Intentional violation monitoring system and method based on electroencephalogram signals
By constructing a Conv1d&Dilated Convolution deep learning model and a perceived benefit experimentation method, the problem of insufficient multi-factor analysis of intentional violations in existing technologies is solved, enabling real-time early warning and intervention for intentional violations and improving the accuracy and applicability of monitoring.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- CENT SOUTH UNIV
- Filing Date
- 2026-03-20
- Publication Date
- 2026-06-19
AI Technical Summary
Current EEG technology has failed to fully cover and deeply analyze multiple factors in the field of intentional violation monitoring, resulting in insufficient understanding of the complex internal mechanisms by which the brain drives intentional violations and affecting prevention effectiveness.
A Conv1d & Dilated Convolution deep learning model is constructed, which combines EEG signal processing and classification components. Through the perceived benefit experiment method, real-time early warning and intervention for intentional violations are achieved, and an instant feedback mechanism is introduced to ensure data accuracy and reliability.
It improves the accuracy of predicting intentional violations, enables real-time early warning and intervention, reduces operational risks, ensures the accuracy and reliability of collected data, and enhances the comprehensiveness and applicability of monitoring.
Smart Images
Figure CN122229470A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of monitoring technology, and in particular to the monitoring of intentional misconduct based on electroencephalogram (EEG) signals. Current applications of EEG technology in the monitoring of intentional misconduct primarily focus on assessing individual aspects such as fatigue levels and inattention, failing to comprehensively cover and deeply analyze the multiple factors that trigger intentional misconduct. This invention provides a system and method for monitoring intentional misconduct based on EEG signals, and constructs a Conv1d & Dilated Convolution deep learning model to address the problem of existing technologies failing to comprehensively analyze the multiple factors contributing to intentional misconduct. Background Technology
[0002] According to Bode's accident causal chain theory, the root cause of accidents resulting in injuries or fatalities is management deficiencies. Management can be divided into functions such as person-to-person and person-to-object control, which are processes for achieving group goals, with people being the core of the management process. Safety managers use management decision-making tools to analyze, evaluate, and select the safest actions to execute. Individual behavior stems directly from their decision-making process. In addition to factors such as fatigue and distraction, perceived benefit is considered to be one of the important factors influencing individual decision-making processes. When people are in a state of high perception, there is a probability of underestimating risks, leading to intentional violations; when people are in a state of low perception, there is a probability of making simplistic decisions, leading to intentional violations.
[0003] Among existing technologies, compared to other methods for monitoring intentional violations, such as on-site monitoring, video image monitoring, or mutual monitoring, electroencephalography (EEG), as a non-invasive technique, can directly monitor the brain's electrophysiological activity, and the recorded characteristic information cannot be actively concealed by individuals. Based on the theoretical framework of neuroscience, in-depth exploration of the connections between psychological phenomena and behavioral activities related to human safety holds the promise of revealing the intrinsic neural mechanisms driving behavior at a fundamental level. This exploration has significant theoretical and practical value for preventing intentional violations, indicating that EEG technology has broad application potential in the field of intentional violation identification and prevention.
[0004] However, the current application of electroencephalography (EEG) technology in the monitoring of intentional misconduct is still in a relatively preliminary and limited stage, mainly focusing on the assessment of individual aspects such as fatigue level and inattention level, without comprehensively covering and deeply analyzing the diverse factors that trigger intentional misconduct. This limitation restricts our in-depth understanding of the complex internal mechanisms by which the brain drives intentional misconduct, thus affecting the effectiveness and potential of EEG technology in preventing intentional misconduct. Summary of the Invention
[0005] The technical problem this invention aims to solve is to address the shortcomings of existing technologies by proposing a violation monitoring system and method based on electroencephalogram (EEG) signals. It also constructs a Conv1d & Dilated Convolution deep learning model to address the failure of existing technologies to comprehensively analyze the multiple factors contributing to intentional violations. This invention applies the perceived benefit experiment method to intentional violation monitoring, improving prediction accuracy. Furthermore, it introduces an instant feedback mechanism to achieve real-time early warning and intervention for intentional violations, reducing operational risks and ensuring the accuracy and reliability of the collected EEG signals and intentional violation data, thus providing a solid foundation for monitoring.
[0006] To achieve the above objectives, this invention proposes a system for monitoring intentional violations based on electroencephalogram (EEG) signals, specifically comprising: The EEG acquisition device includes dry electrodes, an EEG cap, an EEG filter and signal amplification chip, and a data transmission chip, used to acquire EEG data from subjects; and to personalize the configuration of the dry electrodes for data acquisition. The EEG signal processing and classification component is connected to the EEG acquisition module and is used to receive EEG signals and perform a preprocessing process, which includes filtering, noise reduction and feature selection. The response device, connected to the EEG signal processing and classification component, receives data on the subject's intentional violation of rules and provides real-time feedback to the subject through visual, auditory, and tactile means. The data storage unit, connected to the EEG signal processing and classification component, is used to store the collected EEG signals and intentional violation data. It analyzes the data based on the intentional violation data and an unsafe behavior decision-making influence model based on changes in theta and alpha power to verify the accuracy of the test results, provide data support for intentional violation monitoring, and feed the analysis results back to the response device.
[0007] Furthermore, personalized electrode configuration and data acquisition include: guiding subjects to correctly set dry electrodes according to the brain region location guidance in the experimental method, setting dry electrodes according to the preset brain region location, and collecting subjects' EEG data through an EEG acquisition device.
[0008] Furthermore, the EEG signal processing and classification component integrates a RISC-V architecture microcontroller, a high-precision analog-to-digital converter, a large-capacity storage medium, and a high-efficiency data transmission interface chip.
[0009] Furthermore, the EEG signal processing and classification component integrates a Conv1d & Dilated Convolution deep learning model to analyze the frequency and time domain characteristics of the subject's EEG signals, assess the subject's perceived benefits, cognitive state and emotional fluctuations, and deliver the processed data to the response device and data storage unit.
[0010] Furthermore, the Conv1d&Dilated Convolution deep learning model consists of an input dimensionality reduction module, a parallel processing branch module, a feature fusion module, and a residual connection module, specifically including: The input dimensionality reduction module is used to perform dimensionality reduction processing on multi-channel EEG signals, where the multi-channel EEG signals are set as follows: Where C is the number of channels and T is the time point, the dimensionality reduction operation of the input EEG signal is calculated using the following formula: (1) in, The output feature tensor after dimensionality reduction operation. * indicates a standard 1D convolution operation. BN is a pseudo-normalization. Paranoia; The parallel processing branches include a standard 1D convolution branch and a dilated convolution branch, used to extract local features and long-range dependency features, respectively. The standard 1D convolution branch is calculated using the following formula: (2) in, For the output feature tensor of a standard 1D convolution branch, K is the kernel size. It is the ReLU activation function. F This represents the total number of feature maps after subsequent stitching. The dilated convolution branch is calculated using the following formula: (3) in, The output feature tensor of the dilated convolution branch, The dilated convolution operator has a dilation rate of d. For hollow convolution kernel and Dimensional consistency; The feature fusion module is used to concatenate and fuse the outputs of the two branches. The dilated convolution output is cropped, taking only the first T time points from each channel and feature map. The calculation formula is as follows: (4) Then, channel splicing is performed, and the calculation formula is as follows: (5) in, This indicates splicing along the channel dimension. Let F be the dimension space of the concatenated feature map, a real matrix of F rows and T columns. The dilated convolution branch after the cropping operation outputs a feature tensor. The concatenated feature tensor; Finally, a 1x1 convolution is performed for fusion, calculated using the following formula: (6) in, The final output feature tensor after feature fusion is... , To determine the number of output feature maps, The bias parameters are for a 1x1 convolution; The residual connection module is used to perform residual connections between the fused features and the input features. The calculation formula for residual connections is as follows: (7) in, This is the output of the entire Conv1d&Dilated Convolution deep learning model. ; (8).
[0011] Furthermore, the response device mainly consists of lights, sound, vibration motors, buttons, and control chips.
[0012] Furthermore, the data storage unit mainly consists of a storage chip, a storage controller, and an interface circuit.
[0013] A method for monitoring intentional violations based on electroencephalogram (EEG) signals includes the following steps: S1. Collect EEG data from subjects using an EEG-based intentional violation monitoring system; S2. According to the pre-set experimental method, a perceived benefit experiment was conducted on the subjects. Three perceived benefit conditions were set for the subjects: control condition, high perceived benefit condition, and low perceived benefit condition. These conditions were used to analyze the changes in the subjects' EEG data under different perceived benefit conditions. S3. Analyze the EEG characteristics of the subjects under different perceived benefit conditions, including the theta and alpha power in the frontal, central, and occipital regions; S4. Based on the analysis results of the Conv1d & Dilated Convolution deep learning model, determine the subjects' tendency to engage in intentional violation behavior under different perceived benefit conditions; S5. Provide real-time feedback to the subject through the response device to remind them to correct their behavior and reduce operational hazards; S6. Store the collected EEG signals and intentional violation data, and establish an unsafe behavior decision-making influence model based on intentional violation data and changes in theta and alpha power. Analyze the established unsafe behavior decision-making influence model to verify the accuracy of the experimental results, provide data support for intentional violation monitoring, and feed the analysis results back to the response device.
[0014] Furthermore, the S2 pre-set experimental method includes: selecting decision-making scenario images and two outcome images corresponding to safe and unsafe behaviors as stimulus materials. Subjects complete three rounds of experiments in the order of three perceived benefit decision conditions, with a 30-minute rest between rounds. The stimulus presentation sequence is as follows: a "+" pattern lasting 500ms; a blank screen for 500ms; presentation of stimulus material for 5s; a blank screen for 800ms, and so on until the end of the experiment.
[0015] Furthermore, perceived benefit trials include: Under controlled conditions, when participants make unsafe behavioral decisions and no accident occurs, there is no additional gain or loss; Under conditions of high perceived benefit, there is a benefit when participants make unsafe behavioral decisions and no accident occurs; Under conditions of low perceived benefit, participants suffer losses regardless of whether an accident occurs when they make unsafe behavioral decisions.
[0016] Compared with the prior art, the present invention achieves the following technical effects: This invention applies the perceived benefit testing method to the field of intentional violation monitoring. By using the stimulus decision-making materials of subjects in their actual work and life, the system objectively analyzes and judges whether they have intentional violations, thereby providing early warnings and prompts to avoid accidents.
[0017] The unsafe behavior decision-making influence model proposed in this invention, based on data of intentional violations and changes in theta and alpha power, can accurately identify intentional violations.
[0018] The personalized electrode configuration and data acquisition process designed in this invention can ensure the accuracy and reliability of the data; and it introduces an instant feedback and behavior adjustment mechanism to achieve real-time early warning and intervention for intentional violations.
[0019] This invention constructs a Conv1d & Dilated Convolution model, which can significantly improve the signal-to-noise ratio of key frequency bands and enhance the detection accuracy of intentional violations while maintaining real-time performance. Attached Figure Description
[0020] For ease of explanation, the present invention will be described in detail below with reference to specific embodiments and accompanying drawings.
[0021] Figure 1 A schematic diagram illustrating the workflow of an intentional violation monitoring system based on electroencephalogram (EEG) signals, provided as an embodiment of the present invention; Figure 2 A schematic flowchart of a method for monitoring intentional violations based on electroencephalogram (EEG) signals provided in an embodiment of the present invention; Figure 3 A flowchart of a perceptual test method for monitoring intentional violations based on electroencephalogram (EEG) signals, provided in an embodiment of the present invention; Figure 4 An operational status diagram of an intentional violation monitoring system based on electroencephalogram (EEG) signals provided in an embodiment of the present invention; Figure 5 The flowchart shows the operation of the Conv1d&Dilated Convolution deep learning model constructed for this invention. Detailed Implementation
[0022] The following are specific embodiments of the present invention, described in conjunction with the accompanying drawings, to further illustrate the technical solutions of the present invention. However, the present invention is not limited to these embodiments. Specific details, such as particular configurations, are provided in the following description merely to aid in a comprehensive understanding of the embodiments of the present invention. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present invention.
[0023] Example 1: This invention provides a system for monitoring intentional misconduct based on electroencephalogram (EEG) signals, the workflow of which is as follows: Figure 1 As shown: An EEG acquisition device is set up, which includes dry electrodes, an EEG cap, an EEG filter and signal amplification chip, and a data transmission chip, for collecting the EEG data of the subject.
[0024] The EEG acquisition device is checked to determine if it can work properly. If it cannot work properly, the EEG acquisition device is modified and its status is reconfirmed until it can work properly. If it can work properly, dry electrodes are set according to the preset brain region locations, and the subject's EEG data is collected through the EEG acquisition device and transmitted to the EEG signal processing and classification component for signal processing and classification.
[0025] The signal processing and classification component receives and records the collected EEG signals, performs a preprocessing procedure, and analyzes the EEG characteristics of the subjects through the Conv1d & Dilated Convolution model to determine their perceived benefit status and tendency to engage in intentional violations.
[0026] Based on the analysis results from the Conv1d & Dilated Convolution model, the response device provides real-time feedback to the subject through visual and auditory means, reminding them to correct their behavior and reduce operational risks.
[0027] Finally, the data storage unit stores the collected EEG signals and data on intentional violations, and analyzes them using an unsafe behavior decision-making influence model based on intentional violation data and changes in theta and alpha power, further verifying the accuracy of the experimental results and providing data support for the monitoring of intentional violations.
[0028] Example 2: like Figure 2 , 3 As shown, this embodiment of the invention provides a method for monitoring intentional violations based on perceived benefits, specifically including the following steps: Experiment preparation phase: 1. Recruit test subjects: Recruit a certain number of volunteers from the target group to participate in the experiment to ensure their representativeness, and put EEG caps on them to ensure good contact between the dry electrodes and the preset brain regions.
[0029] 2. Set perceived benefit conditions: Three perceived benefit conditions are set: a control condition, a high perceived benefit (HPB) condition, and a low perceived benefit (LPB) condition. The rules for each perceived benefit condition are then defined. Control condition: There is no additional gain or loss when participants make unsafe behavioral decisions and no accident occurs; HPB condition: There is a benefit when participants make unsafe behavioral decisions and no accident occurs; LPB condition: When participants make unsafe behavioral decisions, there will be losses regardless of whether an accident occurs; 3. Stimulus material selection: Select the initial decision-making scenario and its corresponding two types of potential outcome images (representing the decision consequences of safe and unsafe behaviors, respectively) as stimulus materials.
[0030] Experimental process phases: 1. Arrange the test rounds. Throughout the process, the subjects need to make behavioral decisions under three perceived benefit conditions in sequence: control, high perceived benefit, and low perceived benefit. There is a 30-minute break between test rounds.
[0031] 2. Stimulus presentation, the system follows as follows: Figure 3 The preset procedure shown presents the stimulus material on the screen. The specific stimulus presentation procedure is as follows: 500ms "+" pattern, 500ms blank screen, 5s stimulus material, 800ms blank screen, and so on until the end of the experiment.
[0032] 3. EEG Data Acquisition and Signal Classification Processing: During the experiment, EEG data was acquired from the subjects using an EEG acquisition device. The theta and alpha powers (input EEG signals) of the frontal, central, and occipital lobe regions were recorded. Subsequently, the signal processing component ran the Conv1d & Dilated Convolution model (its operation flow is as follows). Figure 5 As shown), the input EEG signals are processed sequentially as follows: Input dimensionality reduction: Use 1x1 convolution to reduce the number of channels and improve computational efficiency; Parallel convolution: One path uses standard 1D convolution to capture local features, while the other path uses dilated convolution to capture long-range temporal dependencies; Feature fusion: The two outputs are cropped, aligned, and then concatenated, and then the features are fused through a 1x1 convolution. Residual connection: The fused features are added to the input to alleviate gradient vanishing and form a module output; The model ultimately outputs a classification result for intentional violations, which is then analyzed to obtain the experimental results: Under high-perceived-gain (HPB) conditions, the theta power in the frontal, central, and occipital lobes was significantly reduced compared to the control condition. Under low-perceived-gain (LPB) conditions, the theta power in the central and occipital regions was significantly lower than that in the control condition.
[0033] Implementation feedback and data storage phase: Based on the comprehensive analysis results, when the model determines that the subject exhibits a high-risk tendency to violate regulations under HPB or LPB conditions, the response device provides real-time feedback to the subject through visual and auditory means, reminding them to correct their behavior, reduce operational hazards, and improve the accuracy and effectiveness of monitoring. All data is simultaneously stored in the data storage unit and analyzed using an unsafe behavior decision-making influence model based on intentional violation data and changes in theta and alpha power, further verifying the accuracy of the experimental results and providing data support for intentional violation monitoring.
[0034] This embodiment demonstrates that the system can effectively distinguish intentional violation tendencies driven by different perceived benefits based on EEG signals and achieve real-time early warning.
[0035] Example 3: This embodiment, based on embodiment 2, adds emotion-inducing factors and emphasizes the closed-loop adaptive characteristics of the system, such as... Figure 2 and Figure 4 The system operation status diagram is shown below. The purpose is to test the system's comprehensive monitoring and intervention capabilities in more complex and realistic scenarios.
[0036] The core steps are similar to those in Example 2, but with the following enhancements and extensions: Experiment preparation phase: 1. Recruit test subjects: Recruit a certain number of people from the target group to participate in the experiment to ensure their representativeness, and put EEG caps on them to ensure good contact between the dry electrodes and the preset brain regions.
[0037] 2. Set perceived benefit conditions: Three perceived benefit conditions are set: control, high perceived benefit (HPB), and low perceived benefit (LPB), and the rules for each condition are clearly defined: Control condition: There is no additional gain or loss when participants make unsafe behavioral decisions and no accident occurs; HPB condition: There is a benefit when participants make unsafe behavioral decisions and no accident occurs; LPB condition: When participants make unsafe behavioral decisions, there will be losses regardless of whether an accident occurs; 3. Stimulus Material Selection: The initial decision-making scenario and its corresponding two types of potential outcome images (representing the decision consequences of safe and unsafe behaviors, respectively) were selected as stimulus materials. Additionally, emotionally induced videos were collected as part of the stimulus materials. By comprehensively considering the stimulus materials, the selected decision-making scenarios and outcome images for intentional violations were more integrated, potentially combining specific work scenarios and emotional contexts to analyze the impact of emotional states on intentional violations.
[0038] During the experimental process: 1. Arrange the test rounds. Throughout the process, the subjects need to make behavioral decisions under three perceived benefit conditions in sequence: control, high perceived benefit, and low perceived benefit. There is a 30-minute break between test rounds.
[0039] 2. Stimulus presentation, the system follows as follows: Figure 3 The preset procedure shown presents the stimulus material on the screen. The specific stimulus presentation procedure is as follows: 500ms "+" pattern, 500ms blank screen, 5s stimulus material, 800ms blank screen, and so on until the end of the experiment.
[0040] 3. EEG Data Acquisition and Signal Classification Processing: During the experiment, EEG data was acquired from the subjects using an EEG acquisition device. The theta and alpha powers (input EEG signals) of the frontal, central, and occipital lobe regions were recorded. Subsequently, the signal processing component ran the Conv1d & Dilated Convolution model (its operation flow is as follows). Figure 5 As shown), the input EEG signals are processed sequentially as follows: Input dimensionality reduction: Use 1x1 convolution to reduce the number of channels and improve computational efficiency; Parallel convolution: One path uses standard 1D convolution to capture local features, while the other path uses dilated convolution to capture long-range temporal dependencies; Feature fusion: The two outputs are cropped, aligned, and then concatenated, and then the features are fused through a 1x1 convolution. Residual connection: The fused features are added to the input to alleviate gradient vanishing and form a module output; The model ultimately outputs a classification result for intentional violations, which is then analyzed to obtain the experimental results: Under high-perceived-gain (HPB) conditions, the theta power in the frontal, central, and occipital lobes was significantly reduced compared to the control condition. Under low-perceived-gain (LPB) conditions, the theta power in the central and occipital regions was significantly lower than that in the control condition.
[0041] When analyzing EEG data, it is possible to determine the tendency of subjects to engage in intentional violation behavior under different emotional states and perceived benefits, and to establish a more comprehensive monitoring model for intentional violation behavior. This makes the closed-loop characteristics of the entire system more prominent. Once the risk level increases, the response device not only issues a warning, but can also automatically adjust the difficulty or feedback intensity of subsequent tests according to the operating status of the monitoring system, forming a closed-loop retrieval mechanism of acquisition-analysis-feedback-adjustment.
[0042] Implementation feedback and data storage phase: Based on the comprehensive analysis results, when the model determines that the subject exhibits a high-risk tendency to violate regulations under HPB or LPB conditions, the response device provides real-time feedback to the subject through visual and auditory means, reminding them to correct their behavior, reduce operational hazards, and improve the accuracy and effectiveness of monitoring. All data is simultaneously stored in the data storage unit and analyzed using an unsafe behavior decision-making influence model based on intentional violation data and changes in theta and alpha power, further verifying the accuracy of the experimental results and providing data support for intentional violation monitoring.
[0043] This embodiment demonstrates that the system of the present invention can not only perform real-time monitoring, but also dynamically adapt to changes in the subject's state through closed-loop feedback, greatly improving the accuracy of monitoring and its applicability in real-world complex environments.
[0044] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other.
[0045] Those skilled in the art to which this application pertains may make various modifications or additions to the specific embodiments described, or adopt similar methods to replace them, without departing from the inventive concept of this application or exceeding the scope defined by the appended claims.
Claims
1. A system for monitoring intentional violations based on electroencephalogram (EEG) signals, characterized in that... include: The EEG acquisition device includes dry electrodes, an EEG cap, an EEG filter and signal amplification chip, and a data transmission chip, used to acquire EEG data from subjects. Personalized configuration and data acquisition of dry electrodes; The EEG signal processing and classification component is connected to the EEG acquisition module and is used to receive EEG signals and perform a preprocessing process, which includes filtering, noise elimination and feature selection. The response device, connected to the EEG signal processing and classification component, receives data on the subject's intentional violation of rules and provides real-time feedback to the subject through visual, auditory, and tactile means. The data storage unit, connected to the EEG signal processing and classification component, is used to store the collected EEG signals and intentional violation behavior data. It also calculates the impact of unsafe behavior decisions based on behavioral indicators of the subjects under different perceived benefit conditions, including but not limited to the number of safe behavior choices, the number of unsafe behavior choices, reaction time, choice tendency, and decision condition labels, using a model of unsafe behavior decision influence based on a preset brain region range. Frequency band (4-7 Hz) and The power values in the frequency band (8-13 Hz) are calculated, and the power change relative to the control conditions is further calculated. and ; behavioral indicators, , The model constructs a combined feature vector based on the cross-brain region differences according to fixed rules to reflect the individual's decision-making state under different perceived benefit conditions; calculates an unsafe behavior risk score based on the model output to predict the probability of an individual engaging in intentional violation under the current perceived benefit conditions; and transmits the risk score as input to the response module of the monitoring system to trigger visual, auditory, or tactile feedback to achieve real-time early warning of intentional violation.
2. The intentional violation monitoring system based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The personalized electrode configuration and data acquisition include: guiding the subject to correctly set the dry electrodes according to the brain region location guidance in the experimental method, setting the dry electrodes according to the preset brain region location, and collecting the subject's EEG data through the EEG acquisition device.
3. The intentional violation monitoring system based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The EEG signal processing and classification component integrates a RISC-V architecture microcontroller, a high-precision analog-to-digital converter, a large-capacity storage medium, and a high-efficiency data transmission interface chip.
4. The intentional violation monitoring system based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The EEG signal processing and classification component integrates a Conv1d & Dilated Convolution deep learning model, which is used to analyze the frequency domain and time domain characteristics of the subject's EEG signals, assess the subject's perceived benefits, cognitive state and emotional fluctuations, and deliver the processed data to the response device and data storage unit.
5. The intentional violation monitoring system based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The Conv1d&Dilated Convolution deep learning model consists of an input dimensionality reduction module, a parallel processing branch module, a feature fusion module, and a residual connection module, specifically including: The input dimensionality reduction module is used to perform dimensionality reduction processing on multi-channel EEG signals, where the multi-channel EEG signals are set as follows: Where C is the number of channels and T is the time point, the dimensionality reduction operation of the input EEG signal is calculated using the following formula: (1) in, The output feature tensor after dimensionality reduction operation. * indicates a standard 1D convolution operation. BN is a pseudo-normalization. Paranoia; The parallel processing branch module includes a standard 1D convolution branch and a dilated convolution branch, used to extract local features and long-range dependency features, respectively. The standard 1D convolution branch is calculated using the following formula: (2) in, For the output feature tensor of a standard 1D convolution branch, K is the kernel size. It is the ReLU activation function. F This represents the total number of feature maps after subsequent stitching. The dilated convolution branch is calculated using the following formula: (3) in, The output feature tensor of the dilated convolution branch, The dilated convolution operator has a dilation rate of d. For hollow convolution kernel and Dimensional consistency; The feature fusion module is used to concatenate and fuse the outputs of the two branches. The dilated convolution output is cropped, taking only the first T time points from each channel and feature map. The calculation formula is as follows: (4) The formula for calculating channel splicing is as follows: (5) in, This indicates splicing along the channel dimension. Let F be the dimension space of the concatenated feature map, a real matrix of F rows and T columns. The dilated convolution branch after the cropping operation outputs a feature tensor. The concatenated feature tensor; Finally, a 1x1 convolution is performed for fusion, calculated using the following formula: (6) in, The final output feature tensor after feature fusion is... , To determine the number of output feature maps, The bias parameters are for a 1x1 convolution; The residual connection module is used to perform residual connections between the fused features and the input features. The calculation formula for residual connections is as follows: (7) in, This is the output of the entire Conv1d&Dilated Convolution deep learning model. ； (8)。 6. The intentional violation monitoring system based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The response device mainly consists of a light, a sound system, a vibration motor, a button, and a control chip.
7. The intentional violation monitoring system based on electroencephalogram (EEG) signals according to claim 1, characterized in that, The data storage unit mainly consists of a storage chip, a storage controller, and an interface circuit.
8. A method for monitoring intentional violations based on electroencephalogram (EEG) signals, as described in claims 1-7, characterized in that... Includes the following steps: S1. Collect EEG data from subjects using an EEG-based intentional violation monitoring system; S2. According to the pre-set experimental method, a perceived benefit experiment was conducted on the subjects. Three perceived benefit conditions were set for the subjects: control condition, high perceived benefit condition, and low perceived benefit condition. These conditions were used to analyze the changes in the subjects' EEG data under different perceived benefit conditions. S3. Analyze the EEG characteristics of the subjects under different perceived benefit conditions, including the theta and alpha power in the frontal, central, and occipital regions; S4. Based on the analysis results of the Conv1d & Dilated Convolution deep learning model, determine the subjects' tendency to engage in intentional violation behavior under different perceived benefit conditions; S5. Provide real-time feedback to the subject through the response device to remind them to correct their behavior and reduce operational hazards; S6. Store the collected EEG signals and intentional violation data, and establish an unsafe behavior decision-making influence model based on intentional violation data and changes in theta and alpha power. Analyze the established unsafe behavior decision-making influence model to verify the accuracy of the experimental results, provide data support for intentional violation monitoring, and feed the analysis results back to the response device.
9. The method for monitoring intentional violations based on electroencephalogram (EEG) signals according to claim 8, characterized in that, The S2 preset test method includes: selecting decision-making scenario images and two outcome images corresponding to safe and unsafe behaviors as stimulus materials. Subjects complete three rounds of tests in the order of three perceived benefit conditions, with a 30-minute rest between test rounds. The stimulus presentation process is as follows: a "+" pattern lasting 500ms; a blank screen for 500ms; presentation of stimulus material for 5s; a blank screen for 800ms, and so on until the end of the test.
10. The method for monitoring intentional violations based on electroencephalogram (EEG) signals according to claim 8, characterized in that, The perceived benefit test includes: Under controlled conditions, when participants make unsafe behavioral decisions and no accident occurs, there is no additional gain or loss; Under conditions of high perceived benefit, there is a benefit when participants make unsafe behavioral decisions and no accident occurs; Under conditions of low perceived benefit, participants suffer losses regardless of whether an accident occurs when they make unsafe behavioral decisions.