Intelligent violent behavior identification and early warning method and system based on BVC scale mode

Abstract

The invention discloses a brute force behavior intelligent identification and early warning method and system based on a BVC scale mode, and the method comprises the steps: extracting behavior characteristics related to a brute force tendency through a natural language processing technology and a computer vision algorithm according to the evaluation content and mode of a BVC evaluation scale in combination with the unstructured data of a patient; based on the extracted behavior features, constructing an intelligent recognition model for violent behaviors by using a deep learning technology; a trained intelligent recognition model is deployed in a monitoring system, real-time analysis is continuously performed on patient behaviors, and when the model recognizes potential violent behaviors, a behavior analysis report is generated and an early warning mechanism is triggered. According to the embodiment of the invention, potential violent behaviors can be quickly responded and intervened through the intelligent identification and early warning method based on the BVC scale, and the accuracy, timeliness and comprehensiveness of identification are improved.

Description

Technical Field

[0001] The present invention belongs to the field of artificial intelligence technology, and in particular to a method and system for intelligently identifying and warning violent behavior based on a BVC scale model. Background Art

[0002] As society pays more and more attention to mental health and violent behavior, how to effectively identify and warn of potential violent behavior has become an important issue that needs to be addressed in the field of medical and psychological intervention. Traditional violent behavior identification mainly relies on the experience judgment and behavioral observation of medical staff. This method is not only highly subjective and inefficient, but also difficult to adapt to the complex and changeable psychological state of patients. With the advancement of science and technology, the use of advanced technical means, especially deep learning, computer vision and natural language processing, to achieve automatic identification and real-time monitoring of violent behavior has become a new research direction.

[0003] Identification of violent behavior often requires comprehensive consideration of multiple factors, including emotional state, verbal expression, and body language. Existing assessment tools such as the BVC (Bio-Violence Assessment Scale) provide a structured approach to assess patients' violent tendencies, but the assessment process still relies on manual observation and subjective judgment, making it difficult to achieve a comprehensive and timely assessment. Summary of the invention

[0004] The purpose of the present invention is to provide a method and system for intelligent identification and early warning of violent behavior based on the BVC scale model, so as to address the deficiencies in the prior art. The method can quickly respond to and intervene in potential violent behavior through an intelligent identification and early warning method based on the BVC scale, thereby improving the accuracy, timeliness and comprehensiveness of identification.

[0005] An embodiment of the present application provides a method for intelligently identifying and warning violent behavior based on the BVC scale model, the method comprising:

[0006] Based on the assessment content and mode of the BVC assessment scale and combined with the patient's unstructured data, behavioral features related to violent tendencies are extracted through natural language processing technology and computer vision algorithms, where the behavioral features include: emotional state, body language and verbal expression;

[0007] Based on the extracted behavioral features, deep learning technology is used to build an intelligent recognition model for violent behavior;

[0008] By deploying the trained intelligent recognition model in the monitoring system, the patient's behavior is continuously analyzed in real time. When the model identifies potential violent behavior, a behavior analysis report is generated and an early warning mechanism is triggered. The behavior analysis report includes: identified behavioral characteristics, emotional changes, and the corresponding relationship with relevant indicators in the BVC scale.

[0009] Optionally, the behavioral features related to violent tendencies are extracted through natural language processing technology and computer vision algorithms based on the assessment content and mode of the BVC assessment scale and in combination with the patient's unstructured data, including:

[0010] Analyze the assessment contents of the BVC scale and determine the multidimensional assessment indicators related to violent tendencies;

[0011] Collect unstructured data from patients, use computer vision technology to extract the patient's body language and movement patterns from the video surveillance data in the unstructured data, encode the body language and movement pattern information into time series features to reflect the patient's dynamic behavior in a specific situation, and establish a feature mapping between the time series features and the BVC scale evaluation indicators;

[0012] Apply deep learning models to analyze facial expressions and quantify emotional states to establish feature mapping between emotional features and BVC scale evaluation indicators;

[0013] Extract acoustic features from audio recordings in unstructured data, use Mel-frequency cepstral coefficients and pitch analysis techniques to identify speech features related to emotional states, and use natural language processing techniques to analyze the emotional features in patients' speech to establish feature mappings between speech features and emotional features and BVC scale evaluation indicators;

[0014] The time series features, emotion features, voice features and sentiment features are integrated, and the principal component analysis or feature selection algorithm is used to reduce the dimension and filter the features to reduce redundancy and retain the most informative features;

[0015] The fused features are stored in a multidimensional behavioral feature database so that the features reflect the relevant indicators of the corresponding mapping in the BVC scale.

[0016] Optionally, the step of constructing an intelligent recognition model for violent behavior based on the extracted behavioral features using deep learning technology includes:

[0017] Extract samples for model training from the collected multi-dimensional behavioral feature database, annotate the data, classify the samples using the BVC scale scoring criteria, indicate the level of violence tendency of each sample, and provide supervision signals for training;

[0018] Construct a hybrid model architecture combining a convolutional neural network (CNN) and a long short-term memory (LSTM) network, where the CNN is used to extract spatial features and the LSTM is used to process and analyze time series data to capture dynamic changes in behavior;

[0019] Obtain a pre-trained model for a large-scale general behavior recognition task as the basis for the CNN part. Fine-tune the last few layers of the CNN model pre-trained on the large-scale general behavior recognition task to adapt to the new violent behavior recognition task, thereby reducing the amount of data and time required for training from scratch.

[0020] The fine-tuned CNN model is trained again, and the LSTM model is trained at the same time to obtain a trained hybrid model architecture as an intelligent recognition model.

[0021] Optionally, the trained intelligent recognition model is deployed in the monitoring system to continuously analyze the patient's behavior in real time. When the model identifies potential violent behavior, a behavior analysis report is generated and an early warning mechanism is triggered, including:

[0022] Extracting a multimodal feature vector of the patient's behavior, wherein the multimodal feature vector corresponds to a feature in a multidimensional behavior feature database;

[0023] The extracted multimodal features are input into the trained intelligent recognition model for real-time recognition. When the model identifies potential violent behavior, an alarm mechanism is triggered. The confidence level of the intelligent recognition model output is compared with a preset threshold. If it exceeds the preset threshold, it is identified as potential violent behavior, a behavioral analysis report is generated, and an early warning mechanism is triggered.

[0024] Another embodiment of the present application provides a violent behavior intelligent identification and early warning system based on the BVC scale model, the system comprising:

[0025] An extraction module is used to extract behavioral features related to violent tendencies based on the assessment content and mode of the BVC assessment scale and in combination with the patient's unstructured data through natural language processing technology and computer vision algorithms, wherein the behavioral features include: emotional state, body language and verbal expression;

[0026] A building module for building an intelligent recognition model for violent behavior based on the extracted behavioral features using deep learning technology;

[0027] The recognition module is used to continuously analyze the patient's behavior in real time by deploying the trained intelligent recognition model in the monitoring system. When the model identifies potential violent behavior, it generates a behavior analysis report and triggers an early warning mechanism. The behavior analysis report includes: the identified behavioral characteristics, emotional changes, and the corresponding relationship with the relevant indicators in the BVC scale.

[0028] Yet another embodiment of the present application provides a storage medium, wherein the storage medium stores a computer program, wherein the computer program is configured to execute any of the above methods when running.

[0029] Yet another embodiment of the present application provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute any of the methods described above.

[0030] Compared with the prior art, the present invention provides an intelligent identification and early warning method for violent behavior based on the BVC scale model. According to the assessment content and model of the BVC assessment scale and in combination with the patient's unstructured data, the behavioral features related to violent tendencies are extracted through natural language processing technology and computer vision algorithms; based on the extracted behavioral features, an intelligent identification model for violent behavior is constructed using deep learning technology; by deploying the trained intelligent identification model in the monitoring system, the patient's behavior is continuously analyzed in real time. When the model identifies potential violent behavior, a behavioral analysis report is generated and an early warning mechanism is triggered, so that the intelligent identification and early warning method based on the BVC scale can quickly respond to and intervene in potential violent behavior, thereby improving the accuracy, timeliness and comprehensiveness of identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 A hardware structure block diagram of a computer terminal for a method for intelligently identifying and warning violent behaviors based on a BVC scale model provided by an embodiment of the present invention;

[0032] Figure 2 A flowchart of a method for intelligently identifying and warning violent behaviors based on the BVC scale model provided by an embodiment of the present invention;

[0033] Figure 3 A schematic diagram of the structure of a violent behavior intelligent identification and early warning system based on the BVC scale model provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0034] The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, but should not be construed as limiting the present invention.

[0035] The embodiment of the present invention firstly provides a method for intelligently identifying and warning violent behaviors based on the BVC scale model. The method can be applied to electronic devices, such as computer terminals, specifically ordinary computers, etc.

[0036] The following describes it in detail by taking running on a computer terminal as an example. Figure 1 The hardware structure block diagram of a computer terminal for a method for intelligently identifying and warning violent behaviors based on the BVC scale model provided by an embodiment of the present invention. Figure 1 As shown, the computer terminal may include one or more ( Figure 1Only one is shown in the figure) a processor 102 (the processor 102 may include but is not limited to a processing device such as a microprocessor MCU or a programmable logic device FPGA) and a memory 104 for storing data. Optionally, the computer terminal may also include a transmission device 106 for communication functions and an input and output device 108. It can be understood by those skilled in the art that Figure 1 The structure shown is only for illustration and does not limit the structure of the above-mentioned computer terminal. Figure 1 More or fewer components as shown, or with Figure 1 Different configurations are shown.

[0037] The memory 104 can be used to store software programs and modules of application software, such as program instructions / modules corresponding to the intelligent identification and early warning method of violent behavior based on the BVC scale model in the embodiment of the present application. The processor 102 executes various functional applications and data processing by running the software programs and modules stored in the memory 104, that is, to implement the above method. The memory 104 may include a high-speed random access memory, and may also include a non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some examples, the memory 104 may further include a memory remotely arranged relative to the processor 102, and these remote memories can be connected to the computer terminal via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0038] The transmission device 106 is used to receive or send data via a network. The specific example of the above network may include a wireless network provided by a communication provider of a computer terminal. In one example, the transmission device 106 includes a network adapter (Network Interface Controller, NIC), which can be connected to other network devices through a base station so as to communicate with the Internet. In one example, the transmission device 106 can be a radio frequency (RF) module, which is used to communicate with the Internet wirelessly.

[0039] See also Figure 2 The embodiment of the present invention provides a method for intelligent identification and early warning of violent behavior based on the BVC scale model, which may include the following steps:

[0040] S201, based on the assessment content and mode of the BVC assessment scale and in combination with the patient's unstructured data, extracting behavioral features related to violent tendencies through natural language processing technology and computer vision algorithms, wherein the behavioral features include: emotional state, body language, and verbal expression;

[0041] By comprehensively analyzing the assessment content in the BVC assessment scale and combining it with unstructured data from patients (including video surveillance and audio recordings), key behavioral characteristics related to violent tendencies are extracted. These characteristics include the patient's emotional state, body language, and verbal expression in specific situations. This multidimensional information collection method provides a more comprehensive and in-depth understanding and can play an important role in the early detection of violent tendencies.

[0042] Specifically, we can analyze each assessment item in the BVC scale and determine multidimensional assessment indicators related to violent tendencies;

[0043] Through in-depth analysis of the various contents of the BVC assessment scale, indicators related to violent tendencies are identified. This step ensures that the focus of subsequent feature extraction is on features closely related to violent behavior, thereby maximizing the accuracy of identification. The core of this analysis step is to build an evaluation framework based on scientific evidence to guide subsequent data collection and feature extraction. Ensure that the extracted features can effectively reflect the patient's potential violent tendencies, thereby improving the model's early warning ability.

[0044] Text mining algorithms can be used to conduct content analysis on the items of the BVC scale to identify key keywords and phrases; topic models (LDA, etc.) can be used to mine potential themes in the scale and determine specific dimensions related to violent behavior; based on the analysis results, multidimensional evaluation indicators can be extracted and weights can be set for each indicator to reflect its degree of influence on violent tendencies.

[0045] Collect unstructured data from patients, use computer vision technology to extract the patient's body language and movement patterns from the video surveillance data in the unstructured data, encode the body language and movement pattern information into time series features to reflect the patient's dynamic behavior in a specific situation, and establish a feature mapping between the time series features and the BVC scale evaluation indicators;

[0046] By using computer vision technology to analyze the patient's behavior, we can obtain key information such as body language and movement patterns. This information is encoded into time series features to reflect the dynamic changes of the patient in a specific situation and provide real-time behavior data for the model. This process can capture subtle behavioral changes of the patient, which may be important precursors to violent tendencies. Therefore, the extraction of time series features provides dynamic and continuous data support for subsequent intelligent recognition.

[0047] The video stream can be captured in real time and frame processed, and the patient's limbs and key points can be identified through the target detection algorithm; the extracted body language and movement patterns can be continuously analyzed using time series analysis methods (such as long short-term memory network LSTM); the time series data is input into the feature extraction module using the sliding window method to generate a series of codes, and the information of body language and movement patterns is converted into a time series feature matrix; combined with the evaluation indicators in the BVC scale, the mapping relationship between the time series features and the evaluation indicators is established through regression analysis.

[0048] Apply deep learning models to analyze facial expressions and quantify emotional states to establish feature mapping between emotional features and BVC scale evaluation indicators;

[0049] Through facial expression analysis, the patient's emotional state can be quantified in real time, thus providing a quantitative basis for emotional fluctuations. This process is particularly important in predicting violent tendencies, as it can provide immediate feedback on emotional changes. Emotional state is a key influencing factor in violent behavior. Quantifying emotional state ensures that the model can take into account the patient's emotional changes when predicting violent tendencies, improving the accuracy of recognition.

[0050] A facial expression recognition model can be established, and a convolutional neural network can be used to analyze video data and extract facial features. An emotion classifier can be used to map facial features to different emotion categories (such as anger, happiness, sadness, etc.). The emotional state can be quantified by setting a scoring system to give different weights to different emotions to form an emotion feature vector. A feature mapping model between emotional features and BVC evaluation indicators can be constructed through methods such as logistic regression or support vector machines.

[0051] Extract acoustic features from audio recordings in unstructured data, use Mel-frequency cepstral coefficients and pitch analysis techniques to identify speech features related to emotional states, and use natural language processing techniques to analyze the emotional features in patients' speech to establish feature mappings between speech features and emotional features and BVC scale evaluation indicators;

[0052] This step aims to extract speech features related to the patient's emotional state through the analysis of audio data. These features will serve as supplementary data for emotional state and support the identification of violent tendencies from multiple perspectives. The extraction of acoustic and emotional features helps to identify the underlying emotions conveyed by the patient during communication and form a comprehensive assessment of the behavioral state. By comprehensively considering the characteristics of audio and speech, the early warning capability of violent behavior is improved.

[0053] Mel-frequency cepstral coefficients (MFCCs) can be extracted from audio recordings as acoustic features, and the spectral characteristics of the audio signal can be obtained through short-time Fourier transform; combined with pitch analysis technology, the pitch changes can be modeled and emotional changes can be identified; natural language processing technology can be used to analyze the emotional characteristics in the patient's speech, and the emotions can be annotated using an emotional dictionary to form an emotional feature set; combined with acoustic features and emotional features, multivariate linear regression or machine learning classifiers can be used to establish feature mapping and associate it with the BVC evaluation index.

[0054] The time series features, emotion features, voice features and sentiment features are integrated, and the principal component analysis or feature selection algorithm is used to reduce the dimension and filter the features to reduce redundancy and retain the most informative features;

[0055] In this step, the dimensionality of the feature space is reduced by fusing multiple features. This process ensures that the model can process data efficiently during training and avoid overfitting problems. The fusion and dimensionality reduction of features not only improves computational efficiency, but also ensures the generalization ability of the model. This can reduce the negative impact of redundant features on model training, making the recognition process more accurate and efficient.

[0056] The extracted time series features, emotional features, speech features and sentiment features can be integrated into a feature matrix; the principal component analysis (PCA) technique is used to reduce the dimension of the features, and the main components are selected to retain the features with the highest variance in the data; feature selection algorithms (such as LASSO regression, random forest, etc.) are used to further screen the features to ensure that the final feature set only contains the most informative features for identifying violent tendencies; finally, the fused features after dimension reduction are stored in a multidimensional behavioral feature database for subsequent use.

[0057] The fused features are stored in a multidimensional behavioral feature database so that the features reflect the relevant indicators of the corresponding mapping in the BVC scale.

[0058] After the previous feature extraction, fusion and dimensionality reduction, the final features are stored in the multi-dimensional behavior feature database. These features can not only reflect the evaluation indicators of the BVC scale, but also provide important training data for the subsequent intelligent recognition model. This database is the core data source of the entire recognition system, storing multi-dimensional and time-series behavior feature information, ensuring that the intelligent recognition model can be trained and reasoned based on comprehensive information and make accurate judgments on violent behavior.

[0059] An efficient database architecture (such as a relational database or a NoSQL database) can be designed to store the fused features; each feature vector will be accompanied by a corresponding BVC evaluation indicator identifier to ensure the convenience of subsequent retrieval and association analysis; data version control is implemented, the feature database is updated regularly, and new behavioral features are integrated to maintain the adaptability and effectiveness of the model; indexing technology is used to improve database query efficiency to ensure that the system can quickly respond to feature query requests for real-time monitoring.

[0060] S202, based on the extracted behavior features, using deep learning technology to build an intelligent recognition model for violent behavior;

[0061] By integrating the previously extracted multi-dimensional behavioral features, an intelligent recognition model is constructed using deep learning technology to effectively identify patients' violent behaviors. This model will integrate multiple features, such as emotional state, body language, verbal expression, etc., to form comprehensive input data, thereby achieving efficient identification of violent behaviors.

[0062] Specifically, samples for model training can be extracted from the collected multi-dimensional behavioral feature database, and data annotation can be performed. The samples can be classified using the scoring criteria of the BVC scale, and the level of violence tendency of each sample can be indicated to provide supervision signals for training.

[0063] In this process, sample data suitable for model training is selected from the multidimensional behavioral feature database, and these samples are classified and labeled based on the scoring criteria of the BVC scale. The labeled violence tendency level provides the necessary supervision signal for the model, so that the model can learn the relationship between different violence tendency levels and corresponding input features. Data labeling is a prerequisite for model training and ensures the effectiveness and accuracy of the training process. The labeled samples provide clear supervision signals for the deep learning model, allowing the model to continuously optimize its parameters during the learning process, thereby improving the accuracy of violent behavior recognition.

[0064] Samples from the behavioral feature database can be selected through statistical analysis to ensure that the samples cover different levels of violence tendency. Text mining technology can be used to structure the scoring criteria of the BVC scale, establish scoring rules, and automatically assign corresponding violence tendency levels to samples. Data enhancement technology (such as rotation, scaling, cropping, etc.) can be applied to training samples to increase the diversity of training samples and improve the generalization ability of the model.

[0065] Construct a hybrid model architecture combining a convolutional neural network (CNN) and a long short-term memory (LSTM) network, where the CNN is used to extract spatial features and the LSTM is used to process and analyze time series data to capture dynamic changes in behavior;

[0066] By designing a hybrid model architecture that combines CNN and LSTM, we can fully utilize the spatial feature extraction capabilities of CNN and the time series analysis capabilities of LSTM to provide a powerful solution to the problem of dynamic identification of violent behavior. This hybrid structure can effectively capture the spatial distribution and temporal evolution of behavioral characteristics and deepen our understanding of the complexity of violent behavior. CNN can identify the patient's physical performance in static images, while LSTM can analyze the changes in the patient's behavior over time to form a comprehensive assessment of violent behavior.

[0067] First, we can use convolutional neural networks to extract spatial features from input data (such as video frames), design convolutional layers and pooling layers for feature extraction; input the extracted spatial features into the long short-term memory network in time series, and design multiple LSTM layers to capture the dynamic evolution of behavior; connect a fusion layer to the output layer of CNN and LSTM, perform weighted combination of the outputs of the two, and generate the final comprehensive feature representation.

[0068] Obtain a pre-trained model for a large-scale general behavior recognition task as the basis for the CNN part. Fine-tune the last few layers of the CNN model pre-trained on the large-scale general behavior recognition task to adapt to the new violent behavior recognition task, thereby reducing the amount of data and time required for training from scratch.

[0069] Introducing a pre-trained model from a large-scale general behavior recognition task as a basis can leverage existing knowledge to accelerate the training process of the violent behavior recognition model. Fine-tuning the last few layers can better adapt to specific application scenarios and improve the performance of the model. This approach not only significantly reduces the demand for training data, but also reduces the time and cost of training a model from scratch. The initial parameters provided by the pre-trained model make training more stable and efficient, thereby accelerating convergence and improving the final recognition accuracy.

[0070] You can choose a CNN model suitable for general behavior recognition tasks (such as ResNet or Inception), download and load its pre-trained weights; freeze the weights of the first few layers, only update the parameters of the last few layers, and adjust the training learning rate slightly to ensure that the output of the model can better adapt to new tasks; during the fine-tuning process, set a validation set, regularly monitor the performance of the model, and adjust the learning rate and optimizer parameters through cross-validation technology.

[0071] The fine-tuned CNN model is trained again, and the LSTM model is trained at the same time to obtain a trained hybrid model architecture as an intelligent recognition model.

[0072] In this step, the fine-tuned CNN model and the LSTM model are trained simultaneously to form a complete hybrid model architecture, and finally generate an intelligent recognition model with strong recognition capabilities. This joint training method can ensure that the feature learning of the two models is carried out simultaneously, making the recognition more consistent and accurate. In addition, through the training of the overall architecture, more complex behavior patterns can be captured, providing stronger support for the identification of violent behavior.

[0073] An end-to-end training method can be used to integrate the training processes of CNN and LSTM at the same time. The input data is directly transmitted to the LSTM module after feature extraction by CNN. During the training process, a learning rate scheduling strategy is adopted to dynamically adjust the learning rate according to the performance of the validation set to improve training efficiency and avoid overfitting.

[0074] S203, by deploying the trained intelligent recognition model in the monitoring system, continuously analyze the patient's behavior in real time. When the model identifies potential violent behavior, it generates a behavior analysis report and triggers an early warning mechanism, wherein the behavior analysis report includes: identified behavioral characteristics, emotional changes, and the corresponding relationship with relevant indicators in the BVC scale.

[0075] By deploying a trained intelligent recognition model in the monitoring system, the patient's behavior is monitored in real time. By analyzing the real-time video stream and other sensor data, the model will continuously determine whether there is potential violent behavior. Once potential violent behavior is identified, the system will automatically generate a behavioral analysis report, which records in detail the identified behavioral characteristics, the patient's emotional changes, and their comparative relationship with the relevant indicators of the BVC scale for medical staff to use.

[0076] This step is of great significance for ensuring personal safety. Through real-time monitoring and intelligent recognition, timely identification and early warning can be achieved to prevent the occurrence of violent incidents. The generated behavioral analysis report provides detailed information reference for medical staff to help them take appropriate intervention measures and deal with potential crises in a timely manner. Therefore, the integration of intelligent recognition technology and real-time monitoring can significantly improve the level of safety and security in the field of mental health.

[0077] Specifically, a multimodal feature vector of the patient's behavior may be extracted, wherein the multimodal feature vector corresponds to a feature in a multidimensional behavior feature database;

[0078] By analyzing and fusing data from different sources (such as video surveillance, audio data, and text data), the patient's multimodal feature vectors are extracted. These feature vectors contain the patient's behavior, emotional state, and verbal expression, and can fully reflect the patient's psychological and behavioral patterns. The extraction of multimodal features provides a rich information basis for subsequent behavioral analysis, allowing the system to consider more comprehensive factors when judging potential violent behavior. By aggregating different types of data, this step ensures the accuracy of information and the effectiveness of decision-making, greatly improving the accuracy of identifying violent tendencies.

[0079] A fusion model based on the self-attention mechanism can be constructed. The model dynamically calculates the importance of each feature to the final recognition result by inputting the features of different modalities into a multi-layer self-attention network. In the process of feature fusion, the correlation matrix is ​​used to quantify the relationship between different modalities. For example, the features are weighted by calculating the interaction between modalities to ensure that the final multimodal feature vector is more informative and representative.

[0080] The extracted multimodal features are input into the trained intelligent recognition model for real-time recognition. When the model identifies potential violent behavior, an alarm mechanism is triggered. The confidence level of the intelligent recognition model output is compared with a preset threshold. If it exceeds the preset threshold, it is identified as potential violent behavior, a behavioral analysis report is generated, and an early warning mechanism is triggered.

[0081] The previously extracted multimodal feature vectors are input into the trained intelligent recognition model to analyze the patient's behavior in real time. Once the model identifies potential violent behavior, it compares the output confidence with the preset threshold and triggers the corresponding alarm mechanism. This process ensures real-time monitoring and timely response to patient behavior, and can provide early warning before violent behavior occurs, thereby providing medical staff with timely intervention measures. The confidence comparison mechanism optimizes the decision-making process, reduces the possibility of false alarms and missed alarms, and significantly improves the reliability and effectiveness of the system.

[0082] A dynamic threshold adjustment mechanism can be introduced, which is optimized based on past monitoring data and real-time feedback. By establishing a feedback loop and adaptively adjusting the preset threshold, the model can flexibly adjust the recognition sensitivity to adapt to individual differences and environmental changes in the context of different patients' behavioral performance and emotional state changes. This flexibility enables the system to maintain efficient and accurate early warning capabilities in a variety of situations.

[0083] Once the intelligent recognition model detects potential violent behavior, the system will automatically generate a behavioral analysis report, detailing the identified behavioral characteristics, the patient's emotional changes, and the correspondence between these changes and the relevant indicators in the BVC scale. At the same time, the system will also trigger the corresponding early warning mechanism to ensure that relevant personnel are aware of the situation in a timely manner. The generated behavioral analysis report provides important feedback information to medical staff, which not only helps to quickly understand the patient's current status, but also provides data support in subsequent evaluations and interventions. At the same time, the triggered early warning mechanism can ensure that the medical team takes quick action to reduce potential risks. This process is crucial to improving the overall quality and safety of medical services.

[0084] The knowledge graph can be used to build a behavior analysis report generation system. By associating the identified behavioral characteristics and emotional states with the BVC scale indicators, the system will automatically map them into a knowledge graph. The graph contains information such as the relationship between various behaviors and emotions, historical data, clinical manifestations, etc. When generating a report, the system automatically generates a comprehensive report with background information and reasoning process based on the latest recognition results and knowledge graph, providing deeper analysis and suggestions. This approach not only improves the accuracy of the report, but also makes the analysis results more clinically valuable.

[0085] It can be seen that according to the assessment content and mode of the BVC assessment scale and combined with the patient's unstructured data, behavioral characteristics related to violent tendencies are extracted through natural language processing technology and computer vision algorithms; based on the extracted behavioral characteristics, deep learning technology is used to build an intelligent recognition model for violent behavior; by deploying the trained intelligent recognition model in the monitoring system, the patient's behavior is continuously analyzed in real time. When the model identifies potential violent behavior, a behavioral analysis report is generated and an early warning mechanism is triggered, so that potential violent behavior can be quickly responded to and intervened through intelligent recognition and early warning methods based on the BVC scale, thereby improving the accuracy, timeliness and comprehensiveness of recognition.

[0086] Another embodiment of the present invention provides a violent behavior intelligent identification and early warning system based on the BVC scale model, see Figure 3 , the system may include:

[0087] The extraction module 301 is used to extract the behavior characteristics related to the violent tendency through natural language processing technology and computer vision algorithm according to the evaluation content and mode of the BVC evaluation scale and in combination with the unstructured data of the patient, wherein the behavior characteristics include: emotional state, body language and verbal expression;

[0088] A construction module 302 is used to construct an intelligent recognition model for violent behavior based on the extracted behavior features using deep learning technology;

[0089] The recognition module 303 is used to continuously analyze the patient's behavior in real time by deploying the trained intelligent recognition model in the monitoring system. When the model identifies potential violent behavior, it generates a behavior analysis report and triggers an early warning mechanism, wherein the behavior analysis report includes: the identified behavioral characteristics, emotional changes, and the corresponding relationship with the relevant indicators in the BVC scale.

[0090] It can be seen that according to the assessment content and mode of the BVC assessment scale and combined with the patient's unstructured data, behavioral characteristics related to violent tendencies are extracted through natural language processing technology and computer vision algorithms; based on the extracted behavioral characteristics, deep learning technology is used to build an intelligent recognition model for violent behavior; by deploying the trained intelligent recognition model in the monitoring system, the patient's behavior is continuously analyzed in real time. When the model identifies potential violent behavior, a behavioral analysis report is generated and an early warning mechanism is triggered, so that potential violent behavior can be quickly responded to and intervened through intelligent recognition and early warning methods based on the BVC scale, thereby improving the accuracy, timeliness and comprehensiveness of recognition.

[0091] An embodiment of the present invention further provides a storage medium, in which a computer program is stored, wherein the computer program is configured to execute the steps of any of the above method embodiments when running.

[0092] Specifically, in this embodiment, the above storage medium may be configured to store a computer program for performing the following steps:

[0093] S201, based on the assessment content and mode of the BVC assessment scale and in combination with the patient's unstructured data, extracting behavioral features related to violent tendencies through natural language processing technology and computer vision algorithms, wherein the behavioral features include: emotional state, body language, and verbal expression;

[0094] S202, based on the extracted behavior features, using deep learning technology to build an intelligent recognition model for violent behavior;

[0095] S203, by deploying the trained intelligent recognition model in the monitoring system, continuously analyze the patient's behavior in real time. When the model identifies potential violent behavior, it generates a behavior analysis report and triggers an early warning mechanism, wherein the behavior analysis report includes: identified behavioral characteristics, emotional changes, and the corresponding relationship with relevant indicators in the BVC scale.

[0096] It can be seen that according to the assessment content and mode of the BVC assessment scale and combined with the patient's unstructured data, behavioral characteristics related to violent tendencies are extracted through natural language processing technology and computer vision algorithms; based on the extracted behavioral characteristics, deep learning technology is used to build an intelligent recognition model for violent behavior; by deploying the trained intelligent recognition model in the monitoring system, the patient's behavior is continuously analyzed in real time. When the model identifies potential violent behavior, a behavioral analysis report is generated and an early warning mechanism is triggered, so that potential violent behavior can be quickly responded to and intervened through intelligent recognition and early warning methods based on the BVC scale, thereby improving the accuracy, timeliness and comprehensiveness of recognition.

[0097] An embodiment of the present invention further provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor is configured to run the computer program to execute the steps in any one of the above method embodiments.

[0098] Specifically, the electronic device may further include a transmission device and an input / output device, wherein the transmission device is connected to the processor, and the input / output device is connected to the processor.

[0099] Specifically, in this embodiment, the processor may be configured to perform the following steps through a computer program:

[0100] S201, based on the assessment content and mode of the BVC assessment scale and in combination with the patient's unstructured data, extracting behavioral features related to violent tendencies through natural language processing technology and computer vision algorithms, wherein the behavioral features include: emotional state, body language, and verbal expression;

[0101] S202, based on the extracted behavior features, using deep learning technology to build an intelligent recognition model for violent behavior;

[0102] S203, by deploying the trained intelligent recognition model in the monitoring system, continuously analyze the patient's behavior in real time. When the model identifies potential violent behavior, it generates a behavior analysis report and triggers an early warning mechanism, wherein the behavior analysis report includes: identified behavioral characteristics, emotional changes, and the corresponding relationship with relevant indicators in the BVC scale.

[0103] It can be seen that according to the assessment content and mode of the BVC assessment scale and combined with the patient's unstructured data, behavioral characteristics related to violent tendencies are extracted through natural language processing technology and computer vision algorithms; based on the extracted behavioral characteristics, deep learning technology is used to build an intelligent recognition model for violent behavior; by deploying the trained intelligent recognition model in the monitoring system, the patient's behavior is continuously analyzed in real time. When the model identifies potential violent behavior, a behavioral analysis report is generated and an early warning mechanism is triggered, so that potential violent behavior can be quickly responded to and intervened through intelligent recognition and early warning methods based on the BVC scale, thereby improving the accuracy, timeliness and comprehensiveness of recognition.

[0104] The above describes in detail the structure, features and effects of the present invention based on the embodiments shown in the drawings. The above is only a preferred embodiment of the present invention, but the present invention is not limited to the scope of implementation shown in the drawings. Any changes made according to the concept of the present invention, or modifications to equivalent embodiments with equivalent changes, which still do not exceed the spirit covered by the description and drawings, should be within the protection scope of the present invention.

Claims

1. A method for intelligent identification and early warning of violent behavior based on the BVC scale model, characterized in that: The method comprises: Based on the assessment content and mode of the BVC assessment scale and combined with the patient's unstructured data, behavioral features related to violent tendencies are extracted through natural language processing technology and computer vision algorithms, where the behavioral features include: emotional state, body language and verbal expression; Based on the extracted behavioral features, deep learning technology is used to build an intelligent recognition model for violent behavior; By deploying the trained intelligent recognition model in the monitoring system, the patient's behavior is continuously analyzed in real time. When the model identifies potential violent behavior, a behavior analysis report is generated and an early warning mechanism is triggered. The behavior analysis report includes: identified behavioral characteristics, emotional changes, and the corresponding relationship with relevant indicators in the BVC scale.

2. The method according to claim 1, characterized in that: Based on the assessment content and mode of the BVC assessment scale, combined with the patient's unstructured data, behavioral features related to violent tendencies are extracted through natural language processing technology and computer vision algorithms, including: Analyze the assessment contents of the BVC scale and determine the multidimensional assessment indicators related to violent tendencies; Collect unstructured data from patients, use computer vision technology to extract the patient's body language and movement patterns from the video surveillance data in the unstructured data, encode the body language and movement pattern information into time series features to reflect the patient's dynamic behavior in a specific situation, and establish a feature mapping between the time series features and the BVC scale evaluation indicators; Apply deep learning models to analyze facial expressions and quantify emotional states to establish feature mapping between emotional features and BVC scale evaluation indicators; Extract acoustic features from audio recordings in unstructured data, use Mel-frequency cepstral coefficients and pitch analysis techniques to identify speech features related to emotional states, and use natural language processing techniques to analyze the emotional features in patients' speech to establish feature mappings between speech features and emotional features and BVC scale evaluation indicators; The time series features, emotion features, voice features and sentiment features are integrated, and the principal component analysis or feature selection algorithm is used to reduce the dimension and filter the features to reduce redundancy and retain the most informative features; The fused features are stored in a multidimensional behavioral feature database so that the features reflect the relevant indicators of the corresponding mapping in the BVC scale.

3. The method according to claim 2, characterized in that The method of constructing an intelligent recognition model for violent behavior based on the extracted behavioral features using deep learning technology includes: Extract samples for model training from the collected multi-dimensional behavioral feature database, annotate the data, classify the samples using the BVC scale scoring criteria, indicate the level of violence tendency of each sample, and provide supervision signals for training; Construct a hybrid model architecture combining a convolutional neural network (CNN) and a long short-term memory (LSTM) network, where the CNN is used to extract spatial features and the LSTM is used to process and analyze time series data to capture dynamic changes in behavior; Obtain a pre-trained model for a large-scale general behavior recognition task as the basis for the CNN part. Fine-tune the last few layers of the CNN model pre-trained on the large-scale general behavior recognition task to adapt to the new violent behavior recognition task, thereby reducing the amount of data and time required for training from scratch. The fine-tuned CNN model is trained again, and the LSTM model is trained at the same time to obtain a trained hybrid model architecture as an intelligent recognition model.

4. The method according to claim 3, characterized in that The trained intelligent recognition model is deployed in the monitoring system to continuously analyze the patient's behavior in real time. When the model identifies potential violent behavior, a behavior analysis report is generated and an early warning mechanism is triggered, including: Extracting a multimodal feature vector of the patient's behavior, wherein the multimodal feature vector corresponds to a feature in a multidimensional behavior feature database; The extracted multimodal features are input into the trained intelligent recognition model for real-time recognition. When the model identifies potential violent behavior, an alarm mechanism is triggered. The confidence level of the intelligent recognition model output is compared with a preset threshold. If it exceeds the preset threshold, it is identified as potential violent behavior, a behavioral analysis report is generated, and an early warning mechanism is triggered.

5. An intelligent identification and early warning system for violent behavior based on the BVC scale model, characterized in that: The system comprises: An extraction module is used to extract behavioral features related to violent tendencies based on the assessment content and mode of the BVC assessment scale and in combination with the patient's unstructured data through natural language processing technology and computer vision algorithms, wherein the behavioral features include: emotional state, body language and verbal expression; A building module for building an intelligent recognition model for violent behavior based on the extracted behavioral features using deep learning technology; The recognition module is used to continuously analyze the patient's behavior in real time by deploying the trained intelligent recognition model in the monitoring system. When the model identifies potential violent behavior, it generates a behavior analysis report and triggers an early warning mechanism. The behavior analysis report includes: the identified behavioral characteristics, emotional changes, and the corresponding relationship with the relevant indicators in the BVC scale.

6. The system according to claim 5, characterized in that The extraction module is specifically used for: Analyze the assessment contents of the BVC scale and determine the multidimensional assessment indicators related to violent tendencies; Collect unstructured data from patients, use computer vision technology to extract the patient's body language and movement patterns from the video surveillance data in the unstructured data, encode the body language and movement pattern information into time series features to reflect the patient's dynamic behavior in a specific situation, and establish a feature mapping between the time series features and the BVC scale evaluation indicators; Apply deep learning models to analyze facial expressions and quantify emotional states to establish feature mapping between emotional features and BVC scale evaluation indicators; Extract acoustic features from audio recordings in unstructured data, use Mel-frequency cepstral coefficients and pitch analysis techniques to identify speech features related to emotional states, and use natural language processing techniques to analyze the emotional features in patients' speech to establish feature mappings between speech features and emotional features and BVC scale evaluation indicators; The time series features, emotion features, voice features and sentiment features are integrated, and the principal component analysis or feature selection algorithm is used to reduce the dimension and filter the features to reduce redundancy and retain the most informative features; The fused features are stored in a multidimensional behavioral feature database so that the features reflect the relevant indicators of the corresponding mapping in the BVC scale.

7. The system according to claim 6, characterized in that The building blocks are specifically used for: Extract samples for model training from the collected multi-dimensional behavioral feature database, annotate the data, classify the samples using the BVC scale scoring criteria, indicate the level of violence tendency of each sample, and provide supervision signals for training; Construct a hybrid model architecture combining a convolutional neural network (CNN) and a long short-term memory (LSTM) network, where the CNN is used to extract spatial features and the LSTM is used to process and analyze time series data to capture dynamic changes in behavior; Obtain a pre-trained model for a large-scale general behavior recognition task as the basis for the CNN part. Fine-tune the last few layers of the CNN model pre-trained on the large-scale general behavior recognition task to adapt to the new violent behavior recognition task, thereby reducing the amount of data and time required for training from scratch. The fine-tuned CNN model is trained again, and the LSTM model is trained at the same time to obtain a trained hybrid model architecture as an intelligent recognition model.

8. The system according to claim 7, characterized in that The identification module is specifically used for: Extracting a multimodal feature vector of the patient's behavior, wherein the multimodal feature vector corresponds to a feature in a multidimensional behavior feature database; The extracted multimodal features are input into the trained intelligent recognition model for real-time recognition. When the model identifies potential violent behavior, an alarm mechanism is triggered. The confidence level of the intelligent recognition model output is compared with a preset threshold. If it exceeds the preset threshold, it is identified as potential violent behavior, a behavioral analysis report is generated, and an early warning mechanism is triggered.

9. A storage medium, characterized in that: The storage medium stores a computer program, wherein the computer program is configured to execute the method according to any one of claims 1 to 4 when executed.

10. An electronic device comprising a memory and a processor, characterized in that: A computer program is stored in the memory, and the processor is configured to run the computer program to perform the method according to any one of claims 1 to 4.

Images