transformed into a matrix form containing node spatiotemporal offset attributes and topological association information; the embedded feature matrix is ​​normalized to adjust the matrix dimension and feature arrangement to optimize the matrix's computational efficiency and expressive power; the normalized matrix is ​​feature-verified to ensure that the matrix can completely retain the topological association and spatiotemporal offset features of the process, resulting in a topology-enhanced process feature matrix.

[0093] The topology embedding process includes core steps such as cleaning of spatiotemporal offset feature data of process nodes, topology parsing, high-dimensional embedding of topology information, feature matrix normalization, and feature verification. These steps are used to integrate the topological association information of the process into the feature matrix representation.

[0094] (3) Perform heterogeneous feature interaction learning on the attention-weighted visual feature tensor and the topology enhancement process feature matrix to generate cross-modal fusion features.

[0095] For example, the biosafety assessment terminal acquires an attention-weighted visual feature tensor and a topology-enhanced process feature matrix. It then compresses the dimension of the attention-weighted visual feature tensor and adjusts its dimensionality to match the dimension of the topology-enhanced process feature matrix. Next, it expands the dimension of the topology-enhanced process feature matrix to align it with the attention-weighted visual feature tensor. Through a cross-modal attention mechanism, it establishes a mapping between the attention-weighted visual feature tensor and the topology-enhanced process feature matrix, mining complementary information between visual and process features. Finally, it initiates a heterogeneous feature interaction learning model, using bidirectional feature transfer and iterative information updates to guide and strengthen the association between the two types of features, generating intermediate features containing cross-modal information. The intermediate features are then integrated and deredundanted, retaining core effective features to generate cross-modal fusion features.

[0096] Among them, heterogeneous feature interaction learning includes core steps such as heterogeneous feature dimension alignment, cross-modal association mapping, bidirectional feature transfer, information iterative update, intermediate feature integration and redundant information removal, which are used to achieve deep interaction and fusion of visual features and process features.

[0097] (4) Dynamic compliance quantification mapping is performed on the cross-modal fusion features to obtain the dynamic quantification index of biosafety compliance. The expression of the dynamic quantification index of biosafety compliance is:

[0098]

[0099] in, This represents a dynamic quantitative indicator of biosafety compliance. Represents the importance weights between modes. Represents the element-level Hadamard product. Represents the tensor outer product. Indicates the feature fusion operator, This represents the feature selection function. Represents the feature transformation function. This represents the dynamic quantization mapping function. Indicates the number of modal interaction channels. Indicates the first Attention-weighted visual feature tensor of the channel Indicates the first The topology enhancement process feature matrix of the channel. This represents the visual feature tensor after spatiotemporal attention enhancement. This represents the process feature matrix after topology embedding. Indicates the channel index.

[0100] For example, the biosafety assessment terminal acquires cross-modal fusion features. A feature selection function is used to filter these features, retaining core features highly relevant to biosafety compliance and removing irrelevant and redundant features. A feature transformation function is used to spatially transform the filtered core features, optimizing their representation and improving their discriminative power. Inter-modal importance weights are introduced to weighted aggregate the fusion features from different channels, strengthening the feature contribution of high-weight channels. Element-level Hadamard product operators are used to process feature interactions within channels, and tensor outer product operators are used to process feature interactions between channels, generating multi-scale interactive features. Then, a feature fusion operator integrates the multi-source interactive features to form a unified fusion feature expression. Finally, a dynamic quantization mapping function transforms the integrated features into quantitative values ​​that characterize the real-time state of biosafety compliance, resulting in a dynamic quantitative index of biosafety compliance.

[0101] The dynamic compliance quantification mapping includes core steps such as feature selection, feature transformation, intermodal weighted aggregation, multi-operator feature interaction, feature fusion, and dynamic quantification mapping, which are used to transform cross-modal fused features into standardized dynamic quantification indicators of biosafety compliance.

[0102] In one embodiment, a spatiotemporal behavioral map is constructed from video stream data in multi-source heterogeneous proxy data to obtain a behavioral trajectory sequence of key biosafety operations, including:

[0103] (1) Detect key operation nodes in the video stream data to obtain a set of biosafety key action segments.

[0104] For example, the biosafety assessment terminal extracts video stream data from multi-source heterogeneous proxy data, performs image preprocessing on the video stream data, removes environmental noise through Gaussian filtering, enhances image contrast through histogram equalization, and unifies the range of image pixel values ​​through pixel normalization to optimize the visual quality and feature recognition of video frames; it then calls a preset biosafety critical operation node detection model, which is built based on a target detection algorithm, traverses each frame of the video stream, identifies the start and end nodes of the core actions corresponding to the biosafety procedures, and outputs the coordinate position and timestamp information of the nodes in the image; subsequently, based on the timestamp and coordinate position of the nodes, it extracts a continuous frame sequence containing complete critical actions from the video stream, with each continuous frame sequence corresponding to a biosafety critical action segment, and all segments are combined to form a biosafety critical action segment set.

[0105] The key operation node detection includes core steps such as Gaussian filtering and noise reduction of video stream data, histogram equalization enhancement, pixel normalization preprocessing, inference of biosafety key operation node detection model, node coordinate and timestamp output, and key action segment extraction, which are used to accurately locate and extract effective action segments related to biosafety from unstructured video streams.

[0106] (2) Perform cross-frame spatiotemporal correlation analysis on the set of biosafety key action segments to obtain the action continuity trajectory.

[0107] For example, the biosafety assessment terminal acquires a set of biosafety key action segments, extracts spatiotemporal features for each segment, extracts optical flow features to capture motion changes in the time dimension, extracts human posture features to capture morphological information in the spatial dimension, and extracts action bounding box features to capture the spatial position of the action. Through a cross-frame spatiotemporal association algorithm, the sequence of different segments is matched based on timestamps, and the rationality of action connection between segments is judged based on the degree of spatial position overlap, establishing a logical association mapping between segments. The associated biosafety key action segments are then sequentially linked according to the temporal order of action execution and spatial logic to construct an action continuity trajectory that includes the complete link of action start, evolution, and end.

[0108] The cross-frame spatiotemporal correlation analysis includes core steps such as segment optical flow feature extraction, human pose feature extraction, action bounding box feature extraction, timestamp matching, spatial position overlap calculation, construction of logical association mapping between segments, and orderly concatenation of action trajectories, which are used to integrate discrete key action segments into a continuous action execution link.

[0109] (3) Perform behavioral semantic encoding on the action continuity trajectory to generate a behavioral trajectory sequence for critical biosafety operations. The expression for the behavioral trajectory sequence is:

[0110]

[0111] in, Represents a sequence of behavioral trajectories. Indicates the first A segment of key biosafety actions, Represents a set of adjacent segments within a spatiotemporal neighborhood. This indicates a feature concatenation operation. Indicates the aggregation of neighborhood features. This represents a Long Short-Term Memory (LSTM) network encoder. Represents the sequence aggregation operator, Indicates the first A neighborhood behavior fragment, Indicates the neighborhood fragment index, This indicates the total number of critical biosafety action segments. Indicates the index of the behavior segment currently being processed.

[0112] For example, the biosafety assessment terminal acquires the action continuity trajectory, performs feature splicing on each biosafety key action segment in the action continuity trajectory, aligns the core features of the current biosafety key action segment with the features of adjacent segments in the spatiotemporal neighborhood in terms of dimension, and splices them into a high-dimensional feature vector through channel fusion; performs neighborhood feature aggregation on the spliced ​​features, and strengthens the feature representation of the neighborhood with high correlation to the current action through an attention-weighted averaging method; inputs the aggregated features into a long short-term memory network encoder, which captures the long-term temporal dependencies of the action trajectory through a gating mechanism and outputs the semantic encoding vector of each segment; through a sequence aggregation operator, the semantic encoding vectors of all segments are globally pooled and feature spliced, and integrated into a unified structured sequence representation to generate a sequence of behavioral trajectories for key biosafety operations.

[0113] Among them, behavioral semantic encoding includes core steps such as feature dimension alignment, channel fusion and splicing, attention-weighted neighborhood feature aggregation, long short-term memory network gating encoding, semantic encoding vector output, global pooling and feature splicing sequence aggregation, which are used to transform continuous action trajectories into structured sequence data with clear semantic associations.

[0114] In one embodiment, the Biosafety Compliance Index (BCI) is used as a time-varying risk weight and injected into a pre-defined brucellosis transmission model to generate a dynamic risk assessment result that incorporates human-caused risks, including:

[0115] (1) The risk transmission path is mapped to the Biosafety Compliance Index (BCI) to generate a spatiotemporal dynamic weight distribution.

[0116] For example, the biosafety assessment terminal acquires the Biosafety Compliance Index (BCI), retrieves a pre-defined knowledge base of brucellosis risk transmission pathways, and identifies risk transmission pathway types such as airborne transmission, contact transmission, and vector transmission within the knowledge base. It then performs feature decomposition of the BCI in both time and spatial dimensions, separating the fluctuation characteristics of the BCI over time and its distribution characteristics in spatial regions. The decomposed spatiotemporal features are matched one by one with each type of risk transmission pathway, calculating the risk weight coefficient corresponding to each pathway. A high BCI corresponds to a low risk weight coefficient, and a low BCI corresponds to a high risk weight coefficient. The risk weight coefficients are then distributed in a gridded manner according to spatiotemporal units, with each spatiotemporal unit corresponding to a unique weight coefficient. Finally, the risk weight coefficients of all spatiotemporal units are integrated to generate a dynamic spatiotemporal weight distribution.

[0117] The risk transmission path mapping includes core steps such as retrieval of the brucellosis risk transmission path knowledge base, identification of risk transmission path types, spatiotemporal feature decomposition of the Biosafety Compliance Index (BCI), feature and path matching, calculation of risk weight coefficients, spatiotemporal unit grid allocation, and weight integration. These steps are used to transform the Biosafety Compliance Index (BCI) into dynamic risk weights with spatiotemporal attributes.

[0118] (2) The spatiotemporal dynamic weight distribution is coupled with the static environmental risk factors in the brucellosis transmission model through multipath risk coupling to obtain the coupled risk field.

[0119] For example, the biosafety assessment terminal acquires the spatiotemporal dynamic weight distribution and the static environmental risk factors in the brucellosis transmission model, aligns the two along the transmission path dimension to ensure a complete match between the path dimension of the spatiotemporal dynamic weight distribution and the path dimension of the static environmental risk factors; through a multi-path risk coupling operator, the dynamic weights and static environmental risk factors on each transmission path are weighted and fused, taking into account the nonlinear correlation between dynamic weights and static risks during the fusion process; the coupling risk value of each transmission path is calculated, and spatial interpolation is performed on the coupling risk values ​​of each path to fill the gaps in risk values ​​between spatiotemporal units; the interpolated multi-path risk values ​​are then fused into a unified risk field to generate a coupling risk field.

[0120] Among them, multi-path risk coupling includes core components such as propagation path dimension alignment, multi-path risk coupling operator invocation, path-by-path risk weighted fusion, coupled risk value calculation, spatial interpolation, and domain fusion, which are used to achieve the coupling and fusion of human-caused risks and environmental risks on multiple propagation paths.

[0121] (3) Propagation dynamics simulation of the coupled risk field is performed to obtain the dynamic risk assessment results of the integrated human-caused risk. The expression for the dynamic risk assessment results is as follows:

[0122] in, This indicates the results of a dynamic risk assessment. Represents the propagation dynamics simulation function. This represents a multi-path risk coupling operator. Indicates static environmental risk factors. This represents the spatiotemporal dynamic weight distribution.

[0123] For example, the biosafety assessment terminal acquires the coupled risk field, performs initial state calibration on the coupled risk field, sets the initial risk threshold, boundary constraints, and time step for the simulation; calls the propagation dynamics simulation function, which is built based on the infectious disease propagation dynamics model framework, to simulate the evolution of risk in the coupled risk field in the time dimension and the diffusion process in the spatial dimension; iteratively updates the risk state value of each spatiotemporal unit, captures the dynamic change trend of risk propagation, including the migration of risk hotspot areas and the fluctuation of risk intensity; and outputs the risk distribution state at the end of the simulation to obtain the dynamic risk assessment result that integrates human-caused risks.

[0124] In one embodiment, the spatiotemporal dynamic weight distribution is coupled with the static environmental risk factors in the brucellosis transmission model through multipath risk to obtain a coupled risk field, including:

[0125] (1) Perform path sensitivity analysis on the spatiotemporal dynamic weight distribution to obtain a multi-path dynamic weight vector set.

[0126] For example, the biosafety assessment terminal acquires the spatiotemporal dynamic weight distribution, retrieves a pre-defined brucellosis transmission path database, and identifies transmission path types such as airborne transmission, contact transmission, and vector transmission included in the database. It then performs path-by-path sensitivity analysis on the spatiotemporal dynamic weight distribution, combining attributes such as the transmission efficiency of the transmission path and the distribution of susceptible populations to quantify the risk transmission efficiency of the spatiotemporal dynamic weight distribution on each transmission path, thus obtaining the risk contribution of each path. Finally, it vectorizes the risk contribution of each transmission path according to the path dimension, with each vector dimension corresponding to a type of transmission path, encapsulating it into a structured vector form to obtain a multi-path dynamic weight vector set.

[0127] Among them, path sensitivity analysis includes core steps such as retrieving the brucellosis transmission path database, identifying transmission path types, quantifying the risk transmission efficiency of each path, and vectorizing the data. This is used to decompose the spatiotemporal dynamic weight distribution into weight vectors corresponding to different transmission paths.

[0128] (2) Perform path-adaptive mapping on static environmental risk factors to generate a path-aligned static risk base.

[0129] For example, the biosafety assessment terminal acquires static environmental risk factors from the brucellosis transmission model, extracts the original features of the static environmental risk factors, including transmission-related attributes such as environmental humidity, vector density, and host population distribution; matches the original features of the static environmental risk factors with the identified transmission path attributes one by one, for example, the environmental humidity feature corresponds to the airborne transmission path, and the vector density feature corresponds to the vector transmission path; adjusts the feature dimensions and data structure of the static environmental risk factors so that the path dimension of the static environmental risk factors is fully matched with the path dimension of the multi-path dynamic weight vector set; normalizes the adjusted static environmental risk factors, uses the Min-Max normalization method to unify the value range of the features, and generates a path-aligned static risk base.

[0130] The path adaptive mapping includes core steps such as original feature extraction, feature and path attribute matching, dimensional structure adjustment, and normalization processing, which are used to achieve precise alignment of static environmental risk factors and multi-path dynamic weight vector sets in the path dimension.

[0131] (3) Nonlinear path coupling is performed on the multi-path dynamic weight vector set and the path-aligned static risk basis to obtain the coupled risk field. The expression of the coupled risk field is:

[0132]

[0133] in, Indicates the coupled risk field. Represents a nonlinear path coupling function. This represents the path alignment operator. This indicates that the path is aligned with the static risk base. This represents a set of dynamic weight vectors for multiple paths.

[0134] For example, the biosafety assessment terminal acquires the multi-path dynamic weight vector set and the path-aligned static risk basis, performs path alignment operation, sorts the two types of data according to the unique identifier of the propagation path to ensure that the two correspond one-to-one in the propagation path dimension; calls the nonlinear path coupling function, which is constructed based on a nonlinear activation mechanism, to perform nonlinear weighted fusion of the multi-path dynamic weight vector set and the path-aligned static risk basis on each propagation path, and calculates the coupling risk value of each propagation path; integrates the coupling risk values ​​of all propagation paths, and uses the Kriging interpolation method to spatially interpolate the risk values ​​between spatiotemporal units to fill the gaps in risk values; integrates the interpolated multi-path risk values ​​into a unified risk field to generate the coupling risk field.

[0135] Nonlinear path coupling includes core components such as path alignment operation, nonlinear path coupling function call, path-by-path risk fusion, risk value integration, and spatial interpolation, which are used to achieve deep coupling of human-caused risks and environmental risks across multiple propagation paths.

[0136] like Figure 2 As shown, multi-source data fusion was performed on the dynamic quantitative indicators of biosafety compliance to obtain the Biosafety Compliance Index (BCI), which includes:

[0137] S201: Perform time-domain alignment processing on heterogeneous data streams in the dynamic quantitative indicators of biosafety compliance to generate a synchronized compliance feature set.

[0138] For example, the biosafety assessment terminal acquires heterogeneous data streams from the dynamic quantitative indicators of biosafety compliance, identifies different types of data included in the heterogeneous data streams, such as visual compliance feature data and process spatiotemporal offset feature data; extracts the original timestamp information of various heterogeneous data streams, and determines a unified time base and time granularity; adjusts the sampling frequency and data collection period of different types of heterogeneous data streams according to the unified time base, so that various types of data form a one-to-one correspondence in the time dimension; performs data integrity verification on the time-aligned heterogeneous data streams, and supplements the missing time-series data; integrates all time-synchronized heterogeneous data to generate a synchronized compliance feature set.

[0139] The time-domain alignment process includes core steps such as heterogeneous data stream type identification, timestamp extraction, time base unification, sampling frequency adjustment, data integrity verification, and synchronous data integration, which are used to eliminate the time dimension differences of heterogeneous data streams.

[0140] S202: Perform attention-based feature importance weighting on the synchronized compliance feature set to obtain a dynamically weighted feature matrix.

[0141] For example, the biosafety assessment terminal acquires a synchronized compliance feature set and constructs a feature weight calculation model based on an attention mechanism; it inputs the synchronized compliance feature set into the feature weight calculation model to analyze the degree of influence of each feature on the biosafety compliance assessment results; it dynamically adjusts the importance weight of each feature by combining biosafety procedure requirements and historical assessment data, strengthening the weight ratio of key compliance features and weakening the influence of secondary features; it performs a weighted operation on each feature in the synchronized compliance feature set with its corresponding importance weight; and it organizes the weighted features in a structured manner according to the feature dimension and the time dimension to generate a dynamic weighted feature matrix.

[0142] Among them, the attention-based feature importance weighting includes core steps such as weight calculation model construction, feature influence degree analysis, dynamic weight adjustment, feature weighting operation, and matrix structured organization, which are used to highlight key feature information for compliance assessment.

[0143] S203: Nonlinear feature compression is performed on the dynamic weighted feature matrix to obtain the biosafety compliance index (BCI).

[0144] For example, the biosafety assessment terminal acquires a dynamic weighted feature matrix and calls a preset nonlinear feature compression model, which is built based on a deep learning network. The nonlinear feature compression model performs high-dimensional feature reduction on the dynamic weighted feature matrix, retaining effective features that can characterize the core state of biosafety compliance and eliminating redundant features and noise information. The core features after dimensionality reduction are subjected to feature fusion and normalization to unify the value range and expression form of the features. The fused features are mapped to a single quantitative index, and the quantitative index is output to obtain the Biosafety Compliance Index (BCI).

[0145] Nonlinear feature compression includes core steps such as compression model invocation, high-dimensional feature dimensionality reduction, core feature fusion, feature normalization, and quantization index mapping, which are used to transform high-dimensional weighted features into a concise compliance quantification index.

[0146] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0147] In one embodiment, such as Figure 3 As shown, this application also provides an intelligent risk assessment system 300 for brucellosis in large-scale dairy farms, the system 300 comprising:

[0148] The multi-source agent data acquisition module 301 is used to collect multi-source heterogeneous agent data related to the execution of biosafety procedures through IoT terminals and visual sensing devices deployed in key areas of dairy farms.

[0149] The behavior process mining module 302 is used to perform computer vision behavior recognition and process mining on multi-source heterogeneous agent data to obtain dynamic quantitative indicators of biosafety compliance.

[0150] BCI fusion generation module 303 is used to fuse multi-source data on the dynamic quantitative indicators of biosafety compliance to obtain the biosafety compliance index BCI;

[0151] The dynamic risk assessment module 304 is used to inject the Biosafety Compliance Index (BCI) as a time-varying risk weight into a preset brucellosis transmission model to generate a dynamic risk assessment result that incorporates human-caused risks.

[0152] Specifically, the biosafety assessment terminal includes a multi-source proxy data acquisition module 301, which deploys IoT terminals and visual sensing devices such as temperature and humidity sensors, infrared cameras, and equipment operation monitoring terminals in key areas of the dairy farm, including the breeding area, disinfection channel, and feed processing area. The IoT terminals collect data in real time, including environmental parameters, equipment start-up and shutdown status, and operation records during the execution of biosafety procedures. The visual sensing devices continuously capture video streams and image data of operator actions and equipment usage processes. The raw data collected by both types of equipment undergoes format standardization conversion, unifying data encoding and storage formats. Data validity screening is conducted, eliminating abnormal fluctuations and invalid collection records. The standardized data is then integrated to form multi-source heterogeneous proxy data related to the execution of biosafety procedures.

[0153] Among them, the multi-source heterogeneous proxy data includes environmental parameter data, equipment operation data, operation record data, video stream data, image data, etc., which are used to comprehensively cover multi-dimensional scenarios such as environment, equipment, and personnel operation in the implementation of biosafety procedures.

[0154] The behavior process mining module 302 receives multi-source heterogeneous proxy data; it uses computer vision behavior recognition technologies such as pose estimation and action sequence matching to extract features such as the standardization of personnel operations and the completeness of process execution from video streams and image data in the data; it performs process node analysis on environmental parameters, equipment operation and operation record data to identify key execution nodes of biosafety processes, node connection timing and execution deviation; it constructs a cross-modal feature fusion model to associate and integrate the action features extracted by visual recognition with the node information obtained by process mining; and it uses a quantification algorithm to transform the fused multi-dimensional features into quantifiable dynamic indicators to generate dynamic quantitative indicators of biosafety compliance.

[0155] Among them, computer vision behavior recognition and process mining includes core links such as pose feature extraction, action sequence matching, process node parsing, cross-modal association fusion, and feature quantization transformation, which are used to transform unstructured and semi-structured data into dynamic quantification results representing compliance states.

[0156] The BCI fusion generation module 303 receives dynamic quantitative indicators of biosafety compliance; extracts timestamp information from heterogeneous data streams from different sources in the indicators to establish a unified time granularity and synchronization benchmark; adjusts the collection frequency and time interval of various data streams according to the synchronization benchmark to achieve temporal alignment; constructs a weight allocation model based on an attention mechanism to analyze the correlation between each feature in the aligned feature set and biosafety compliance, and dynamically assigns differentiated importance weights; performs weighted calculations on the features and corresponding weights, constructs a structured matrix according to feature category and time dimension, and obtains a dynamically weighted feature matrix; uses a deep learning dimensionality reduction model to compress the matrix into high-dimensional features, retaining core effective features and removing redundant information; normalizes the compressed core features and maps them to a single-dimensional quantitative index to obtain the Biosafety Compliance Index (BCI).

[0157] The multi-source data fusion includes core steps such as temporal synchronization alignment, attention weight allocation, high-dimensional feature compression, core feature normalization, and quantitative index mapping, which are used to integrate multi-dimensional dynamic quantitative indicators into a concise and unified compliance assessment index.

[0158] The dynamic risk assessment module 304 receives the Biosafety Compliance Index (BCI); retrieves a pre-defined brucellosis risk transmission path knowledge base, maps the BCI to risk weights for different transmission paths, and generates a spatiotemporal dynamic weight distribution; calls a pre-defined brucellosis transmission model, extracting static environmental risk factors such as geographical environment and host distribution included in the model; uses a multi-path risk coupling algorithm to couple the spatiotemporal dynamic weight distribution with the static environmental risk factors one by one according to the transmission path type, forming a coupled risk field covering the entire transmission path; starts the transmission dynamics simulation engine, sets the simulation boundary conditions and iteration step size, and simulates the evolution of the coupled risk field in the time dimension and the diffusion path in the spatial dimension; iteratively updates the risk state values ​​of each spatiotemporal unit, capturing the dynamic change characteristics of risk; and outputs the risk distribution data after the simulation is completed, obtaining a dynamic risk assessment result that integrates human-caused risks.

[0159] The dynamic risk assessment includes core components such as risk transmission path mapping, multi-path coupling calculation, transmission dynamics simulation, risk state iterative update, and assessment result output. It is used to integrate human-caused risks into traditional transmission models to achieve dynamic assessment of brucellosis risk.

[0160] Behavioral process mining module 302 is also used for:

[0161] Spatiotemporal behavioral maps were constructed from video stream data in multi-source heterogeneous proxy data to obtain behavioral trajectory sequences of key biosafety operations.

[0162] An attention-based action compliance assessment is performed on the behavioral trajectory sequence to generate a multi-dimensional visual compliance feature vector.

[0163] Perform process variation structure analysis on log data in multi-source heterogeneous proxy data to extract spatiotemporal offset features of process nodes;

[0164] By fusing multi-dimensional visual compliance feature vectors with spatiotemporal offset features of process nodes across modalities, a dynamic quantitative index of biosafety compliance is obtained.

[0165] Behavioral process mining module 302 is also used for:

[0166] Spatiotemporal attention enhancement is performed on multi-dimensional visual compliance feature vectors to generate attention-weighted visual feature tensors;

[0167] Topological structure embedding is performed on the spatiotemporal offset features of process nodes to obtain a topology-enhanced process feature matrix;

[0168] Heterogeneous feature interaction learning is performed on the attention-weighted visual feature tensor and the topology-enhanced process feature matrix to generate cross-modal fusion features;

[0169] Dynamic compliance quantification mapping is performed on cross-modal fusion features to obtain a dynamic quantitative index of biosafety compliance. The expression for the dynamic quantitative index of biosafety compliance is as follows:

[0170]

[0171] in, This represents a dynamic quantitative indicator of biosafety compliance. Represents the importance weights between modes. Represents the element-level Hadamard product. Represents the tensor outer product. Indicates the feature fusion operator, This represents the feature selection function. Represents the feature transformation function. This represents the dynamic quantization mapping function. Indicates the number of modal interaction channels. Indicates the first Attention-weighted visual feature tensor of the channel Indicates the first The topology enhancement process feature matrix of the channel. This represents the visual feature tensor after spatiotemporal attention enhancement. This represents the process feature matrix after topology embedding. Indicates the channel index.

[0172] Behavioral process mining module 302 is also used for:

[0173] By detecting key operation nodes in video stream data, a set of key biosafety action segments is obtained;

[0174] Cross-frame spatiotemporal correlation analysis was performed on a set of key biosafety action segments to obtain the trajectory of action continuity;

[0175] Behavioral semantic encoding is performed on the continuous trajectory of actions to generate a sequence of behavioral trajectories for critical biosafety operations. The expression for the sequence of behavioral trajectories is as follows:

[0176]

[0177] in, Represents a sequence of behavioral trajectories. Indicates the first A segment of key biosafety actions, Represents a set of adjacent segments within a spatiotemporal neighborhood. This indicates a feature concatenation operation. Indicates the aggregation of neighborhood features. This represents a Long Short-Term Memory (LSTM) network encoder. Represents the sequence aggregation operator, Indicates the first A neighborhood behavior fragment, Indicates the neighborhood fragment index, This indicates the total number of critical biosafety action segments. Indicates the index of the behavior segment currently being processed.

[0178] The dynamic risk assessment module 304 is also used for:

[0179] Risk transmission path mapping is performed on the Biosafety Compliance Index (BCI) to generate a spatiotemporal dynamic weight distribution;

[0180] By coupling the spatiotemporal dynamic weight distribution with the static environmental risk factors in the brucellosis transmission model through multipath risk, a coupled risk field is obtained.

[0181] Propagation dynamics simulation of the coupled risk field yields dynamic risk assessment results that incorporate human-caused risks. The expression for the dynamic risk assessment results is as follows:

[0182] in, This indicates the results of a dynamic risk assessment. Represents the propagation dynamics simulation function. This represents a multi-path risk coupling operator. Indicates static environmental risk factors. This represents the spatiotemporal dynamic weight distribution.

[0183] The dynamic risk assessment module 304 is also used for:

[0184] Path sensitivity analysis is performed on the spatiotemporal dynamic weight distribution to obtain a multi-path dynamic weight vector set;

[0185] Path-adaptive mapping is performed on static environmental risk factors to generate a path-aligned static risk base.

[0186] By performing nonlinear path coupling between the multi-path dynamic weight vector set and the path-aligned static risk basis, a coupled risk field is obtained. The expression for the coupled risk field is:

[0187]

[0188] in, Indicates the coupled risk field. Represents a nonlinear path coupling function. This represents the path alignment operator. This indicates that the path is aligned with the static risk base. This represents a set of dynamic weight vectors for multiple paths.

[0189] The BCI fusion generation module 303 is also used for:

[0190] Time-domain alignment processing is performed on heterogeneous data streams in the dynamic quantitative indicators of biosafety compliance to generate a synchronized compliance feature set;

[0191] The synchronized compliance feature set is weighted based on the importance of features using an attention mechanism to obtain a dynamically weighted feature matrix;

[0192] The Biosafety Compliance Index (BCI) is obtained by nonlinear feature compression of the dynamically weighted feature matrix.

[0193] In one embodiment, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0194] In one embodiment, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps in the above-described method embodiments.

[0195] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0196] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A method for intelligent risk assessment of brucellosis in large-scale dairy farms, characterized in that, The method includes: By deploying IoT terminals and visual sensing devices in key areas of dairy farms, multi-source heterogeneous proxy data related to the execution of biosafety procedures is collected. Computer vision behavior recognition and process mining were performed on the multi-source heterogeneous agent data to obtain dynamic quantitative indicators of biosafety compliance. The biosafety compliance dynamic quantitative indicators were fused from multiple sources to obtain the biosafety compliance index (BCI). The biosafety compliance index (BCI) is used as a time-varying risk weight and injected into a pre-defined brucellosis transmission model to generate a dynamic risk assessment result that incorporates human-caused risks.

2. The intelligent risk assessment method for brucellosis in large-scale dairy farms according to claim 1, characterized in that, The process of performing computer vision behavior recognition and process mining on the multi-source heterogeneous proxy data to obtain dynamic quantitative indicators of biosafety compliance includes: Spatiotemporal behavioral maps are constructed from the video stream data in the multi-source heterogeneous proxy data to obtain behavioral trajectory sequences of key biosafety operations. The behavior trajectory sequence is evaluated for action compliance based on an attention mechanism, and a multi-dimensional visual compliance feature vector is generated. Perform process variation structure analysis on the log data in the multi-source heterogeneous proxy data to extract the spatiotemporal offset features of process nodes; The multi-dimensional visual compliance feature vector is fused with the spatiotemporal offset feature of the process node through cross-modal feature fusion to obtain the dynamic quantitative index of biosafety compliance.

3. The intelligent risk assessment method for brucellosis in large-scale dairy farms according to claim 2, characterized in that, The process of fusing the multi-dimensional visual compliance feature vector with the spatiotemporal offset features of the process nodes across modalities to obtain the dynamic quantitative index of biosafety compliance includes: Spatiotemporal attention enhancement is performed on the multi-dimensional visual compliance feature vector to generate an attention-weighted visual feature tensor; Topological structure embedding is performed on the spatiotemporal offset features of the process nodes to obtain a topology-enhanced process feature matrix; Heterogeneous feature interaction learning is performed on the attention-weighted visual feature tensor and the topology enhancement process feature matrix to generate cross-modal fusion features; The cross-modal fusion features are dynamically mapped to compliance quantification to obtain the dynamic quantification index of biosafety compliance. The expression of the dynamic quantification index of biosafety compliance is as follows: in, This represents a dynamic quantitative indicator of biosafety compliance. Represents the importance weights between modes. Represents the element-level Hadamard product. Represents the tensor outer product. Indicates the feature fusion operator, This represents the feature selection function. Represents the feature transformation function. This represents the dynamic quantization mapping function. Indicates the number of modal interaction channels. Indicates the first Attention-weighted visual feature tensor of the channel Indicates the first The topology enhancement process feature matrix of the channel. This represents the visual feature tensor after spatiotemporal attention enhancement. This represents the process feature matrix after topology embedding. Indicates the channel index.

4. The intelligent risk assessment method for brucellosis in large-scale dairy farms according to claim 2, characterized in that, The construction of spatiotemporal behavioral maps from the video stream data in the multi-source heterogeneous proxy data to obtain behavioral trajectory sequences of key biosafety operations includes: By detecting key operation nodes in the video stream data, a set of key biosafety action segments is obtained; Cross-frame spatiotemporal correlation analysis was performed on the set of biosafety key action segments to obtain the action continuity trajectory; The continuity trajectory of the action is semantically encoded to generate a sequence of behavioral trajectories for the critical biosafety operation. The expression for the sequence of behavioral trajectories is as follows: in, Represents a sequence of behavioral trajectories. Indicates the first A segment of key biosafety actions, Represents a set of adjacent segments within a spatiotemporal neighborhood. This indicates a feature concatenation operation. Indicates the aggregation of neighborhood features. This represents a Long Short-Term Memory (LSTM) network encoder. Represents the sequence aggregation operator, Indicates the first A neighborhood behavior fragment, Indicates the neighborhood fragment index, This indicates the total number of critical biosafety action segments. Indicates the index of the behavior segment currently being processed.

5. The intelligent risk assessment method for brucellosis in large-scale dairy farms according to claim 1, characterized in that, The step of using the Biosafety Compliance Index (BCI) as a time-varying risk weight and injecting it into a pre-defined brucellosis transmission model to generate a dynamic risk assessment result that incorporates human-caused risks includes: Risk transmission path mapping is performed on the Biosafety Compliance Index (BCI) to generate a spatiotemporal dynamic weight distribution; The spatiotemporal dynamic weight distribution is coupled with the static environmental risk factors in the brucellosis transmission model through multipath risk coupling to obtain a coupled risk field. The propagation dynamics of the coupled risk field are simulated to obtain the dynamic risk assessment results of the fused human-cause risk.

6. The intelligent risk assessment method for brucellosis in large-scale dairy farms according to claim 5, characterized in that, The step of coupling the spatiotemporal dynamic weight distribution with the static environmental risk factors in the brucellosis transmission model through multipath risk to obtain a coupled risk field includes: The spatiotemporal dynamic weight distribution is analyzed for path sensitivity to obtain a multi-path dynamic weight vector set; Path-adaptive mapping is performed on the static environmental risk factors to generate a path-aligned static risk base. The multi-path dynamic weight vector set is nonlinearly coupled to the path-aligned static risk basis to obtain the coupled risk field.

7. The intelligent risk assessment method for brucellosis in large-scale dairy farms according to claim 1, characterized in that, The process of fusing multi-source data to obtain the Biosafety Compliance Index (BCI) from the dynamic quantitative indicators of biosafety compliance includes: The heterogeneous data streams in the dynamic quantitative indicators of biosafety compliance are time-domain aligned to generate a synchronized compliance feature set; The synchronized compliance feature set is weighted based on an attention mechanism to obtain a dynamically weighted feature matrix; The biosafety compliance index (BCI) is obtained by performing nonlinear feature compression on the dynamic weighted feature matrix.

8. A smart risk assessment system for brucellosis in large-scale dairy farms, characterized in that, The system includes: The multi-source agent data acquisition module is used to collect multi-source heterogeneous agent data related to the execution of biosafety procedures through IoT terminals and visual sensing devices deployed in key areas of dairy farms. The behavior process mining module is used to perform computer vision behavior recognition and process mining on the multi-source heterogeneous agent data to obtain dynamic quantitative indicators of biosafety compliance. The BCI fusion generation module is used to perform multi-source data fusion on the dynamic quantitative indicators of biosafety compliance to obtain the biosafety compliance index (BCI). The dynamic risk assessment module is used to inject the Biosafety Compliance Index (BCI) as a time-varying risk weight into a preset brucellosis transmission model to generate a dynamic risk assessment result that incorporates human-caused risks.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the intelligent risk assessment method for brucellosis in large-scale dairy farms as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of any one of claims 1 to 7 for a smart risk assessment method for brucellosis in large-scale dairy farms.

Images