A safety education effectiveness audit method and system based on artificial intelligence

Abstract

The present invention discloses an artificial intelligence-based safety education effectiveness auditing method and system, wherein an artificial intelligence-based safety education effectiveness auditing method comprises: collecting workers' identity information, historical attendance data and real-time entry and exit records, and storing them in a database in a cloud backend; when workers enter a work area, performing face recognition on them; confirming the identity of the workers and recording attendance information based on the face recognition results; analyzing the workers' entry and exit events and whereabouts according to the database to determine whether the workers leave the site after attendance or enter the site without receiving safety education, and if there is an abnormality, generating an alarm or prompt information; the present invention can realize non-sensing attendance and in-depth analysis of workers' entry and exit events and whereabouts, timely discover and prevent potential safety risks in the work area, and improve the safety and stability of the working environment.

Description

Technical Field

[0001] The present invention relates to the technical field of effectiveness auditing of worker safety education, and in particular to an artificial intelligence-based safety education effectiveness auditing method and system. Background Art

[0002] With the rapid development of artificial intelligence and cloud computing technologies, safety education management has become increasingly important in the industrial and construction industries. Traditional methods often rely on manual inspections and manual records, which are inefficient, error-prone, and difficult to monitor in real time. In addition, workers who leave the site after attendance or enter the site without receiving safety education are difficult to detect and correct in time, which increases safety risks.

[0003] Nowadays, many companies and construction sites use attendance systems based on facial recognition to manage employee access and attendance. These systems capture employee facial information through cameras and compare it with information stored in a database to confirm employee identity and attendance. In addition, the application of cloud computing technology makes the storage and processing of large-scale data more convenient and efficient, which provides technical support for real-time monitoring and analysis of employee activities.

[0004] However, existing attendance systems based on facial recognition have some limitations. First, although these systems can accurately record employees' entry and exit times, they are often unable to directly assess whether employees have received the necessary safety education. Second, traditional attendance systems are usually unable to comprehensively monitor and analyze employees' real-time behavior in the work area, which may result in safety hazards not being discovered and handled in a timely manner. Summary of the invention

[0005] The purpose of the present invention is to provide a safety education effectiveness audit method and system based on artificial intelligence to solve the problems that existing methods are inefficient, error-prone, and difficult to directly evaluate whether employees have received necessary safety education.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions, including: collecting workers' identity information, historical attendance data and real-time entry and exit records, and storing them in a database in the cloud background; performing facial recognition on workers when they enter the work area; confirming the worker's identity based on the facial recognition result and recording the attendance information; analyzing the workers' entry and exit events and whereabouts according to the database to determine whether the workers leave the site after attendance or enter the site without receiving safety education, and if there is any abnormality, generating an alarm or prompt information.

[0007] As a preferred solution of the artificial intelligence-based safety education effectiveness audit method described in the present invention, the face recognition includes: preprocessing the input face image and dividing it into multiple local areas; extracting local area features; performing L2 norm normalization on the local area features; inputting the feature vector obtained after normalization into the face recognition model, and optimizing the face recognition model through the loss function.

[0008] As a preferred solution of the artificial intelligence-based safety education effectiveness audit method described in the present invention, wherein: extracting local area features includes: each local area is modeled as a graph structure, the graph structure includes multiple nodes and multiple edges, and the feature points or pixels in each local area are modeled as a feature descriptor, wherein each feature descriptor is regarded as a node, and the node has its own feature vector representation, and the connection relationship of the edge is determined according to the spatial relationship or feature similarity between the feature points in the local area; a multi-layer graph attention network is applied to the graph of each local area to obtain a first feature representation; a label is assigned to each node respectively, and the label represents the category of the feature; the dependency relationship between each node is defined; the first feature representation, the label and the dependency relationship between each node are learned by maximizing the conditional probability to generate the final feature representation of each local area.

[0009] As a preferred solution of the artificial intelligence-based safety education effectiveness audit method of the present invention, the loss function includes:

[0010]

[0011] In the formula, is the loss function value, N is the total number of samples, θy,i is the angle of the ith sample of the yth class, θ j，i is the angle of the i-th sample of the j-th class, s is the scaling parameter, m is the margin parameter, μ is the weight parameter, and n is the number of samples in each domain. is the kth sample in the source domain s, is the lth sample in the target domain t, and φ is the feature mapping function.

[0012] As a preferred solution of the artificial intelligence-based safety education effectiveness audit method described in the present invention, the analysis of workers' entry and exit events and movement trajectories includes: preprocessing the time information and spatial information of workers' entry and exit records and movement trajectories; setting distance thresholds and time windows; representing the trajectory data of each worker as a trajectory object; running the TraCluster algorithm on the preprocessed data set to analyze the spatial distribution and time distribution of each cluster to determine whether the workers have left the site after attendance or entered the site without receiving safety education.

[0013] As a preferred solution of the artificial intelligence-based safety education effectiveness audit system described in the present invention, it is characterized in that it includes: a data acquisition module, which is configured to collect workers' identity information, historical attendance data and real-time entry and exit records, and store them in a database in the cloud background; a face recognition module, which is configured to perform face recognition on workers when they enter the work area; a contactless attendance module, which is configured to confirm the identity of the workers according to the face recognition results and record the attendance information; an analysis module, which is configured to analyze the entry and exit events and whereabouts of the workers according to the database, and determine whether the workers leave the site after attendance or enter the site without receiving safety education; an early warning module, which is configured to generate an alarm or prompt information if it is determined that the worker's behavior is abnormal.

[0014] As a preferred solution of the artificial intelligence-based safety education effectiveness audit system described in the present invention, it is characterized in that the face recognition module is specifically configured to perform: preprocessing the input face image and dividing it into multiple local areas; extracting local area features; performing L2 norm normalization on the local area features; inputting the feature vector obtained after normalization into the face recognition model, and optimizing the face recognition model through the loss function.

[0015] As a preferred solution of the artificial intelligence-based safety education effectiveness audit system described in the present invention, the face recognition module is specifically configured to perform: extracting local area features includes: each local area is modeled as a graph structure, the graph structure includes multiple nodes and multiple edges, the feature points or pixels in each local area are modeled as a feature descriptor, wherein each feature descriptor is a node, has its own feature vector representation, and the connection relationship of the edges is determined according to the spatial relationship or feature similarity between the feature points in the local area; a multi-layer graph attention network is applied to the graph of each local area to obtain a first feature representation; a label is assigned to each node respectively, and the label represents the category of the feature; the dependency relationship between each node is defined; the first feature representation, the label and the dependency relationship between each node are learned by maximizing the conditional probability to generate the final feature representation of each local area.

[0016] As a preferred solution of the artificial intelligence-based safety education effectiveness audit system of the present invention, the face recognition module is specifically configured to execute the construction of the loss function:

[0017]

[0018] In the formula, is the loss function value, N is the total number of samples, θy,i is the angle of the ith sample of the yth class, θ j，iis the angle of the i-th sample of the j-th class, s is the scaling parameter, m is the margin parameter, μ is the weight parameter, and n is the number of samples in each domain. is the kth sample in the source domain s, is the first sample in the target domain t, and φ is the feature mapping function.

[0019] As a preferred solution of the artificial intelligence-based safety education effectiveness audit system described in the present invention, the analysis module is specifically configured to perform: preprocessing the time information and spatial information of the workers' entry and exit records and movement trajectories; setting distance thresholds and time windows; representing the trajectory data of each worker as a trajectory object; running the TraCluster algorithm on the preprocessed data set to analyze the spatial distribution and time distribution of each clustering cluster to determine whether the workers have left the site after attendance or entered the site without receiving safety education.

[0020] Beneficial effects of the present invention: The present invention integrates workers' identity information, historical attendance data and real-time entry and exit records, and combines the database storage and analysis capabilities of the cloud background to achieve seamless attendance and in-depth analysis of workers' entry and exit events and whereabouts, determine whether workers leave the site after attendance or enter the site without receiving safety education, and generate alarms or prompts in time, thereby improving workers' safety awareness, being able to timely discover and prevent potential safety risks in the work area, and improving the safety and stability of the working environment. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for describing the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative labor. Among them:

[0022] Figure 1 A schematic diagram of a flow chart of a safety education effectiveness audit method based on artificial intelligence according to the first embodiment of the present invention;

[0023] Figure 2 The figure is a schematic diagram of the face recognition process according to the first embodiment of the present invention. DETAILED DESCRIPTION

[0024] In order to make the above-mentioned purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are described in detail below in conjunction with the drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, but not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary persons in the art without creative work should fall within the scope of protection of the present invention.

[0025] In the following description, many specific details are set forth to facilitate a full understanding of the present invention, but the present invention may also be implemented in other ways different from those described herein, and those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0026] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The term "in one embodiment" that appears in different places in this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive with other embodiments.

[0027] The present invention is described in detail with reference to schematic diagrams. When describing the embodiments of the present invention, for the sake of convenience, the cross-sectional diagrams showing the device structure will not be partially enlarged according to the general scale, and the schematic diagrams are only examples, which should not limit the scope of protection of the present invention. In addition, in actual production, the three-dimensional dimensions of length, width and depth should be included.

[0028] At the same time, in the description of the present invention, it should be noted that the directions or positional relationships indicated by the terms "upper, lower, inner and outer" are based on the directions or positional relationships shown in the drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore cannot be understood as limiting the present invention. In addition, the terms "first, second or third" are only used for descriptive purposes and cannot be understood as indicating or implying relative importance.

[0029] In the present invention, unless otherwise clearly specified and limited, the terms "install, connect, connect" should be understood in a broad sense, for example: it can be a fixed connection, a detachable connection or an integral connection; it can also be a mechanical connection, an electrical connection or a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0030] Example 1

[0031] Reference Figure 1-2 , which is the first embodiment of the present invention, and provides a safety education effectiveness audit method based on artificial intelligence, comprising:

[0032] S1: Collect workers’ identity information, historical attendance data, and real-time entry and exit records, and store them in a database in the cloud backend.

[0033] All workers’ real-name information, historical attendance data, and real-time entry and exit records are stored in the cloud-based backend database to ensure data integrity and accessibility.

[0034] S2: When workers enter the work area, facial recognition is performed on them.

[0035] Install a high-definition camera at the entrance of the work area. When a worker enters the work area through the security door, the camera automatically starts when the worker approaches the door, clearly capturing the worker's facial features. The AI ​​edge box analyzes the camera data, performs facial recognition on the worker, and matches it with the real-name information database. Once the match is successful, the system records the worker's entry time and location, and displays it on the data screen.

[0036] Specifically, refer to Figure 2 , the steps of face recognition include:

[0037] (1) Preprocess the input face image and divide it into multiple local areas, such as eyes, nose, mouth, ears, etc.

[0038] (2) Extract local area features;

[0039] Although traditional convolutional neural networks perform well in processing regular structures in image data, they may have limitations in processing complex structures and interdependencies in local areas. By combining graph attention networks and conditional random fields, the present invention can better capture the spatial relationships and contextual information in local areas, thereby improving the accuracy and representation ability of local features. Specifically:

[0040] ① Each local area is modeled as a graph structure, which includes multiple nodes and multiple edges. The feature points or pixels in each local area are modeled as a feature descriptor, where each feature descriptor is regarded as a node with its own feature vector representation, and the connection relationship of the edges is determined according to the spatial relationship or feature similarity between the feature points in the local area.

[0041] ② Apply a multi-layer graph attention network to the graph of each local area to obtain the first feature representation; through the multi-layer graph attention network, learn the relationship and importance between different nodes, and dynamically adjust the weight of each node so that important feature points can more effectively influence the final feature representation.

[0042] ③ Assign labels to each node, where the labels represent the categories of the features.

[0043] ④ Define the dependency relationship P(v 1 , ..., v q ):

[0044]

[0045] In the formula, v q is the feature vector of the qth local region, v p is the feature vector of the pth local region, v r is the eigenvector of the rth local region, Z is the normalization factor, T is the transposed sign, and W 2 and W 3 is the parameter matrix of CRF.

[0046] ⑤ By maximizing the conditional probability, the dependency between the first feature representation, label and each node is learned to further improve the consistency and accuracy of local features and generate the final feature representation of each local area.

[0047] (3) L2 norm normalization of local area features;

[0048] (4) The normalized feature vector is input into the face recognition model, and the face recognition model is optimized and trained through the loss function.

[0049] Specifically, the face recognition model is constructed through the following steps:

[0050] Create a basic model that includes an input layer and two fully connected layers (Dense layers).

[0051] The specific steps are as follows:

[0052] Create a sequence model.

[0053] Add a Flatten layer to flatten the input 2D image data into a 1D vector.

[0054] Add a Dense layer with 64 neurons and ReLU activation function.

[0055] Add an output layer with the number of neurons equal to the number of categories and use the softmax activation function.

[0056] The model is compiled using the adam optimizer, and the loss function is The evaluation indicator is accuracy. When training the model, set the number of training rounds to 200.

[0057] Furthermore, a new sequence model is created based on the basic model.

[0058] Add a Flatten layer to flatten the input 2D image data into a 1D vector.

[0059] Add a Dense layer with 256 neurons and a ReLU activation function.

[0060] Add a Dense layer with 128 neurons and a ReLU activation function.

[0061] Add a Dense layer with 64 neurons and ReLU activation function.

[0062] Add an output layer with the number of neurons equal to the number of categories and use the softmax activation function.

[0063] Similarly, the model compilation uses the adam optimizer, and the loss function is The evaluation indicator is accuracy. When training the model, set the number of training rounds to 100.

[0064] Considering ArcFace loss and domain classification loss at the same time, the loss function is defined as:

[0065]

[0066] In the formula, is the loss function value, N is the total number of samples, θy,i is the angle of the ith sample of the yth class, θ j，i is the angle of the i-th sample of the j-th class, s is the scaling parameter, m is the margin parameter, which is used to adjust the classification interval, μ is the weight parameter, and n is the number of samples in each domain. is the kth sample in the source domain s, is the first sample in the target domain t, and φ is the feature mapping function.

[0067] The normalized feature vector is input into the face recognition model, and the domain label prediction of the feature is obtained through the face recognition model to obtain the face recognition result.

[0068] S3: Confirm the worker’s identity based on the face recognition results and record attendance information.

[0069] The facial recognition results are compared with the information stored in the database in the cloud background to confirm whether the worker’s identity is correct. Once the worker’s identity is confirmed, the attendance information is recorded immediately, including entry time and work location.

[0070] S4: Analyze the workers' entry and exit events and movement trajectories according to the database to determine whether the workers leave the site after attendance or enter the site without receiving safety education. If there is any abnormality, an alarm or prompt message is generated.

[0071] Analyze the workers' entry and exit events and whereabouts according to the database. The specific steps are as follows:

[0072] (1) Preprocessing the time and space information of workers’ entry and exit records and movement trajectories, including removing abnormal data (excluding abnormal data points in time or space, such as records outside the normal working area or non-working hours) and filling missing values ​​(handling any missing time or space data, interpolation methods or simple filling methods can be used to handle missing values).

[0073] (2) Set the distance threshold and time window.

[0074] Set the time window to 15 minutes and the distance threshold to 100 meters.

[0075] (3) Represent the trajectory data of each worker as a trajectory object, including time and space dimensions.

[0076] For example, a group of workers’ entry and exit events and movement trajectory data are collected, including timestamps and geographic location information. The specific data is as follows:

[0077] Table 1: Workers’ entry and exit events and movement trajectory data

[0078] Worker ID Timestamp longitude latitude 1 2023-05-01 08:00:00 120.101 30.241 2 2023-05-01 08:05:00 120.105 30.245 3 2023-05-01 12:00:00 120.102 30.240

[0079] The worker’s trajectory data is represented as [(120.101,30.241,'2023-05-01 08:00:00'),(120.105,30.245,'2023-05-01 08:05:00'),(120.102,30.240,'2023-05-01 12:00:00')].

[0080] (4) Run the TraCluster algorithm on the preprocessed data set to analyze the spatial and temporal distribution of each cluster and determine whether workers leave the site immediately after attendance or enter the site without receiving safety education.

[0081] Based on the set distance threshold and time window, the algorithm divides the workers’ trajectory data into three clusters, as shown below:

[0082] Cluster 1: Worker 1’s work area, including entry and exit events and work trajectories in the morning.

[0083] Cluster 2: Worker 2’s work area, including the morning work trajectory.

[0084] Cluster 3: Worker 3’s work area, which may contain entry and exit events in the afternoon.

[0085] The spatial and temporal distribution of each cluster was analyzed to determine whether there were dense work areas or specific worker behavior patterns.

[0086] Check the clustering results for anomalies, such as smaller clusters or trajectory patterns that do not match normal work patterns, which may indicate unusual worker behavior, such as unreasonable work hours or changes in work location.

[0087] This will determine whether the team has engaged in wage fraud by asking workers to leave immediately after clocking in, or whether workers have entered the site without receiving safety education, and generate warnings or prompts in a timely manner. If there are any abnormalities in the workers' entry and exit events and whereabouts, warnings or prompts will be generated, and on-site managers will intervene and handle them in a timely manner based on the warning information to ensure the effectiveness of safety education and attendance.

[0088] Example 2

[0089] This embodiment is different from the first embodiment in that it provides a safety education effectiveness audit system based on artificial intelligence, including:

[0090] A data collection module is configured to collect workers' identity information, historical attendance data, and real-time entry and exit records, and store them in a database in the cloud background;

[0091] A face recognition module is configured to perform face recognition on a worker when the worker enters a work area;

[0092] The non-contact attendance module is configured to confirm the identity of the worker based on the face recognition result and record the attendance information;

[0093] An analysis module is configured to analyze workers' entry and exit events and movement trajectories according to a database to determine whether workers leave the site immediately after attendance or enter the site without receiving safety education;

[0094] The early warning module is configured to generate an alarm or prompt information if it is determined that the worker's behavior is abnormal.

[0095] The face recognition module is specifically configured to perform:

[0096] Install a high-definition camera at the entrance of the work area. When a worker enters the work area through the security door, the camera automatically starts when the worker approaches the door, clearly capturing the worker's facial features. The AI ​​edge box analyzes the camera data, performs facial recognition on the worker, and matches it with the real-name information database. Once the match is successful, the system records the worker's entry time and location, and displays it on the data screen.

[0097] Specifically, the steps of face recognition include:

[0098] (1) Preprocess the input face image and divide it into multiple local areas, such as eyes, nose, mouth, ears, etc.

[0099] (2) Extract local area features;

[0100] Although traditional convolutional neural networks perform well in processing regular structures in image data, they may have limitations in processing complex structures and interdependencies in local areas. By combining graph attention networks and conditional random fields, the present invention can better capture the spatial relationships and contextual information in local areas, thereby improving the accuracy and representation ability of local features. Specifically:

[0101] ① Each local area is modeled as a graph structure, which includes multiple nodes and multiple edges. The feature points or pixels in each local area are modeled as a feature descriptor, where each feature descriptor is regarded as a node with its own feature vector representation, and the connection relationship of the edges is determined according to the spatial relationship or feature similarity between the feature points in the local area.

[0102] ② Apply a multi-layer graph attention network to the graph of each local area to obtain the first feature representation; through the multi-layer graph attention network, learn the relationship and importance between different nodes, and dynamically adjust the weight of each node so that important feature points can more effectively influence the final feature representation.

[0103] ③ Assign labels to each node, where the labels represent the categories of the features.

[0104] ④ Define the dependency relationship P(v 1 , ..., v q ):

[0105]

[0106] In the formula, v q is the feature vector of the qth local region, v p is the feature vector of the pth local region, v r is the eigenvector of the rth local region, Z is the normalization factor, T is the transposed sign, and W 2 and W 3 is the parameter matrix of CRF.

[0107] ⑤ By maximizing the conditional probability, the dependency between the first feature representation, label and each node is learned to further improve the consistency and accuracy of local features and generate the final feature representation of each local area.

[0108] (3) L2 norm normalization of local area features;

[0109] (4) The normalized feature vector is input into the face recognition model, and the face recognition model is optimized and trained through the loss function.

[0110] Specifically, the face recognition model is constructed through the following steps:

[0111] Create a basic model that includes an input layer and two fully connected layers (Dense layers).

[0112] The specific steps are as follows:

[0113] Create a sequence model.

[0114] Add a Flatten layer to flatten the input 2D image data into a 1D vector.

[0115] Add a Dense layer with 64 neurons and ReLU activation function.

[0116] Add an output layer with the number of neurons equal to the number of categories and use the softmax activation function.

[0117] The model is compiled using the adam optimizer, and the loss function is The evaluation indicator is accuracy. When training the model, set the number of training rounds to 200.

[0118] Furthermore, a new sequence model is created based on the basic model.

[0119] Add a Flatten layer to flatten the input 2D image data into a 1D vector.

[0120] Add a Dense layer with 256 neurons and a ReLU activation function.

[0121] Add a Dense layer with 128 neurons and a ReLU activation function.

[0122] Add a Dense layer with 64 neurons and ReLU activation function.

[0123] Add an output layer with the number of neurons equal to the number of categories and use the softmax activation function.

[0124] Similarly, the model compilation uses the adam optimizer, and the loss function is The evaluation indicator is accuracy. When training the model, set the number of training rounds to 100.

[0125] Considering ArcFace loss and domain classification loss at the same time, the loss function is defined as:

[0126]

[0127] In the formula, is the loss function value, N is the total number of samples, θy,i is the angle of the ith sample of the yth class, θ j，i is the angle of the i-th sample of the j-th class, s is the scaling parameter, m is the margin parameter, which is used to adjust the classification interval, μ is the weight parameter, and n is the number of samples in each domain. is the kth sample in the source domain s, is the lth sample in the target domain t, and φ is the feature mapping function.

[0128] The normalized feature vector is input into the face recognition model, and the domain label prediction of the feature is obtained through the face recognition model to obtain the face recognition result.

[0129] The analysis module is specifically configured to perform:

[0130] (1) Preprocessing the time and space information of workers’ entry and exit records and movement trajectories, including removing abnormal data (excluding abnormal data points in time or space, such as records outside the normal working area or non-working hours) and filling missing values ​​(handling any missing time or space data, interpolation methods or simple filling methods can be used to handle missing values).

[0131] (2) Set the distance threshold and time window.

[0132] Set the time window to 15 minutes and the distance threshold to 100 meters.

[0133] (3) Represent the trajectory data of each worker as a trajectory object, including time and space dimensions.

[0134] (4) Run the TraCluster algorithm on the preprocessed data set to analyze the spatial and temporal distribution of each cluster and determine whether workers leave the site immediately after attendance or enter the site without receiving safety education.

[0135] It should be appreciated that embodiments of the present invention may be implemented or enforced by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable memory. The method may be implemented in a computer program using standard programming techniques - including a non-transitory computer-readable storage medium configured with a computer program, wherein the storage medium so configured causes the computer to operate in a specific and predefined manner - according to the methods and drawings described in the specific embodiments. Each program may be implemented in a high-level procedural or object-oriented programming language to communicate with a computer system. However, if desired, the program may be implemented in an assembly or machine language. In any case, the language may be a compiled or interpreted language. In addition, the program may be run on a programmed ASIC for this purpose.

[0136] Furthermore, the operations of the processes described herein may be performed in any suitable order unless otherwise indicated herein or otherwise clearly contradicted by context. The processes described herein (or variations and / or combinations thereof) may be performed under the control of one or more computer systems configured with executable instructions, and may be implemented as code (e.g., executable instructions, one or more computer programs, or one or more applications) that is executed collectively on one or more processors, by hardware, or a combination thereof. The computer program includes a plurality of instructions that may be executed by one or more processors.

[0137] Further, the method can be implemented in any type of computing platform that is operably connected to a suitable computer, including but not limited to a personal computer, a minicomputer, a mainframe, a workstation, a network or distributed computing environment, a separate or integrated computer platform, or in communication with a charged particle tool or other imaging device, etc. Aspects of the present invention can be implemented in machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, an optical read and / or write storage medium, a RAM, a ROM, etc., so that it can be read by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the process described herein. In addition, the machine-readable code, or part thereof, can be transmitted via a wired or wireless network. When such media includes instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor, the invention described herein includes these and other different types of non-transitory computer-readable storage media. When programmed according to the methods and techniques described in the present invention, the present invention also includes the computer itself. The computer program can be applied to input data to perform the functions described herein, thereby converting the input data to generate output data stored in a non-volatile memory. The output information can also be applied to one or more output devices such as a display. In a preferred embodiment of the present invention, the converted data represents a physical and tangible object, including a specific visual depiction of the physical and tangible object produced on a display.

[0138] As used in this application, the terms "component", "module", "system", etc. are intended to refer to a computer-related entity, which can be hardware, firmware, a combination of hardware and software, software, or software in operation. For example, a component can be, but is not limited to: a process running on a processor, a processor, an object, an executable file, a thread in execution, a program, and / or a computer. As an example, an application running on a computing device and the computing device can both be components. One or more components can exist in a process and / or thread in execution, and a component can be located in a computer and / or distributed between two or more computers. In addition, these components can be executed from various computer-readable media having various data structures thereon. These components can communicate in a local and / or remote process manner, such as according to a signal having one or more data packets (e.g., data from a component that interacts with another component in a local system, a distributed system, and / or interacts with other systems in a signal manner through a network such as the Internet).

[0139] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. A safety education effectiveness audit method based on artificial intelligence, characterized in that: include: Collect workers’ identity information, historical attendance data, and real-time entry and exit records, and store them in a cloud-based backend database; When workers enter the work area, they are identified by facial recognition; Confirm worker identities and record attendance information based on facial recognition results; Analyze workers' entry and exit events and whereabouts according to the database to determine whether workers leave the site immediately after attendance or enter the site without receiving safety education. If there is any abnormality, generate an alarm or prompt information; Wherein, the face recognition includes: Preprocess the input face image and divide it into multiple local areas; Extract local area features; Perform L2 norm normalization on local area features; The normalized feature vector is input into the face recognition model, and the face recognition model is optimized and trained through the loss function; Extracting local area features includes: Each local region is modeled as a graph structure, which includes multiple nodes and multiple edges. The feature points or pixels in each local region are modeled as a feature descriptor, wherein each feature descriptor is used as a node, and the node has its own feature vector representation. The connection relationship of the edge is determined according to the spatial relationship or feature similarity between the feature points in the local region. Apply a multi-layer graph attention network to the graph of each local region to obtain the first feature representation; Assign a label to each node, wherein the label represents the category of the feature; Define the dependencies between each node; The final feature representation of each local region is generated by learning the dependencies between the first feature representation, the label, and each node by maximizing the conditional probability.

2. The artificial intelligence-based safety education effectiveness audit method according to claim 1, characterized in that: The loss function includes: In the formula, is the loss function value, N is the total number of samples, θ y,i is the angle of the i-th sample of the y-th class, θ j,i is the angle of the i-th sample of the j-th class, s is the scaling parameter, m is the margin parameter, μ is the weight parameter, and n is the number of samples in each domain. is the kth sample in the source domain s, is the lth sample in the target domain t, and φ is the feature mapping function.

3. The artificial intelligence-based safety education effectiveness audit method according to claim 2, characterized in that: The analysis of workers' entry and exit events and movement trajectories includes: Pre-process the time and space information of workers’ entry and exit records and movement trajectories; Set distance thresholds and time windows; Represent each worker's trajectory data as a trajectory object; The TraCluster algorithm is run on the preprocessed data set to analyze the spatial and temporal distribution of each cluster and determine whether workers leave the site immediately after attendance or enter the site without receiving safety education.

4. An artificial intelligence-based safety education effectiveness audit system, characterized in that: include: A data collection module is configured to collect workers' identity information, historical attendance data, and real-time entry and exit records, and store them in a database in the cloud background; A face recognition module is configured to perform face recognition on a worker when the worker enters a work area; The non-contact attendance module is configured to confirm the identity of the worker based on the face recognition result and record the attendance information; An analysis module is configured to analyze workers' entry and exit events and movement trajectories according to a database to determine whether workers leave the site immediately after attendance or enter the site without receiving safety education; The early warning module is configured to generate an alarm or prompt information if it is determined that the worker's behavior is abnormal; The face recognition module is specifically configured to perform: Preprocess the input face image and divide it into multiple local areas; Extract local area features; Perform L2 norm normalization on local area features; The normalized feature vector is input into the face recognition model, and the face recognition model is optimized and trained through the loss function; The face recognition module is specifically configured to perform: Extracting local area features includes: Each local region is modeled as a graph structure, which includes multiple nodes and multiple edges. The feature points or pixels in each local region are modeled as a feature descriptor, where each feature descriptor is a node with its own feature vector representation, and the connection relationship of the edge is determined according to the spatial relationship or feature similarity between the feature points in the local region; Apply a multi-layer graph attention network to the graph of each local region to obtain the first feature representation; Assign a label to each node, wherein the label represents the category of the feature; Define the dependencies between each node; The final feature representation of each local region is generated by learning the dependencies between the first feature representation, the label, and each node by maximizing the conditional probability.

5. The artificial intelligence-based safety education effectiveness audit system according to claim 4, characterized in that: The face recognition module is specifically configured to execute the constructed loss function: In the formula, is the loss function value, N is the total number of samples, θ y,i is the angle of the i-th sample of the y-th class, θ j,i is the angle of the i-th sample of the j-th class, s is the scaling parameter, m is the margin parameter, μ is the weight parameter, and n is the number of samples in each domain. is the kth sample in the source domain s, is the lth sample in the target domain t, and φ is the feature mapping function.

6. The artificial intelligence-based safety education effectiveness audit system according to claim 5, characterized in that: The analysis module is specifically configured to perform: Pre-process the time and space information of workers’ entry and exit records and movement trajectories; Set distance thresholds and time windows; Represent each worker's trajectory data as a trajectory object; The TraCluster algorithm is run on the preprocessed data set to analyze the spatial and temporal distribution of each cluster and determine whether workers leave the site immediately after attendance or enter the site without receiving safety education.

Images