Nuclear power personnel safety intelligent management and control system and method
By building an intelligent control system at the nuclear power plant site, and combining deep learning and multi-task detection algorithms, the problems of low efficiency and high misjudgment rate in nuclear power plant foreign object prevention management under traditional manual supervision have been solved, and efficient and accurate real-time detection and management of nuclear power plant personnel and materials have been achieved.
Patent Information
- Application Number
- CN202511929312.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-12-19
- Publication Date
- 2026-04-21
AI Technical Summary
Traditional methods for managing foreign objects in nuclear power plants and inspecting personal protective equipment rely on manual supervision. These methods are characterized by high workload, low efficiency, high misjudgment rate, and susceptibility to fatigue and errors. Furthermore, they cannot provide real-time confirmation and verification, resulting in coarse-grained management of foreign objects and difficulty in improving efficiency and accuracy in the complex nuclear power plant environment.
An intelligent management and control system for nuclear power plant personnel safety is constructed using image acquisition equipment, security inspection equipment, and servers. It combines facial recognition modules, foreign object detection modules, and personnel wearable detection modules, and utilizes deep learning neural networks and multi-task detection algorithms to achieve real-time identification and detection of personnel and objects. The detection accuracy is improved by fusing multi-frame information.
It has enabled intelligent safety management of nuclear power plant workers entering and exiting the KX control area, improved the efficiency and accuracy of foreign object prevention management, reduced the false detection rate, ensured the traceability of items entering and exiting and personnel entering and exiting, and achieved high-accuracy target detection tasks.
Smart Images
Figure CN121904685A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of nuclear power technology, specifically relating to an intelligent management and control system and method for the safety of nuclear power personnel. Background Technology
[0002] Foreign object (FOO) management and the inspection of personal protective equipment (PPE) are crucial aspects of nuclear power plant safety management. FOOs have a significant impact on the safe operation of nuclear power plants, and most safety hazards and accidents occurring in industrial and construction environments are related to whether workers are wearing the correct PPE.
[0003] Traditional methods for managing foreign object debris (FOD) and inspecting personal protective equipment (PPE) rely on manual supervision. FOD methods involve specialists manually counting and recording items and personnel entry and exit through paper documents. This method suffers from high workload, low efficiency, high error rate, and susceptibility to fatigue and mistakes. Furthermore, FOD registration results can only be used for post-event traceability and cannot be confirmed or verified in real time, resulting in coarse-grained FOD management. PPE management methods are also inefficient, susceptible to subjective factors, and prone to overlooking inspections of smaller safety protective equipment. Therefore, in the complex working environment of nuclear power plants, there is an urgent need to improve the efficiency and accuracy of FOD management. Summary of the Invention
[0004] To overcome the problems existing in related technologies, a nuclear power personnel safety intelligent management and control system and method are provided.
[0005] According to one aspect of the present disclosure, a nuclear power plant personnel safety intelligent management and control system is provided. The system includes: an image acquisition device, a security inspection device, and a server. The server integrates a face recognition module, a foreign object detection module, and a personnel wear detection module.
[0006] Image acquisition equipment is used to capture images of people and transmit them to a server;
[0007] Security screening equipment is used to inspect personal belongings carried by individuals.
[0008] The facial recognition module is used to perform facial recognition and identity verification on personnel based on the acquired images;
[0009] The foreign object detection module is used to identify and measure items and output an item list;
[0010] The personnel wear detection module is used to detect the key points of a person's body and whether the corresponding equipment is worn at the key points of the body based on the acquired images, thereby determining whether the person is wearing protective equipment correctly.
[0011] In one possible implementation, the personnel wearing detection module includes a first feature extraction unit, a first feature fusion unit, and a first prediction unit. The first feature extraction unit is used to extract features from the personnel image, and the first feature fusion unit is used to fuse the extracted features. The first prediction unit performs ARCSA convolution processing on the fused features. The ARCSA convolution processing combines spatial and channel attention mechanisms and uses the receptive field to adaptively and dynamically adjust the convolution kernel size. The first prediction unit also includes a human keypoint regression submodule and a target detection submodule. The regression submodule is used to predict and determine the location of the personnel's human keypoints based on the results of the convolution processing, and the target detection submodule is used to determine whether the personnel are wearing corresponding protective equipment at the location of their human keypoints.
[0012] In one possible implementation, ARCSA convolution processing includes:
[0013] Step 11: Use a basic convolutional layer (which can be a 3x3 convolution) to perform preliminary feature extraction on the input original feature map to obtain the feature map F_base;
[0014] Step 12: Perform convolution operations on F_base at multiple different scales to obtain feature maps at multiple different scales. Then, perform global average pooling on the feature maps at each scale to obtain the corresponding channel statistics. Input the statistics of each channel into the fully connected layer and obtain the weights after softmax normalization. Multiply the feature map at each scale with the corresponding weights and sum them to obtain the feature map F_rf that integrates multi-scale information.
[0015] Step 13: After performing global average pooling on F_rf, the first output is obtained through a multilayer perceptron (MLP). After performing max pooling on F_rf, the second output is obtained through the MLP. The first output and the second output are added together and then input into the sigmoid activation function to obtain the channel attention weight Mc. Mc is multiplied by F_rf channel by channel to obtain the channel attention-weighted feature map F_ch.
[0016] Step 14: Perform average pooling along the channel dimension of F_ch to obtain the first feature map, and perform max pooling along the channel dimension of F_ch to obtain the second feature map. Concatenate the first and second feature maps along the channel dimension, then reduce the number of channels to 1 through a convolutional layer, and then pass the sigmoid activation function to obtain the spatial attention weight Ms. Multiply Ms with F_ch point by point to obtain the spatial attention weighted feature map F_sp.
[0017] Step 15: Perform a residual connection between F_sp and the original feature map or F_base to obtain the final output.
[0018] In one possible implementation, the first prediction unit also employs a continuous multi-frame information fusion judgment mechanism: taking continuous multi-frame images as a detection cycle, the distance between the key points of the human body in each frame image and the center point of the detection bounding box of the protective equipment is calculated. If the distance is less than or equal to a set threshold, the type of the equipment is determined to be correctly detected; if the distance is greater than the set threshold, the type of the equipment is determined to be incorrectly detected. The final detection result is output by combining the judgment results of multiple frames within a detection cycle.
[0019] In one possible implementation, the foreign object detection module includes a second feature extraction unit, a second feature fusion unit, and a second prediction unit;
[0020] The second feature extraction unit is a Darknet network based on the lightweight network structure C2f module, containing multiple convolutional modules, used to extract features of different scales from the input image layer by layer and output feature maps of different scales; the second feature fusion unit is a multi-scale feature aggregation network, which fuses feature maps of different scales by combining a feature pyramid network and a path aggregation structure; the second prediction unit uses a decoupled head structure, including a category prediction branch and a location prediction branch, to generate the category and location of the target item.
[0021] In one possible implementation, a low-level feature map P2 is retained in the second prediction unit to form a multi-level prediction scale from P2 to P5, with different prediction scales used to detect targets of different sizes.
[0022] In one possible implementation, the foreign object detection module uses a depth transform convolutional structure DConv, which performs a 3×3 convolution after concatenating features in the channel dimension through four parallel downsampling operations.
[0023] According to another aspect of the embodiments of this disclosure, a method for intelligent management and control of nuclear power plant personnel safety is provided. The method is based on the system described above and is used for personnel entry verification, including:
[0024] Step 21: Collect facial images of people waiting to enter the venue, and perform facial recognition and identity verification;
[0025] Step 22: After verifying the identity of the person, capture images of the personal belongings placed on the tray, and use the foreign object detection module to identify and measure the items;
[0026] Step 23: Use security screening equipment to check the items carried by the person. After confirming that the person has no items, simultaneously identify the obtained facial information.
[0027] Step 24: Generate an entry identification item form and send it to personnel for confirmation and administrator for approval;
[0028] Step 25: Collect video images of the personnel's appearance, use a personnel wearing detection module based on multi-frame fusion to identify the wearing status of protective equipment, and determine whether the personnel are wearing the protective equipment correctly based on the identification results;
[0029] Step 26: If the personnel are judged to be dressed correctly, generate a confirmation form and confirm and approve it; if the personnel are judged to be dressed incorrectly, issue an alarm and output the violation item.
[0030] In one possible implementation, the method is also used for personnel departure verification, including:
[0031] Step 31: Collect facial images of people waiting to leave and perform facial recognition and identity verification;
[0032] Step 32: After verifying the identity of the personnel, capture images of the personal items placed on the tray, use the small target consumable identification algorithm model to identify and measure the items, and compare the identified list of outgoing items with the corresponding list of incoming items.
[0033] Step 33: Use security screening equipment to check the items carried by the person. After confirming that the person has no items, synchronize the facial information.
[0034] Step 34: Generate an item comparison form, record and explain any discrepancies, and have it confirmed by personnel and reviewed by the administrator.
[0035] According to another aspect of the present disclosure, a non-volatile computer-readable storage medium is provided, on which computer program instructions are stored, which, when executed by a processor, implement the above-described method.
[0036] The beneficial effects of this disclosure are as follows: This disclosure extracts sequential multi-frame images from nuclear power plant site videos, combines them with human body key point detection technology, and constructs a multi-task personnel wear detection algorithm model based on multi-frame fusion. This achieves accurate and automatic detection of personnel appearance. After multiple images are input into the multi-task detection model, false detections with similar features are initially screened out based on key points and personnel wear detection results. Further screening is performed based on continuous sequential multi-frame information to remove missed detections and false detections. Finally, the detection results are output. This achieves a target detection task with high accuracy and low false detection rate. It enables intelligent safety management of nuclear power plant workers entering and exiting the KX control area. Attached Figure Description
[0037] Figure 1 This is a block diagram of a nuclear power plant personnel safety intelligent management and control system as shown in an embodiment of this disclosure.
[0038] Figure 2 This is a schematic diagram of key human body points in an application example of this disclosure.
[0039] Figure 3 This is a schematic diagram of a personnel wear detection module shown in an embodiment of this disclosure.
[0040] Figure 4 This is a schematic diagram of the detection results of the personnel wear detection module in an application example of this disclosure.
[0041] Figure 5 This is a schematic diagram of the foreign object detection module shown in an embodiment of this disclosure.
[0042] Figure 6 This is a schematic diagram of the multi-frame information fusion judgment mechanism of the first prediction unit shown in the embodiments of this disclosure. Detailed Implementation
[0043] The present disclosure will now be described in further detail with reference to the accompanying drawings and specific embodiments.
[0044] Unless otherwise defined, the technical and scientific terms used in this disclosure have the same meaning as commonly understood by one of ordinary skill in the art to which this disclosure pertains; the terminology used in this disclosure is for the purpose of describing particular embodiments only and is not intended to be limiting of this disclosure; the term "comprising" and any variations thereof in this disclosure are intended to cover non-exclusive inclusion. Clearly, the embodiments described in this disclosure are only a part of the embodiments of this disclosure, and not all of them. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of this disclosure without inventive effort are within the scope of protection of this disclosure.
[0045] In this disclosure, the reference to "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this disclosure. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.
[0046] Figure 1 This is a block diagram of a nuclear power plant personnel safety intelligent management and control system as shown in an embodiment of this disclosure, such as... Figure 1 As shown, the system includes: image acquisition equipment, security inspection equipment, and a server. The server integrates a face recognition module, a foreign object detection module, and a personnel wear detection module.
[0047] Image acquisition equipment is used to acquire images of people and transmit them to the server; security inspection equipment is used to inspect people's personal belongings; face recognition module is used to perform face recognition and identity verification on people based on the acquired images; foreign object detection module is used to identify and measure items and output an item list; personnel wearing detection module is used to detect the key points of people based on the acquired images and whether the key points are wearing the corresponding equipment, thereby determining whether the personnel are wearing protective equipment correctly.
[0048] This disclosure allows for the pre-construction of a dataset using 17 key points of the human body. The names and corresponding locations of each key point are as follows: Figure 2 As shown, the dataset consists of images of different people in various working postures at different construction sites. Image data annotation was performed using tools such as LabelImg to label the targets to be identified in the images. The annotations included the coordinates of the bounding boxes surrounding the targets and their categories. The labeled images and annotation files were divided into training and validation sets in an 8:2 ratio for model training and evaluation, respectively, thus constructing the personnel clothing detection module.
[0049] See Figure 3 The personnel wear detection module is based on a deep learning neural network method and adopts an end-to-end multi-task detection network structure. The personnel wear detection module includes a first feature extraction unit (Backbone1), a first feature fusion unit (PANet), and a first prediction unit (Heads). The first feature extraction unit is a feature extraction network that extracts feature information from the input original image. The first feature fusion unit is a feature fusion module that integrates the extracted feature information to obtain fused feature information. Based on the fused feature information output by the first feature fusion unit, the first feature extraction unit outputs the location of key points of the human body.
[0050] The first feature extraction unit, the first feature fusion unit, and the first prediction unit are used to address issues such as missed detections, false detections, and slow speeds in keypoint and personnel wear detection caused by lighting, background, and occlusion in construction site environments. The first prediction unit performs ARCSA convolution processing on the fused features. ARCSA convolution processing combines spatial attention and channel attention mechanisms, using a receptive field adaptive module to dynamically adjust the kernel size of the convolutional layer, thereby dynamically changing the receptive field. Simultaneously, the spatial attention and channel attention mechanisms extract spatial and channel feature information, respectively.
[0051] As an example of this embodiment, ARCSA convolution processing includes:
[0052] Step 11: Use a basic convolutional layer (which can be a 3x3 convolution) to perform preliminary feature extraction on the input original feature map to obtain the feature map F_base;
[0053] Step 12: Perform convolution operations at multiple different scales (e.g., 3x3, 5x5, 7x7) on F_base to obtain feature maps at multiple different scales (e.g., F_3x3, F_5x5, F_7x7). Then, perform global average pooling on the feature maps at each scale to obtain the corresponding channel statistics. Input the statistics of each channel into the fully connected layer and obtain the weights after softmax normalization. Multiply the feature map at each scale with the corresponding weights and sum them to obtain the feature map F_rf that integrates multi-scale information.
[0054] Step 13: After performing global average pooling on F_rf, the first output is obtained through a multilayer perceptron (MLP). After performing max pooling on F_rf, the second output is obtained through the MLP. The first output and the second output are added together and then input into the sigmoid activation function to obtain the channel attention weight Mc. Mc is multiplied by F_rf channel by channel to obtain the channel attention-weighted feature map F_ch.
[0055] Step 14: Perform average pooling along the channel dimension of F_ch to obtain the first feature map, and perform max pooling along the channel dimension of F_ch to obtain the second feature map. Concatenate the first and second feature maps along the channel dimension, then reduce the number of channels to 1 through a convolutional layer (usually a 7x7 convolution), and then pass the sigmoid activation function to obtain the spatial attention weight Ms. Multiply Ms by F_ch point by point to obtain the spatial attention-weighted feature map F_sp.
[0056] Step 15: Perform a residual concatenation between F_sp and the original feature map or F_base (whichever you choose according to your processing needs) to obtain the final output.
[0057] The first prediction unit also includes a key point regression submodule and an object detection submodule. The regression submodule is used to make predictions based on the results of convolution processing to determine the key point positions of personnel, and the object detection submodule is used to determine whether personnel are wearing corresponding protective equipment at their key point positions.
[0058] In the complex and ever-changing nuclear power production environment, typical multi-frame target detection algorithms rely heavily on single image features (such as color and shape). While they can effectively identify and track multiple targets, their recognition capabilities are limited when faced with targets with similar features, making false detections highly likely. When detecting gloves and shoes, both are white, have a small target scale at certain viewpoints, a low pixel count, and similar shapes, making it difficult for detection algorithms to accurately classify them.
[0059] The first prediction unit of this disclosure also determines the distance between the location of the human body key point and the center point of the bounding box obtained by the personnel wearing detection. If the distance between the location of the human body key point and the center point of the bounding box obtained by the personnel wearing detection is less than or equal to a set threshold (e.g., 5), the type detection of the equipment is determined to be correct (is_correct); otherwise, it is a false detection, as indicated by:
[0060]
[0061] In the formula, K = (k x ,k y ) represents the predicted center point coordinates of key human body points, B = (b x ,b y The coordinates of the center point of the bounding box of the personnel appearance detection result are represented by d(K,B), and d(K,B) is the Euclidean distance between the two points. During real-time video detection, three consecutive frames constitute one detection cycle, and the detection results for each frame are acquired separately. The multi-task network efficiently integrates the results of keypoint detection and appearance detection, considering not only the confidence level of individual task detection results but also the correlation between different task detection results, thus achieving more accurate and robust target recognition. In the integrated results, gloves and shoes, which were originally difficult to distinguish due to similar colors and small pixel size, can now be clearly distinguished, reducing the false detection rate. The detection results are as follows: Figure 4 As shown.
[0062] In one possible implementation, the foreign object detection module includes a second feature extraction unit, a second feature fusion unit, and a second prediction unit. The second feature extraction unit is a Darknet network based on the lightweight C2f network structure, containing multiple convolutional modules to extract features of different scales from the input image layer by layer and output feature maps of different scales. The second feature fusion unit is a multi-scale feature aggregation network that fuses feature maps of different scales by combining a feature pyramid network and a path aggregation structure. The second prediction unit uses a decoupled head structure, including a category prediction branch and a location prediction branch, to generate the category and location of the target object. The low-level feature map P2 is retained in the second prediction unit, forming a multi-level prediction scale from P2 to P5, with different prediction scales used to detect targets of different sizes. The foreign object detection module employs a depth transform convolutional structure DConv, which performs 3×3 convolutions after concatenating features in the channel dimension through four parallel downsampling passes.
[0063] In one possible implementation, the first prediction unit also employs a multi-frame information fusion judgment mechanism to construct a multi-task personnel wear detection algorithm model based on multi-frame fusion. This overcomes the limitations of single-frame detection, improves detection accuracy, and reduces missed detections. See [link to relevant documentation]. Figure 6The first prediction unit determines whether the detection results of multiple consecutive images are consistent. If the detection results are consistent, it outputs the detection result; otherwise, it re-detects the multiple consecutive images. This method, by continuously analyzing the target state changes in multiple frames, can more comprehensively capture the dynamic features of the target, thus more accurately determining the target's state. Even if the target is occluded or its features are blurred in some frames, it can still make a more accurate judgment by using the correlation information between consecutive frames.
[0064] By utilizing the target motion consistency algorithm, the prediction results of the multi-task personnel protective equipment detection algorithm model and the small target foreign object detection model based on multi-frame fusion are optimized, thereby improving the accuracy of the model in real-time detection in dynamically changing video data.
[0065] The core idea of the target motion consistency algorithm is that the position of a target should not change significantly across three consecutive frames. If the position of the detection box changes too much, it is determined that it is not the same target. The Euclidean distance between the same target in two frames can be expressed as:
[0066]
[0067] In the formula, the coordinates of the center point of the target bounding box detected in the current frame are B. T (x T ,y T The coordinates of the center point of the target bounding box detected in the previous frame are B. T-1 (x T-1 ,y T-1 The motion consistency threshold is Δ. If d(B) T B T-1 If the value of the target detected in the two frames is greater than or equal to Δ, then the two frames are considered to contain different targets. Furthermore, the detection results from the previous few frames are averaged and used as the bounding box coordinates for the target in three consecutive frames to reduce noise. The calculation method is as follows:
[0068]
[0069] In the formula, the coordinates of the center point of the detected target bounding box in the current frame and the two previous frames are B. T B T-1 B T-2 .
[0070] After four parallel downsampling operations, the feature layers are concatenated along the channel dimension and then convolutional with a 3×3 kernel is performed. This not only achieves multi-scale feature extraction of small target images but also further reduces semantic feature loss and improves the accuracy of small target recognition and localization.
[0071] In one possible implementation, see Figure 5Because 3×3 convolutional kernels easily lose fine-grained information when extracting features from small objects, they limit the neural network's ability to learn key features of small objects, making it difficult for the model to accurately detect small objects. To solve this problem, this disclosure designs a depth transform convolutional structure (DConv) to replace the general 3×3 convolutional layer. After multiple parallel downsampling operations (such as max pooling, stride convolution, etc.), the feature maps obtained from the four downsampling operations are concatenated along the channel dimension. Finally, a 3×3 convolution operation is performed on the concatenated feature map. This achieves multi-scale feature extraction of small object images while further reducing image feature loss and improving the accuracy of small object recognition and localization.
[0072] In one possible implementation, a method for intelligent management and control of nuclear power personnel safety is provided, in which a detection area is set up in the KX plant control area, and the above-mentioned system is deployed in the area to automatically register and compare all items carried by workers when entering and leaving the KX control area.
[0073] The method is used for pre-entry personnel verification, including:
[0074] Step 21: Collect facial images of people waiting to enter the venue, and perform facial recognition and identity verification;
[0075] Step 22: After verifying the identity of the person, capture images of the personal belongings placed on the tray, and use the foreign object detection module to identify and measure the items;
[0076] Step 23: Use security screening equipment to check the items carried by the person. After confirming that the person has no items, simultaneously identify the obtained facial information.
[0077] Step 24: Generate an entry identification item form and send it to personnel for confirmation and administrator for approval;
[0078] Step 25: Collect video images of the personnel's appearance, use a personnel wearing detection module based on multi-frame fusion to identify the wearing status of protective equipment, and determine whether the personnel are wearing the protective equipment correctly based on the identification results;
[0079] Step 26: If the personnel are judged to be dressed correctly, generate a confirmation form and confirm and approve it; if the personnel are judged to be dressed incorrectly, issue an alarm and output the violation item.
[0080] In one possible implementation, the method is also used for pre-departure personnel verification, including:
[0081] Step 31: Collect facial images of people waiting to leave and perform facial recognition and identity verification;
[0082] Step 32: After verifying the identity of the personnel, capture images of the personal items placed on the tray, use the small target consumable identification algorithm model to identify and measure the items, and compare the identified list of outgoing items with the corresponding list of incoming items.
[0083] Step 33: Use security screening equipment to check the items carried by the person. After confirming that the person has no items, synchronize the facial information.
[0084] Step 34: Generate an item comparison form, record and explain any discrepancies, and have it confirmed by personnel and reviewed by the administrator.
[0085] This disclosed method enhances the compliance of nuclear power plant workers' protective equipment wearing and improves the management of foreign object registration when carrying tools and other items into and out of the KX20M control area. It proactively prevents foreign objects from being left behind or falling into the control area, strictly implements item registration, and ensures that the entry and exit of items and personnel are traceable and verifiable. A nuclear power plant entry and exit safety control system needs to be built, utilizing AI computer vision algorithms, data comparison and analysis, and other technologies to replace existing paper registration with an information-based system. This facilitates rapid registration of personnel entry and exit, efficient traceability and verification, and achieves refined management of foreign object prevention.
[0086] In this disclosure, the foreign object detection module combines AI computer vision technology to construct a deep convolutional neural network, enabling intelligent detection of items carried by personnel. Before entering the site, workers place items such as consumables and equipment on the foreign object detection tray. The image acquisition device captures images and uploads them to the server. The foreign object detection module's foreign object recognition model identifies the type and quantity of items in the video and images and automatically registers them.
[0087] This disclosure can be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this disclosure.
[0088] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.
[0089] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.
[0090] Computer program instructions used to perform the operations of this disclosure may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages such as Smalltalk, C++, etc., and conventional procedural programming languages such as the "C" language or similar programming languages. The computer-readable program instructions may execute entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuitry, such as programmable logic circuitry, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), is personalized by utilizing the status information of the computer-readable program instructions to implement various aspects of this disclosure.
[0091] Various aspects of this disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this disclosure. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.
[0092] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.
[0093] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.
[0094] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction containing one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions marked in the blocks may occur in a different order than those shown in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.
[0095] The various embodiments of this disclosure have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical application, or improvement of the technology in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.
Claims
1. A nuclear power plant personnel safety intelligent management and control system, characterized in that, The system includes: image acquisition equipment, security inspection equipment, and a server. The server integrates a face recognition module, a foreign object detection module, and a personnel wear detection module. Image acquisition equipment is used to capture images of people and transmit them to a server; Security screening equipment is used to inspect personal belongings carried by individuals. The facial recognition module is used to perform facial recognition and identity verification on personnel based on the acquired images; The foreign object detection module is used to identify and measure items and output an item list; The personnel wear detection module is used to detect the key points of a person's body based on the acquired images and whether the person is wearing the corresponding equipment at those key points, thereby determining whether the person is wearing protective equipment correctly.
2. The nuclear power safety management and control system according to claim 1, characterized in that, The personnel wearing detection module includes a first feature extraction unit, a first feature fusion unit, and a first prediction unit. The first feature extraction unit is used to extract features from the personnel image, and the first feature fusion unit is used to fuse the extracted features. The first prediction unit performs ARCSA convolution processing on the fused features. The ARCSA convolution processing combines spatial and channel attention mechanisms and uses the receptive field to adaptively and dynamically adjust the convolution kernel size. The first prediction unit also includes a human keypoint regression submodule and a target detection submodule. The regression submodule is used to predict and determine the location of the personnel's human keypoints based on the results of the convolution processing, and the target detection submodule is used to determine whether the personnel are wearing corresponding protective equipment at the location of their human keypoints.
3. The system according to claim 2, characterized in that, ARCSA convolution processing includes: Step 11: Use a basic convolutional layer (which can be a 3x3 convolution) to perform preliminary feature extraction on the input original feature map to obtain the feature map F_base; Step 12: Perform convolution operations on F_base at multiple different scales to obtain feature maps at multiple different scales. Then, perform global average pooling on the feature maps at each scale to obtain the corresponding channel statistics. Input the statistics of each channel into the fully connected layer and obtain the weights after softmax normalization. Multiply the feature map at each scale with the corresponding weights and sum them to obtain the feature map F_rf that integrates multi-scale information. Step 13: After performing global average pooling on F_rf, the first output is obtained through a multilayer perceptron (MLP). After performing max pooling on F_rf, the second output is obtained through the MLP. The first output and the second output are added together and then input into the sigmoid activation function to obtain the channel attention weight Mc. Mc is multiplied by F_rf channel by channel to obtain the channel attention-weighted feature map F_ch. Step 14: Perform average pooling along the channel dimension of F_ch to obtain the first feature map, and perform max pooling along the channel dimension of F_ch to obtain the second feature map. Concatenate the first and second feature maps along the channel dimension, then reduce the number of channels to 1 through a convolutional layer, and then pass the sigmoid activation function to obtain the spatial attention weight Ms. Multiply Ms with F_ch point by point to obtain the spatial attention weighted feature map F_sp. Step 15: Perform a residual connection between F_sp and the original feature map or F_base to obtain the final output.
4. The system according to claim 2, characterized in that, The first prediction unit also adopts a continuous multi-frame information fusion judgment mechanism: taking continuous multi-frame images as a detection cycle, it calculates the distance between the key points of the human body in each frame image and the center point of the detection bounding box of the protective equipment. If the distance is less than or equal to a set threshold, it is determined that the type of the equipment is correctly detected. If the distance is greater than the set threshold, it is determined that the type of the equipment is incorrectly detected. The final detection result is output by combining the judgment results of multiple frames in a detection cycle.
5. The system according to claim 1, characterized in that, The foreign object detection module includes a second feature extraction unit, a second feature fusion unit, and a second prediction unit. The second feature extraction unit is a Darknet network based on the lightweight network structure C2f module, which contains multiple convolutional modules to extract features of different scales from the input image layer by layer and output feature maps of different scales. The second feature fusion unit is a multi-scale feature aggregation network that combines feature pyramid networks and path aggregation structures to fuse feature maps of different scales. The second prediction unit uses a decoupled head structure, including a category prediction branch and a location prediction branch, to generate the category and location of the target item.
6. The system according to claim 1, characterized in that, In the second prediction unit, the low-level feature map P2 is retained to form a multi-level prediction scale from P2 to P5. Different prediction scales are used to detect targets of different sizes.
7. The system according to claim 1, characterized in that, The foreign object detection module uses a depth transform convolutional structure DConv, which performs a 3×3 convolution after concatenating features in the channel dimension through four parallel downsampling operations.
8. A method for intelligent management and control of nuclear power plant personnel safety, said method being implemented based on the system described in any one of claims 1 to 6, characterized in that, The method is used for personnel entry verification, including: Step 21: Collect facial images of people waiting to enter the venue, and perform facial recognition and identity verification; Step 22: After verifying the identity of the person, capture images of the personal belongings placed on the tray, and use the foreign object detection module to identify and measure the items; Step 23: Use security screening equipment to check the items carried by the person. After confirming that the person has no items, simultaneously identify the obtained facial information. Step 24: Generate an entry identification item form and send it to personnel for confirmation and administrator for approval; Step 25: Collect video images of the personnel's appearance, use a personnel wearing detection module based on multi-frame fusion to identify the wearing status of protective equipment, and determine whether the personnel are wearing the protective equipment correctly based on the identification results; Step 26: If the personnel are judged to be dressed correctly, generate a confirmation form and confirm and approve it; if the personnel are judged to be dressed incorrectly, issue an alarm and output the violation item.
9. The method according to claim 7, characterized in that, The method is also used for personnel departure verification, including: Step 31: Collect facial images of people waiting to leave and perform facial recognition and identity verification; Step 32: After verifying the identity of the personnel, capture images of the personal items placed on the tray, use the small target consumable identification algorithm model to identify and measure the items, and compare the identified list of outgoing items with the corresponding list of incoming items. Step 33: Use security screening equipment to check the items carried by the person. After confirming that the person has no items, synchronize the facial information. Step 34: Generate an item comparison form, record and explain any discrepancies, and have it confirmed by personnel and reviewed by the administrator.
10. A non-volatile computer-readable storage medium storing computer program instructions thereon, characterized in that, When the computer program instructions are executed by the processor, they implement the method of claim 8 or 9.