A work safety risk intelligent perception and early warning method based on electroencephalogram data

Abstract

The embodiment of the present application provides a kind of based on electroencephalogram data's job safety risk intelligent perception and early warning method, belong to data processing technical field, specifically include: obtain field environment image data;Call visual recognition model to identify field environment image data;Image recognition result is matched with preset accident hidden danger category library, generates the first kind of risk assessment result for representing job safety risk level;Acquire the electroencephalogram data of hard hat wearer;Feature extraction is carried out to electroencephalogram data, generates electroencephalogram risk feature vector;Electroencephalogram risk feature vector is analyzed and handled, generates the second kind of risk assessment result for representing the job safety risk level perceived by wearer;According to the first kind of risk assessment result and the second kind of risk assessment result, integrated risk value is calculated;When integrated risk value is higher than preset threshold value, generate early warning information and send management terminal.The scheme of the present application improves the early warning accuracy and real-time performance.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method for intelligent perception and early warning of workplace safety risks based on electroencephalogram (EEG) data. Background Technology

[0002] Currently, with the accelerated development of intelligent technology in high-risk operations and the continuous upgrading of operational safety management technologies, smart safety helmets have emerged as a key piece of equipment to ensure the safety of workers. At present, smart safety helmet technology has gradually evolved from its early basic protective functions towards data collection and status monitoring.

[0003] In traditional smart safety helmet technology, the assessment of operational safety risks primarily relies on built-in high-precision sensors of various types and basic data transmission mechanisms. These high-precision sensors can specifically collect key environmental parameters such as temperature, humidity, harmful gas concentration, and dust. Some also integrate proximity sensors to monitor potential hazards such as high-voltage electricity. When the detected data exceeds a preset safety threshold, local warnings are triggered through sound, light, vibration, and other means, and the data is transmitted to the back-end management system in real time, enabling real-time perception and alarm of environmental risks.

[0004] However, current smart safety helmet technology has significant limitations in intelligent perception and early warning of operational safety risks: First, it has poor adaptability to complex scenarios. In conditions such as dust, backlighting, and equipment obstruction, sensors are easily interfered with, and a single vision or sensing module cannot reliably capture risks. Second, it lacks multi-dimensional fusion capabilities, mostly monitoring only isolated parameters such as temperature, humidity, and gas concentration, lacking integrated analysis of risk information such as personnel protective status and body posture, and thus unable to identify complex hazards. Early warning response is also delayed, with some algorithms having high inference latency, and fixed thresholds cannot be dynamically adjusted with the environment, making it difficult to respond to sudden risks in a timely manner.

[0005] It is evident that there is an urgent need for an intelligent perception and early warning method for workplace safety risks based on EEG data, which offers high accuracy and real-time performance. Summary of the Invention

[0006] In view of this, embodiments of the present invention provide a method for intelligent perception and early warning of work safety risks based on electroencephalogram (EEG) data, which at least partially solves the problems of poor accuracy and real-time performance of early warning in the prior art.

[0007] This invention provides a method for intelligent perception and early warning of workplace safety risks based on electroencephalogram (EEG) data, applied to a smart safety helmet. The smart safety helmet includes a helmet body and a camera mounted on the helmet body.

[0008] Step 1: Acquire real-time environmental image data from the camera;

[0009] Step 2: Call the pre-trained visual recognition model to recognize the on-site environmental image data and generate image recognition results;

[0010] Step 3: Match the image recognition results with a pre-defined accident hazard category library to generate a first-class risk assessment result that characterizes the level of operational safety risks;

[0011] Step 4: Obtain the EEG data of the helmet wearer;

[0012] Step 5: Extract features from the EEG data to generate an EEG risk feature vector;

[0013] Step 6: Analyze and process the EEG risk feature vector to generate a second type of risk assessment result that characterizes the wearer's perceived level of job safety.

[0014] Step 7: Calculate the comprehensive risk value based on the results of the first type of risk assessment and the results of the second type of risk assessment;

[0015] Step 8: When the overall risk value is higher than the preset threshold, generate an early warning message and send it to the management terminal.

[0016] According to a specific implementation of an embodiment of the present invention, step 3 specifically includes:

[0017] Step 3.1: Extract multi-dimensional feature information related to safety hazards from the image recognition results;

[0018] Step 3.2: Match the multi-dimensional feature information with the accident hazard category library to obtain the first matching result. The first matching result includes the matched hazard category and its corresponding basic risk value. The accident hazard category library includes multiple hazard categories and risk weight coefficients corresponding to each category.

[0019] Step 3.3: Based on the type of the current work scenario, the basic risk value is adjusted by weighting using a scenario adaptation correction algorithm to generate the first type of risk assessment result.

[0020] According to a specific implementation of an embodiment of the present invention, step 3.2 specifically includes:

[0021] Step 3.2.1: Input multi-dimensional feature information into the pre-built AI hazard classification model to determine the target hazard category;

[0022] Step 3.2.2: Based on the target hazard category, search for and obtain the corresponding basic risk value from the accident hazard category library to form the first matching result.

[0023] According to a specific implementation of an embodiment of the present invention, step 3.2.1 specifically includes:

[0024] Step 3.2.1.1: Perform fusion processing on the multi-dimensional feature information to generate a multimodal feature vector;

[0025] Step 3.2.1.2: Calculate the similarity score between the multimodal feature vector and the feature template vector of each hazard category in the accident hazard category library;

[0026] Step 3.2.1.3: Determine the hazard category with the highest matching degree as the target hazard category based on the similarity score.

[0027] According to a specific implementation of the present invention, the expression for the similarity score is:

[0028] ;

[0029] in, It is the first of the multimodal eigenvectors Each dimension of feature value, It is the number of feature dimensions of the multimodal feature vector. It is the first feature template vector of a certain hazard category in the accident hazard category library. Each dimension of feature value.

[0030] According to a specific implementation of an embodiment of the present invention, step 6 specifically includes:

[0031] Step 6.1: Compare the EEG risk feature vector with the preset individual EEG risk benchmark library to obtain a second matching result that characterizes the degree of preliminary matching between the vector and each risk perception level interval in the library. The individual EEG risk benchmark library is constructed based on the wearer's historical EEG data and corresponding work scenario risk records, and includes EEG feature vector intervals and weights corresponding to different risk perception levels.

[0032] Step 6.2: Calculate the similarity measure between the EEG risk feature vector and the benchmark vector corresponding to each risk perception level in the individual EEG risk benchmark database. The similarity measure includes Euclidean distance and / or feature overlap.

[0033] Step 6.3: Calculate the perceptual matching value based on the similarity metric and the complexity coefficient of the current task scenario;

[0034] Step 6.4: Determine the wearer's risk perception level based on the preset perception range where the perception matching value is located, and generate a second type of risk assessment result. The perception range includes a low perception range, a medium perception range, and a high perception range.

[0035] According to a specific implementation of an embodiment of the present invention, before step 6.3, the method further includes:

[0036] Obtain multi-dimensional environmental feature parameters of the current work scenario, wherein the multi-dimensional environmental feature parameters include equipment spatial distribution density, personnel movement and interaction frequency, and environmental electromagnetic interference intensity;

[0037] Based on the preset environmental feature weights, the environmental feature parameters of each dimension are normalized to obtain the normalized parameter values ​​of each dimension.

[0038] The complexity coefficient is obtained by weighted summation of the normalized parameter values ​​for each dimension.

[0039] According to a specific implementation of an embodiment of the present invention, step 7 specifically includes:

[0040] Step 7.1: Normalize the results of the first type of risk assessment and the second type of risk assessment, mapping them to a unified scaling range;

[0041] Step 7.2: Using a weighted fusion algorithm, the normalized first-type risk assessment results and the second-type risk assessment results are fused and calculated to output a comprehensive risk value.

[0042] The intelligent perception and early warning scheme for work safety risks based on EEG data in this embodiment of the invention includes: Step 1, acquiring real-time on-site environmental image data collected by a camera; Step 2, calling a pre-trained visual recognition model to recognize the on-site environmental image data and generating image recognition results; Step 3, matching the image recognition results with a preset accident hazard category library to generate a first type of risk assessment result characterizing the level of work safety risks; Step 4, acquiring the EEG data of the helmet wearer; Step 5, extracting features from the EEG data to generate an EEG risk feature vector; Step 6, analyzing and processing the EEG risk feature vector to generate a second type of risk assessment result characterizing the level of work safety perceived by the wearer; Step 7, calculating a comprehensive risk value based on the first and second type of risk assessment results; Step 8, when the comprehensive risk value is higher than a preset threshold, generating an early warning message and sending it to the management terminal.

[0043] The beneficial effects of this invention are as follows: By combining visual environment recognition with the analysis of the wearer's electroencephalographic (EEG) responses, a dual-dimensional assessment system of "objective environmental risk" and "subjective risk perception" is constructed. This method utilizes a trained visual model and a database of potential hazards to accurately identify external risks from images. Simultaneously, it analyzes the wearer's neural responses based on an individualized EEG benchmark database to assess their intrinsic risk perception level. Furthermore, it integrates both to generate a comprehensive risk value, achieving a real-time, comprehensive, and personalized dynamic assessment of the overall operational risk. This effectively overcomes the shortcomings of traditional methods, such as poor scenario adaptability, limited information dimensions, and delayed early warnings. It significantly improves the accuracy, timeliness, and reliability of risk warnings in high-risk operational environments, providing intelligent protection for worker safety. Attached Figure Description

[0044] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0045] Figure 1 This is a flowchart illustrating an intelligent perception and early warning method for workplace safety risks based on electroencephalogram (EEG) data, provided as an embodiment of the present invention. Detailed Implementation

[0046] The embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0047] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that, in the absence of conflict, the following embodiments and features in the embodiments can be combined with each other. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0048] It should be noted that various aspects of embodiments within the scope of the appended claims are described below. It will be apparent that the aspects described herein can be embodied in a wide variety of forms, and any particular structure and / or function described herein is merely illustrative. Based on this invention, those skilled in the art will understand that one aspect described herein can be implemented independently of any other aspect, and two or more of these aspects can be combined in various ways. For example, any number of aspects set forth herein can be used to implement the device and / or practice the method. Additionally, this device and / or method can be implemented using structures and / or functionalities other than one or more of the aspects set forth herein.

[0049] It should also be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. The illustrations only show the components related to the present invention and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the form, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0050] Furthermore, specific details are provided in the following description to facilitate a thorough understanding of the examples. However, those skilled in the art will understand that the described aspects can be practiced without these specific details.

[0051] This invention provides a method for intelligent perception and early warning of work safety risks based on electroencephalogram (EEG) data. The method can be applied to the risk perception process in high-risk work scenarios such as critical construction operations, special operations in the power industry, mining operations, and transportation operations.

[0052] See Figure 1 This is a flowchart illustrating a method for intelligent perception and early warning of workplace safety risks based on electroencephalogram (EEG) data, provided by an embodiment of the present invention. Figure 1 As shown, the method mainly includes the following steps:

[0053] Step 1: Acquire real-time environmental image data from the camera;

[0054] In practice, the on-site environmental image data is collected in real time by the camera. The on-site environmental image data is environmental image data that reflects the current environmental conditions of the work area and is collected in real time by the camera installed on the smart safety helmet itself.

[0055] Specifically, the controller connects to the camera through a preset communication method and receives real-time environmental image data of the safety helmet wearer at the work site, i.e., on-site environmental image data.

[0056] Step 2: Call the pre-trained visual recognition model to recognize the on-site environmental image data and generate image recognition results;

[0057] In practical implementation, the visual recognition model is a pre-trained model used to analyze and recognize image data of the on-site environment captured by the camera on the smart safety helmet. The visual recognition model is trained based on a large number of work environment image samples and corresponding safety hazard annotation data. It can accurately identify key information related to safety hazards based on the on-site environmental image data in the current work scenario, such as unsafe personnel behavior, abnormal equipment status, and environmental risk factors, and generate corresponding structured recognition information containing this hazard-related information.

[0058] Image recognition results are structured recognition information obtained by processing and analyzing on-site environmental image data through a trained visual recognition model.

[0059] Specifically, the controller calls a pre-trained visual recognition model, inputs the acquired on-site environmental image data into the model, performs real-time image reasoning analysis on the information related to safety hazards in the work area in the image, and finally generates structured recognition information containing these hazard-related information, i.e., image recognition results.

[0060] Step 3: Match the image recognition results with a pre-defined accident hazard category library to generate a first-class risk assessment result that characterizes the level of operational safety risks;

[0061] In practice, the accident hazard category library is a pre-built structured database used to match and evaluate safety hazards in the work environment. It contains common safety hazard categories in various work scenarios, as well as the risk weight coefficient and basic risk value corresponding to each type of hazard.

[0062] The first type of risk assessment result is obtained by quantitatively assessing the level of operational safety risks based on image recognition results and a pre-set accident hazard category library.

[0063] Specifically, the controller first extracts information related to safety hazards from the image recognition results, then matches the information related to safety hazards with the hazard categories, risk weight coefficients and basic risk values ​​in the preset accident hazard category library, then quantitatively assesses the risk level of the current working environment, and finally generates the first type of risk assessment result to characterize the level of work safety risks.

[0064] For example, in one specific embodiment, generating a first-class risk assessment result based on image recognition results and a pre-defined accident hazard category library may include the following steps:

[0065] Step S301: Extract multi-dimensional feature information related to safety hazards from the image recognition results.

[0066] Among them, multi-dimensional feature information refers to the set of structured features with multiple angles and attributes that are related to safety hazards extracted from image recognition results.

[0067] Specifically, the controller performs feature analysis on the image recognition results, extracts multi-dimensional feature information associated with safety hazards, and uses feature vectors. Perform structured characterization, whereby... Indicates the first The specific numerical values ​​of the dimensional features, This represents the total dimension of the features.

[0068] Step S302: Match the multi-dimensional feature information with the preset accident hazard category library to generate a first matching result; the first matching result includes the matched hazard category and the basic risk value corresponding to the hazard category; the preset accident hazard category library includes multiple hazard categories and the risk weight coefficient of the corresponding hazard.

[0069] The first matching result is an intermediate result generated after matching multi-dimensional feature information with a preset accident hazard category library. It is used to determine the correspondence between image recognition content and hazard category and its basic risk level.

[0070] Specifically, the controller calculates the similarity between multi-dimensional feature information and feature templates in the accident hazard category library, determines the best matching hazard category through preset matching rules, and generates a first matching result containing the correspondence between hazard category and basic risk value based on preset data corresponding to that category in the hazard library. ,in, It is a matching hazard category identifier. It is the basic risk value corresponding to the hazard category.

[0071] Step S303: Based on the type of the current work scenario, the basic risk value is weighted and adjusted using a scenario adaptation correction algorithm to generate a first-class risk assessment result; the first-class risk assessment result is used to characterize the level of work safety risk.

[0072] Among them, the basic risk value is the initial risk quantification value set for each type of hazard in the preset accident hazard category library, which is used to characterize the inherent danger level of this type of hazard in standard operating scenarios.

[0073] The scenario adaptation correction algorithm is a calculation method that dynamically weights and adjusts the basic risk value based on the current work scenario type, and is used to match the risk assessment results with the risk characteristics of a specific work scenario.

[0074] The first type of risk assessment result is a quantitative assessment of the operational safety risk level based on image recognition results, combined with a pre-set accident hazard category library, and adjusted by a scenario adaptation correction algorithm.

[0075] Specifically, the controller calls the corresponding scenario weight coefficient based on the current job scenario type. Through scene adaptation correction algorithm Basic risk value Dynamic adjustments are made, among which, The revised risk value is ultimately calculated using the following formula to generate the quantified Type I risk assessment result. :

[0076] ;

[0077] in, It is the number of matched hazard categories. This is the result of the first risk assessment.

[0078] In this embodiment, the controller uses a dynamic weighting mechanism that adapts to different scenarios to achieve a quantitative assessment of risk levels under different working environments, effectively improving the accuracy of safety hazard identification and scenario adaptability.

[0079] Furthermore, matching the multi-dimensional feature information with a pre-defined accident hazard category library to generate a first matching result may include the following steps:

[0080] Based on multi-dimensional feature information, the target hazard category is determined through a pre-built AI hazard classification model.

[0081] Among them, the AI ​​hazard classification model is a computational model built with artificial intelligence technology to automatically analyze and classify multi-dimensional feature information in order to identify and determine the category of target hazards.

[0082] The target hazard category refers to the specific hazard category that best matches the safety hazards in the current working environment, determined by analyzing and identifying multi-dimensional feature information through an AI hazard classification model.

[0083] Specifically, the controller inputs multi-dimensional feature information into a pre-built AI hazard classification model, and calculates the matching probability distribution between features and categories through the multi-layer neural network inside the model. ,in Indicates belonging to the first The probability of each type of hazard. This represents the total number of hazard categories, and the category with the highest probability value is ultimately selected. As a target hazard category.

[0084] Based on the target hazard category, the corresponding basic risk value is obtained from the accident hazard category library to generate the first matching result.

[0085] Specifically, once the target hazard category is determined, the controller calls a pre-set accident hazard category library, searches the library based on the unique identifier of the target hazard category, extracts the corresponding basic risk value, and sets it as the target hazard category. Then, a first matching result containing the correspondence between target hazard category and basic risk value is generated, which can be represented by structured data. ,in, It is the name of the target hazard category. This is the basic risk value retrieved.

[0086] In this embodiment, the controller achieves automated identification of safety hazards and rapid acquisition of basic risk values ​​through AI intelligent classification and matching with the accident hazard category library, significantly improving the efficiency and accuracy of risk assessment.

[0087] Furthermore, based on multi-dimensional feature information, determining the target hazard category through a pre-built AI hazard classification model may include the following steps:

[0088] Feature extraction is performed on multi-dimensional feature information to generate multimodal feature vectors.

[0089] Among them, multimodal feature vectors refer to structured vector representations that can comprehensively represent the characteristics of various types of safety hazards, generated by fusing and abstracting multidimensional feature information.

[0090] Specifically, the controller employs a feature fusion algorithm to process multi-dimensional feature information. Preprocessing is performed to remove redundant data and normalize the features of each dimension.

[0091] Furthermore, the controller abstracts and integrates the normalized multi-dimensional features through a feature fusion algorithm to generate a unified multimodal feature vector. ,in Indicates the first Each fused feature component Indicates the number of feature dimensions.

[0092] The matching degree between the multimodal feature vector and the feature templates of each hazard category in the accident hazard category library is calculated using a vector similarity algorithm to obtain the similarity score corresponding to the multimodal feature vector.

[0093] Similarity score is a numerical index calculated by vector similarity algorithm to quantify the degree of matching between multimodal feature vectors and hazard category feature templates.

[0094] Specifically, the controller first retrieves the feature templates of each hazard category stored in the accident hazard category library, and then uses a vector similarity algorithm to calculate the matching degree between the multimodal feature vector and the feature templates of each hazard category in the accident hazard category library, thereby obtaining the similarity score corresponding to the multimodal feature vector.

[0095] The matching hazard category is determined based on the similarity score and used as the target hazard category.

[0096] Specifically, the controller first aggregates the similarity scores corresponding to the multimodal feature vectors obtained from S502, and constructs a score set. ,in, It is the first Similarity scores of multimodal feature vectors This represents the total number of hazard categories.

[0097] Furthermore, the controller uses the formula Determine the index of the maximum value in the set of fractions. Finally, the index The corresponding hazard category is determined as the target hazard category.

[0098] In this embodiment, the controller preprocesses and fuses multi-dimensional feature information to generate multimodal feature vectors, and then uses a vector similarity algorithm to calculate the matching degree and determine the hazard category corresponding to the maximum similarity. This achieves automated identification of operational safety hazard categories and provides an accurate basis for subsequent risk assessment.

[0099] In one embodiment, a vector similarity algorithm is used to calculate the matching degree between the multimodal feature vector and the feature templates of each hazard category in the accident hazard category library, to obtain the similarity score corresponding to the multimodal feature vector, including:

[0100] Calculate the similarity score using the following formula:

[0101] ;

[0102] in, It is a similarity score. It is the first of the multimodal eigenvectors Each dimension of feature value, It is the number of feature dimensions of the multimodal feature vector. It is the first feature template vector of a certain hazard category in the accident hazard category library. Each dimension of feature value.

[0103] In this embodiment, the controller accurately quantifies the degree of feature matching through the cosine similarity algorithm, providing a reliable numerical basis for hazard classification and effectively improving the accuracy and objectivity of target hazard category determination.

[0104] Step 4: Obtain the EEG data of the helmet wearer;

[0105] In practice, EEG data is physiological electrical signal data that reflects the state of brain neural activity, collected from the head of a smart helmet wearer by sensing devices, such as EEG acquisition instruments.

[0106] Specifically, the controller acquires the helmet wearer's electroencephalogram (EEG) data by receiving real-time physiological electrical signal data from the helmet wearer's head collected by sensing devices, which reflects the state of the helmet wearer's brain neural activity.

[0107] Step 5: Extract features from the EEG data to generate an EEG risk feature vector;

[0108] In practice, the feature extraction algorithm is a multi-dimensional EEG feature extraction method based on time-frequency analysis and nonlinear dynamics. Specifically, it includes wavelet transform for extracting time-frequency domain features, sample entropy for quantifying the complexity of EEG signals, and power spectral density analysis for obtaining energy distribution features in different frequency bands. The feature extraction algorithm can separate brain neural activity features related to risk perception state from the original EEG signal.

[0109] EEG risk feature vectors are vector data that can quantitatively characterize risk perception-related features in the wearer's brain neural activity.

[0110] Specifically, the controller processes the acquired EEG data of the helmet wearer, calls a feature extraction algorithm to extract brain neural activity features related to risk perception from the EEG data, and quantifies these features into a structured vector form, ultimately generating an EEG risk feature vector that can quantify and characterize the neural activity features related to risk perception of the helmet wearer.

[0111] Step 6: Analyze and process the EEG risk feature vector to generate a second type of risk assessment result to characterize the level of occupational safety risk perceived by the wearer;

[0112] In practice, the second type of risk assessment result is based on the EEG risk feature vector of the helmet wearer. By analyzing the neurophysiological response of the wearer in the current working environment, an assessment conclusion is generated to quantitatively characterize the risk level.

[0113] Specifically, the controller uses the generated EEG risk feature vector as a basis to analyze the wearer's neurophysiological response reflected by the vector, and assesses the risk of the helmet wearer in the current working environment. Finally, it generates a second type of risk assessment result to quantify the level of occupational safety risk.

[0114] In one specific embodiment, generating a second type of risk assessment result based on the EEG risk feature vector may include the following steps:

[0115] Step S601: Match the EEG risk feature vector with the EEG feature vector intervals corresponding to each risk perception level in the preset individual EEG risk benchmark library to obtain the second matching result; the individual EEG risk benchmark library is constructed based on the historical EEG data of the helmet wearer and the corresponding work scenario risk records, including feature weights corresponding to different risk perception levels; the second matching result is used to characterize the preliminary matching degree between the EEG risk features and each risk perception level interval.

[0116] Among them, the EEG risk feature vector refers to the structured vector data generated by the controller after processing the EEG data collected from the head of the helmet wearer through a feature extraction algorithm.

[0117] The Individual EEG Risk Benchmark Database refers to a personalized reference database constructed based on the historical EEG data of helmet wearers and their corresponding work scenario risk records. It includes EEG feature vector intervals and corresponding feature weights at different risk levels.

[0118] The EEG feature vector interval refers to the range of multidimensional vector values ​​defined within the feature space for different risk levels in an individual EEG risk benchmark database.

[0119] The second matching result refers to the descriptive result of the initial matching degree between the real-time acquired EEG risk feature vector and the EEG feature vector intervals corresponding to each risk level in the preset individual EEG risk benchmark database.

[0120] Specifically, the controller will acquire EEG risk feature vectors in real time. Compared with the risk perception levels in the individual EEG risk benchmark database Corresponding EEG feature vector interval A comparison was performed one by one, and then the EEG risk feature vector was calculated using the following formula. Falling into each interval membership degree :

[0121] ;

[0122] in, It is an indicator function, when Belongs to the interval The value is 1 if the condition is met, and 0 otherwise. It is the number of feature dimensions. It is the degree of membership.

[0123] Furthermore, the controller determines the membership degree. The magnitude of the numerical values ​​determines the initial matching degree between the EEG risk feature vector and each risk perception level interval, generating a second matching result. .

[0124] Step S602: Calculate the similarity measure between the EEG risk feature vector and the benchmark vector of the EEG feature vector interval corresponding to each perception level in the individual EEG risk benchmark database; the similarity measure includes Euclidean distance and feature overlap; the similarity measure is used to characterize the quantitative matching degree between the EEG risk feature vector and the benchmark vector corresponding to each perception level in the individual EEG risk benchmark database.

[0125] The similarity measure is a quantitative value of the degree of similarity or closeness between the real-time EEG risk feature vector and the benchmark vector corresponding to each risk level in the individual EEG risk benchmark library.

[0126] Specifically, the controller targets each risk level in the individual EEG risk benchmark database. The following formula is used to calculate the EEG risk feature vector. The reference vector corresponding to this level Euclidean distance between :

[0127] ;

[0128] in, It is an EEG risk feature vector In the The numerical value of dimension, It is the first The baseline vector corresponding to each level of risk perception is at the _ ... The numerical value of dimension, It is Euclidean distance.

[0129] Furthermore, the controller calculates the EEG risk feature vector using a preset algorithm. The reference vector corresponding to this level Feature overlap .

[0130] Furthermore, the controller is based on the calculated Euclidean distance. overlap with features This yields a set of similarity metrics at that perceptual level. .

[0131] Step S603: Based on the similarity metric and combined with the complexity coefficient of the current work scenario, obtain the perceptual matching value.

[0132] The complexity coefficient refers to the numerical index obtained by the controller after quantifying the multi-dimensional environmental feature parameters of the current work scenario, which is used to characterize the complexity of the work scenario.

[0133] Perceptual matching value is used to characterize the degree of final agreement between a helmet wearer's real-time EEG risk characteristics and their personal historical risk perception patterns in the current specific work environment.

[0134] Specifically, the controller is based on the calculated set of similarity metrics. First, the Euclidean distance The normalized proximity score is calculated using the following formula. :

[0135] ;

[0136] in, It is the proximity score. It is Euclidean distance.

[0137] Furthermore, the controller uses the following formula to correlate proximity scores with feature overlap. By performing linear combination, the basic physiological matching degree is obtained. :

[0138] ;

[0139] in, It is the basic physiological compatibility. and It is a preset combination weight and , It is the proximity score. It is the feature overlap degree.

[0140] Furthermore, the controller incorporates the complexity coefficient of the current work scenario. The perceptual matching value is calculated using the following formula. :

[0141] ;

[0142] in, It is a perceived matching value. It is the basic physiological compatibility. It is a dynamic moderating factor of environmental impact. It is the complexity coefficient.

[0143] Step S604: Generate a second type of risk assessment result based on the perception interval where the perception matching value is located; the perception interval includes a low perception interval, a medium perception interval, and a high perception interval; the second type of risk assessment result is used to characterize the level of work safety risk perceived by the helmet wearer.

[0144] The perception interval is a continuous range that is pre-divided on the numerical axis to define and determine the risk level to which the perceived matching value belongs.

[0145] The second type of risk assessment result is an assessment conclusion derived from the analysis of the helmet wearer's real-time electroencephalogram (EEG) data, used to quantitatively characterize the helmet wearer's current risk level.

[0146] Specifically, the controller will calculate the obtained perception matching value With the preset perception range threshold Comparison: If If so, it is determined to be in the low perception range and a second assessment result of low risk perception level is generated; if If it falls within the medium perception range, a second-type assessment result with a medium risk level is generated; if If so, it is determined to be in the high-perception range and generates a second-type assessment result with a high-risk level; among which, The pre-defined interval threshold, and satisfying .

[0147] In this embodiment, the controller integrates electroencephalographic matching degree and environmental complexity to achieve personalized assessment of the risk perception level of helmet wearers, effectively improving the accuracy of risk warning.

[0148] Furthermore, the complexity coefficient is obtained through the following method, which may include the following steps:

[0149] Obtain multi-dimensional environmental characteristic parameters of the current work scenario; environmental characteristic parameters include equipment spatial distribution density, personnel movement and interaction frequency, and environmental electromagnetic interference intensity.

[0150] Among them, multi-dimensional environmental characteristic parameters refer to a set of multiple types of data indicators obtained by the controller from the current working scenario, which are used to quantify the correlation between the complexity of the scenario and potential risks.

[0151] Specifically, the controller obtains the spatial distribution density of equipment through a unit area statistical method. The frequency of personnel movement and interaction is obtained by cross-statistical analysis of personnel trajectories within a time window. The intensity of environmental electromagnetic interference is obtained through an electromagnetic signal strength detector. This completes the set of multi-dimensional environmental feature parameters. Synchronous acquisition.

[0152] Based on the preset environmental feature weights, the environmental feature parameters of each dimension are normalized to obtain the normalized parameter values ​​corresponding to the environmental feature parameters of each dimension.

[0153] Among them, the environmental feature weight refers to the numerical coefficient pre-assigned to each multi-dimensional environmental feature parameter to characterize its relative importance in calculating the complexity coefficient.

[0154] Specifically, the controller invokes a preset environmental feature weight vector. ,in Corresponding to the spatial distribution density of equipment Frequency of personnel movement and interaction and environmental electromagnetic interference intensity The weights are then determined, and the max-min normalization method is used to apply the weights to the multi-dimensional environmental feature parameter set. The process involves calculating the normalized values ​​of each parameter to obtain a set of normalized parameter values.

[0155] The complexity coefficient is obtained by weighted summation of the normalized parameter values.

[0156] Specifically, the controller performs a dot product operation between the normalized set of parameter values ​​and a preset environmental feature weight vector to calculate the complexity coefficient. Ultimately, ensure the complexity coefficient To take values ​​in Scalars within the range.

[0157] Step 7: Calculate the comprehensive risk value based on the results of the first type of risk assessment and the results of the second type of risk assessment;

[0158] In practice, the comprehensive risk value is a quantitative indicator used to characterize the overall risk level of the operation by integrating the results of the first type of risk assessment and the results of the second type of risk assessment and then calculating it using a preset algorithm.

[0159] Specifically, the controller inputs the generated first-type risk assessment results and second-type risk assessment results into a preset algorithm. The algorithm then performs a fusion calculation on the first-type risk assessment results and second-type risk assessment results to obtain a comprehensive risk value that can quantitatively reflect the overall risk level of the current operation.

[0160] Step 8: When the overall risk value is higher than the preset threshold, generate an early warning message and send it to the management terminal.

[0161] In practice, the early warning information refers to the risk alarm notification that the controller automatically generates and sends to the management terminal when the comprehensive risk value is higher than the preset threshold.

[0162] The management terminal is a back-end management device that receives and processes early warning information sent from the controller. Connected to the controller via a communication network, the management terminal receives early warning information and associates it with the specific safety management personnel responsible for the work environment.

[0163] Specifically, the controller first compares the calculated comprehensive risk value with a preset threshold. If the comprehensive risk value is lower than the preset threshold, it generates an early warning message containing the current safety risk situation of the operation, and then sends the early warning message to the corresponding management terminal through a preset communication method.

[0164] This embodiment provides an intelligent perception and early warning method for workplace safety risks based on electroencephalogram (EEG) data. By combining visual environment recognition with analysis of the wearer's EEG physiological responses, it constructs a dual-dimensional assessment system of "objective environmental risk" and "subjective risk perception." This method utilizes a trained visual model and a database of potential hazards to accurately identify external environmental risks from images. Simultaneously, it analyzes the wearer's neural responses based on an individualized EEG benchmark database to assess their intrinsic risk perception level. Finally, it integrates both to generate a comprehensive risk value, achieving a real-time, comprehensive, and personalized dynamic assessment of the overall risk of the workplace environment. This effectively overcomes the shortcomings of traditional methods, such as poor scenario adaptability, limited information dimensions, and delayed early warnings. It significantly improves the accuracy, timeliness, and reliability of risk warnings in high-risk operations, providing intelligent protection for worker safety.

[0165] The method of the present invention will be further described below with reference to a specific embodiment. In the practical application scenario of the smart safety helmet based on EEG data to identify work safety risks, after the worker wears the smart safety helmet, the controller will synchronously acquire on-site environmental image data through the camera set on the safety helmet body, and enter the subsequent work safety risk assessment process, as follows:

[0166] Step 1: Call the trained visual recognition model to perform image recognition on the collected on-site environmental image data and generate image recognition results containing information related to safety hazards.

[0167] Step 2: Based on the image recognition results, a matching operation is performed using a pre-set accident hazard category library to generate the first type of risk assessment result that characterizes the level of operational safety risks.

[0168] Specifically, the pre-defined accident hazard category library contains the following twenty-two most common accident hazard categories:

[0169] Category 1 of Accident Hazards (Unsafe Human Behaviors): Neglecting safety, ignoring warnings, incorrect operation, causing safety devices to malfunction, using unsafe equipment, using hands instead of tools, improper storage of objects, entering dangerous locations at risk, climbing or sitting in unsafe positions, behaviors that interfere with or distract attention, neglecting or failing to use personal protective equipment and tools, unsafe attire, and incorrect handling of flammable, explosive, and other hazardous materials. A total of eleven accident hazards are included.

[0170] Category 2 of accident hazards (unsafe conditions of objects): lack or defective protective, safety, and signaling devices; defective equipment, facilities, tools, and accessories; lack or defective personal protective equipment; and poor working environment at production and construction sites. There are a total of four types of accident hazards.

[0171] Category 3 of Accident Hazards (Management Deficiencies): Management deficiencies include technical and design defects, insufficient safety production education and training, unreasonable labor organization, lack of inspection or guidance of on-site work, lack of safety production management rules and regulations, safety operating procedures and safety signs, lack of accident prevention and emergency measures or incomplete measures, and ineffective rectification of accident hazards, totaling seven types of accident hazards.

[0172] Step 3: Use the EEG data acquisition component associated with the helmet to acquire EEG data of the helmet wearer's head in real time.

[0173] Step 4: Use a feature extraction algorithm to process the acquired EEG data, extract features related to risk perception, and quantify them to generate an EEG risk feature vector.

[0174] Step 5: Based on the EEG risk feature vector, and combined with the pre-set individual EEG risk benchmark database, analyze to generate a second type of risk assessment result that characterizes the level of work safety risk perceived by the helmet wearer.

[0175] Step 6: Input the results of the first type of risk assessment and the results of the second type of risk assessment into the preset fusion algorithm to calculate the comprehensive risk value that represents the overall risk level of the operation.

[0176] Step 7: Compare the comprehensive risk value with the preset threshold. If the coefficient is higher than the threshold, an early warning message will be automatically generated and sent to the management terminal through the preset transmission mechanism.

[0177] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof.

[0178] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present invention should be included within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for intelligent perception and early warning of workplace safety risks based on electroencephalogram (EEG) data, applied to a smart safety helmet, wherein the smart safety helmet includes a helmet body and a camera disposed on the helmet body, characterized in that, include: Step 1: Acquire real-time environmental image data from the camera; Step 2: Call the pre-trained visual recognition model to recognize the on-site environmental image data and generate image recognition results; Step 3: Match the image recognition results with a pre-defined accident hazard category library to generate a first-class risk assessment result that characterizes the level of operational safety risks; Step 3 specifically includes: Step 3.1: Extract multi-dimensional feature information related to safety hazards from the image recognition results; Step 3.2: Match the multi-dimensional feature information with the accident hazard category library to obtain the first matching result. The first matching result includes the matched hazard category and its corresponding basic risk value. The accident hazard category library includes multiple hazard categories and risk weight coefficients corresponding to each category. Step 3.3: Based on the type of the current work scenario, the basic risk value is adjusted using a scenario adaptation correction algorithm to generate the first type of risk assessment result; Step 3.2 specifically includes: Step 3.2.1: Input multi-dimensional feature information into the pre-built AI hazard classification model to determine the target hazard category; Step 3.2.2: Based on the target hazard category, search for and obtain the corresponding basic risk value from the accident hazard category library to form the first matching result; Step 4: Obtain the EEG data of the helmet wearer; Step 5: Extract features from the EEG data to generate an EEG risk feature vector; Step 6: Analyze and process the EEG risk feature vector to generate a second type of risk assessment result that characterizes the wearer's perceived level of job safety. Step 6 specifically includes: Step 6.1: Compare the EEG risk feature vector with the preset individual EEG risk benchmark library to obtain a second matching result that characterizes the degree of preliminary matching between the vector and each risk perception level interval in the library. The individual EEG risk benchmark library is constructed based on the wearer's historical EEG data and corresponding work scenario risk records, and includes EEG feature vector intervals and weights corresponding to different risk perception levels. Step 6.2: Calculate the similarity measure between the EEG risk feature vector and the benchmark vector corresponding to each risk perception level in the individual EEG risk benchmark database. The similarity measure includes Euclidean distance and / or feature overlap. Step 6.3: Calculate the perceptual matching value based on the similarity metric and the complexity coefficient of the current task scenario; Step 6.4: Determine the wearer's risk perception level based on the preset perception interval where the perception matching value is located, and generate a second type of risk assessment result, wherein the perception interval includes a low perception interval, a medium perception interval, and a high perception interval. Step 7: Calculate the comprehensive risk value based on the results of the first type of risk assessment and the results of the second type of risk assessment; Step 8: When the overall risk value exceeds the preset threshold, generate an early warning message and send it to the management terminal; Prior to step 6.3, the method further includes: Obtain multi-dimensional environmental feature parameters of the current work scenario, wherein the multi-dimensional environmental feature parameters include equipment spatial distribution density, personnel movement and interaction frequency, and environmental electromagnetic interference intensity; Based on the preset environmental feature weights, the environmental feature parameters of each dimension are normalized to obtain the normalized parameter values ​​of each dimension. The complexity coefficient is obtained by weighted summation of the normalized parameter values ​​for each dimension.

2. The method according to claim 1, characterized in that, Step 3.2.1 specifically includes: Step 3.2.1.1: Perform fusion processing on the multi-dimensional feature information to generate a multimodal feature vector; Step 3.2.1.2: Calculate the similarity score between the multimodal feature vector and the feature template vector of each hazard category in the accident hazard category library; Step 3.2.1.3: Determine the hazard category with the highest matching degree as the target hazard category based on the similarity score.

3. The method according to claim 2, characterized in that, The expression for the similarity score is: in, It is the first of the multimodal eigenvectors Each dimension of feature value, It is the number of feature dimensions of the multimodal feature vector. It is the first feature template vector of a certain hazard category in the accident hazard category library. Each dimension of feature value.

4. The method according to claim 3, characterized in that, Step 7 specifically includes: Step 7.1: Normalize the results of the first type of risk assessment and the second type of risk assessment, mapping them to a unified scaling range; Step 7.2: Using a weighted fusion algorithm, the normalized first-type risk assessment results and the second-type risk assessment results are fused and calculated to output a comprehensive risk value.

Images