Method, system and equipment for detecting dangerous behaviors of machine room operation and maintenance personnel and medium

Abstract

The invention discloses a dangerous behavior detection method, system and device for machine room operation and maintenance personnel and a medium. The method comprises the steps of collecting original image data in a machine room; preprocessing the original image data to obtain first image data; performing image enhancement processing on the first image data through an image enhancement algorithm to obtain target image data; and inputting the target image data into an improved YOLOv8 model, and detecting the dangerous behavior of the machine room operation and maintenance personnel in the target image data through the improved YOLOv8 model to obtain a dangerous behavior detection result. According to the method, dangerous behaviors possibly occurring in the operation and maintenance process can be identified and positioned more accurately, and the overall detection accuracy and the real-time response capability are improved. The method can be widely applied to the technical field of behavior detection.

Description

Technical Field

[0001] The present application relates to the field of behavior detection technology, and in particular to a method, system, device and medium for detecting dangerous behaviors of computer room operation and maintenance personnel. Background Art

[0002] At present, with the rapid advancement of urbanization in my country, the number and scale of computer rooms are also increasing. In order to ensure the normal operation and maintenance of computer room equipment, computer room operation and maintenance personnel need to frequently enter the computer room to perform various operations. However, the computer room environment is complex and there are many potential dangers, such as high-voltage equipment, complex line layout and high temperature environment, which makes it easy for operation and maintenance personnel to engage in dangerous behaviors during operation, thus causing safety accidents.

[0003] In the related technologies, in order to timely discover and warn of dangerous behaviors of operation and maintenance personnel, thereby improving the safety and stability of computer room operation, there are some methods for detecting dangerous behaviors of operation and maintenance personnel in computer rooms. These include the use of posture recognition technology based on machine vision, which uses cameras to capture the posture and movements of operation and maintenance personnel in real time, and uses deep learning algorithms to analyze whether their behaviors comply with safe operation specifications. However, the current recognition technology is greatly affected by light conditions in the computer room environment. In the case of insufficient light, the recognition effect may decrease, and the practicality is poor.

[0004] In summary, the problems existing in related technologies need to be solved urgently. Summary of the invention

[0005] The purpose of this application is to solve one of the technical problems existing in the related art to at least a certain extent.

[0006] To this end, one purpose of the embodiments of the present application is to provide a method, system, device and medium for detecting dangerous behaviors of computer room operation and maintenance personnel.

[0007] In order to achieve the above technical objectives, the technical solutions adopted by the embodiments of the present application include:

[0008] On the one hand, an embodiment of the present application provides a method for detecting dangerous behaviors of a computer room operation and maintenance personnel, which is applied to a playback terminal, and the method includes:

[0009] Collect the original image data in the computer room;

[0010] Preprocessing the original image data to obtain first image data;

[0011] Performing image enhancement processing on the first image data by using an image enhancement algorithm to obtain target image data;

[0012] The target image data is input into an improved YOLOv8 model, and the dangerous behaviors of the computer room operation and maintenance personnel in the target image data are detected by the improved YOLOv8 model to obtain dangerous behavior detection results; wherein the dangerous behavior detection results are used to characterize the categories of dangerous behaviors of the computer room operation and maintenance personnel in the target image data or the absence of dangerous behaviors of the computer room operation and maintenance personnel in the target image data; the improved YOLOv8 model includes a multi-scale dynamic spatiotemporal attention module.

[0013] In addition, the dangerous behavior detection method for computer room operation and maintenance personnel according to the above embodiment of the present application may also have the following additional technical features:

[0014] Furthermore, in one embodiment of the present application, the collecting of raw image data in the computer room includes:

[0015] Query the historical time node when the original image data was last collected;

[0016] If the time length between the historical time node and the current time node reaches a predetermined time interval, new original image data is acquired;

[0017] Alternatively, when it is determined that a predetermined event is triggered in the machine room, the original image data is acquired.

[0018] Furthermore, in one embodiment of the present application, the preprocessing of the original image data includes:

[0019] Performing denoising processing on the original image data;

[0020] Alternatively, the original image data is cropped;

[0021] Alternatively, the original image data is normalized.

[0022] Furthermore, in one embodiment of the present application, performing image enhancement processing on the first image data by using an image enhancement algorithm to obtain target image data includes:

[0023] Inputting the first image data into a trained convolutional neural network, and performing multi-scale feature fusion and adaptive batch normalization processing on the first image data through the convolutional neural network to obtain second image data;

[0024] By using an improved color visibility restoration module, color intensity averaging and histogram equalization are performed on each color channel of the second image data to obtain third image data;

[0025] The third image data is input into a trained generative adversarial network, and the third image data is optimized and enhanced by the generative adversarial network to obtain the target image data.

[0026] Furthermore, in one embodiment of the present application, inputting the third image data into a trained generative adversarial network, and optimizing and enhancing the third image data through the generative adversarial network, comprises:

[0027] Detecting the light intensity in the computer room, and determining the light parameters according to the light intensity;

[0028] The illumination parameters and the third image data are input into a trained generative adversarial network, and the third image data is optimized and enhanced through the generative adversarial network.

[0029] Furthermore, in one embodiment of the present application, the generative adversarial network includes a generator and a discriminator, and the generative adversarial network is trained by the following steps:

[0030] Get a sample image;

[0031] Inputting the sample image into the generator, and generating a generated image corresponding to the sample image through the generator;

[0032] Inputting the sample image and the generated image into the discriminator for discrimination to obtain a discrimination result;

[0033] According to the discrimination result, determining a corresponding loss value;

[0034] According to the loss value, the parameters of the generative adversarial network are updated.

[0035] Furthermore, in one embodiment of the present application, the multi-scale dynamic spatiotemporal attention module is used to extract spatiotemporal features and perform feature fusion on the spatiotemporal features through dynamically adjusted spatiotemporal attention.

[0036] On the other hand, an embodiment of the present application provides a dangerous behavior detection system for computer room operation and maintenance personnel, the system comprising:

[0037] A collection unit, used to collect raw image data in the computer room;

[0038] A processing unit, configured to pre-process the original image data to obtain first image data;

[0039] an enhancement unit, configured to perform image enhancement processing on the first image data by using an image enhancement algorithm to obtain target image data;

[0040] A prediction unit is used to input the target image data into an improved YOLOv8 model, and detect the dangerous behavior of the computer room operation and maintenance personnel in the target image data through the improved YOLOv8 model to obtain a dangerous behavior detection result; wherein the dangerous behavior detection result is used to characterize the category of the dangerous behavior of the computer room operation and maintenance personnel in the target image data or the absence of the dangerous behavior of the computer room operation and maintenance personnel in the target image data; the improved YOLOv8 model includes a multi-scale dynamic spatiotemporal attention module.

[0041] On the other hand, an embodiment of the present application provides an electronic device, including:

[0042] at least one processor;

[0043] at least one memory for storing at least one program;

[0044] When the at least one program is executed by the at least one processor, the at least one processor implements the above-mentioned dangerous behavior detection method for computer room operation and maintenance personnel.

[0045] On the other hand, an embodiment of the present application further provides a computer-readable storage medium, which stores a program executable by a processor. When the program executable by the processor is executed by the processor, it is used to implement the above-mentioned method for detecting dangerous behaviors of computer room operation and maintenance personnel.

[0046] The advantages and benefits of the present application will be partially given in the following description, and partially become apparent from the following description, or be understood through the practice of the present application:

[0047] The dangerous behavior detection method, system, device and medium of the computer room operation and maintenance personnel disclosed in the embodiment of the present application collects the original image data in the computer room; pre-processes the original image data to obtain the first image data; performs image enhancement processing on the first image data through the image enhancement algorithm to obtain the target image data; inputs the target image data into the improved YOLOv8 model, and detects the dangerous behavior of the computer room operation and maintenance personnel in the target image data through the improved YOLOv8 model to obtain the dangerous behavior detection result. This method can more accurately identify and locate the dangerous behaviors that may occur in the operation and maintenance process, and improve the overall detection accuracy and real-time response capability. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following introduction is made to the drawings of the embodiments of the present application or the related technical solutions in the prior art. It should be understood that the drawings introduced below are only for the convenience of clearly describing some embodiments of the technical solutions of the present application. For technical personnel in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0049] Figure 1 A schematic diagram of an implementation environment of a method for detecting dangerous behaviors of computer room operation and maintenance personnel provided in an embodiment of the present application;

[0050] Figure 2 A flowchart of a method for detecting dangerous behaviors of computer room operation and maintenance personnel provided in an embodiment of the present application;

[0051] Figure 3 A schematic diagram of the structure of a convolutional neural network provided in an embodiment of the present application;

[0052] Figure 4 A schematic diagram of the principle of image enhancement provided in an embodiment of the present application;

[0053] Figure 5 A schematic diagram of the structure of YOLOv8 provided in the related art;

[0054] Figure 6 A schematic diagram of the structure of an improved YOLOv8 provided in an embodiment of the present application;

[0055] Figure 7 A schematic diagram of a processing flow of a multi-scale dynamic spatiotemporal attention module provided in an embodiment of the present application;

[0056] Figure 8 A schematic diagram of the implementation principle of a method for detecting dangerous behaviors of computer room operation and maintenance personnel provided in an embodiment of the present application;

[0057] Fig. 9 A schematic diagram of the structure of a dangerous behavior detection system for computer room operation and maintenance personnel provided in an embodiment of the present application;

[0058] Fig.10 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0059] The present application is further described below in conjunction with the accompanying drawings and specific embodiments. The described embodiments should not be regarded as limiting the present application, and all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present application.

[0060] In the following description, reference is made to “some embodiments”, which describe a subset of all possible embodiments, but it will be understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0061] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0062] 1) YOLOv8: YOLOv8 is the latest version in the YOLO series and is a real-time target detection model. It is used in the field of computer vision to identify and locate objects in images. It combines efficient convolutional neural networks and advanced optimization techniques to achieve extremely fast processing speeds while maintaining high detection accuracy. YOLOv8 provides higher accuracy and lower latency by further improving the network structure and training strategy.

[0063] 2) CNN: CNN is a technology based on convolutional neural network (CNN). It automatically learns and extracts features from images through a deep learning model, performs denoising, brightness and contrast adjustment, etc., thereby significantly improving the clarity and detail of images under low-light conditions, making them more suitable for subsequent target detection and analysis tasks.

[0064] 3) CVR: The Color Visibility Restoration (CVR) module is designed to improve the visual quality of low-light images. CVR evenly distributes the pixel values ​​of the image and improves the brightness and contrast of the image by averaging the color intensity and histogram equalization of each color channel. This process can effectively restore the meaningful information in the image, making the enhanced image more visually natural and clear.

[0065] 4) GAN: Generative Adversarial Network (GAN) consists of a generator and a discriminator, which is mainly used to further optimize the quality of the enhanced image. The generator is responsible for generating realistic images, while the discriminator is used to distinguish between real images and generated images. During the adversarial training process, the generator tries to deceive the discriminator by continuously improving the details and quality of the generated images, thereby capturing and restoring the subtle details in low-light images and effectively suppressing noise, making the final generated image more realistic and natural.

[0066] 5) FPGA: It is a highly flexible integrated circuit that can be programmed and configured after production to implement specific logic functions. It has parallel processing capabilities and real-time processing performance, and is widely used in high-speed data processing, digital signal processing, and video encoding and decoding. FPGA allows users to customize hardware functions according to their needs, providing higher processing speeds and lower latency than software implementations, and is suitable for application scenarios that require high performance and real-time response.

[0067] 6) H.265: It is an advanced video compression standard that can reduce data transmission by about 50% compared to H.264 (AVC) at the same image quality. It improves compression efficiency and image quality through more efficient encoding technology and improved prediction methods, and is particularly suitable for videos with resolutions of 4K and above. H.265 is widely used in scenarios that require efficient video transmission and storage, such as streaming services, HD video conferencing, and ultra-high-definition television.

[0068] With the rapid advancement of urbanization in my country, the number and scale of computer rooms are also increasing. In order to ensure the normal operation and maintenance of computer room equipment, computer room operation and maintenance personnel need to frequently enter the computer room to perform various operations. However, the computer room environment is complex and there are many potential dangers, such as high-voltage equipment, complex line layout and high temperature environment, which makes it easy for operation and maintenance personnel to engage in dangerous behaviors during operations, thereby causing safety accidents. In order to timely detect and warn of dangerous behaviors of operation and maintenance personnel, thereby improving the safety and stability of computer room operation, there are some dangerous behavior detection methods for operation and maintenance personnel in computer rooms.

[0069] At present, the methods for detecting dangerous behaviors of maintenance personnel in computer rooms mainly include the following types: First, combined with environmental perception technology, sensors are used to monitor environmental parameters such as temperature, humidity, and electromagnetic fields in the computer room. When maintenance personnel enter dangerous areas or environmental parameters exceed the safe range, the system will issue an alarm. Secondly, wearable devices such as smart bracelets or helmets can be used to monitor the physiological state and position of maintenance personnel in real time. When abnormal conditions are detected, the system will immediately notify maintenance personnel and management personnel. In addition, posture recognition technology based on machine vision can be used to capture the posture and movements of maintenance personnel in real time through cameras, and use deep learning algorithms to analyze whether their behaviors comply with safe operating specifications. Through the combination of multiple technical means, the safety management level of computer room operation and maintenance can be comprehensively improved, and the occurrence of safety accidents can be effectively prevented.

[0070] However, current strategies often have various drawbacks. For example, environmental perception technology cannot directly monitor human behavior, relies on the accuracy and reliability of sensors, and high-precision sensors and monitoring systems are expensive. Wearable devices require maintenance personnel to wear them consciously, battery life limitations may affect the continuous use of the equipment, and real-time monitoring of physiological status and location may involve privacy issues. The disadvantage of machine vision gesture recognition technology in the computer room environment is that it is greatly affected by light conditions and computing resource requirements. In the case of insufficient light, the recognition effect may decrease; and in the case of insufficient computing resources, the system may lag, resulting in dangerous behavior but delayed warnings, thereby increasing risks. Therefore, real-time video analysis requires higher computing resources and processing power, which increases the cost and complexity of the system.

[0071] In view of this, a method for detecting dangerous behaviors of maintenance personnel in a computer room is provided in an embodiment of the present application. The method collects original image data in the computer room; pre-processes the original image data to obtain first image data; performs image enhancement processing on the first image data through an image enhancement algorithm to obtain target image data; inputs the target image data into an improved YOLOv8 model, and detects dangerous behaviors of maintenance personnel in the computer room in the target image data through the improved YOLOv8 model to obtain dangerous behavior detection results. This method can more accurately identify and locate dangerous behaviors that may occur during the operation and maintenance process, and improve the overall detection accuracy and real-time response capability.

[0072] Please refer to Figure 1 , Figure 1 The implementation environment diagram of a method for detecting dangerous behaviors of maintenance personnel in a computer room provided in an embodiment of the present application is shown. In the implementation environment, the main hardware and software entities involved include a terminal device 110 and a backend server 120. The terminal device 110 and the backend server 120 are connected in communication.

[0073] Specifically, the dangerous behavior detection method for the computer room operation and maintenance personnel provided in the embodiment of the present application can be executed based on the data interaction between the terminal device 110 and the background server 120. Among them, the terminal device 110 in the above embodiment may include a mobile phone, a computer, a smart wearable device, an intelligent voice interaction device, etc., but is not limited to this. The background server 120 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms.

[0074] The terminal device 110 and the backend server 120 may establish a communication connection via a wireless network or a wired network. The wireless network or wired network uses standard communication technology and / or protocols, and the network may be set to the Internet or any other network, such as but not limited to a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a private network or any combination of a virtual private network.

[0075] Of course, it is understandable that Figure 1 The implementation environment in the embodiment of the present application is only some optional application scenarios of the dangerous behavior detection method for computer room operation and maintenance personnel provided in the embodiment of the present application, and the actual application is not fixed. Figure 1 The hardware and software environment shown.

[0076] Below, in combination with the introduction of the aforementioned implementation environment, a dangerous behavior detection method for computer room operation and maintenance personnel provided in an embodiment of the present application is introduced and explained.

[0077] Please refer to Figure 2 , Figure 2 Schematic diagram of a method for detecting dangerous behaviors of computer room operation and maintenance personnel provided in an embodiment of the present application. Specifically, the method for detecting dangerous behaviors of computer room operation and maintenance personnel provided in an embodiment of the present application includes but is not limited to:

[0078] Step 210: collecting original image data in the computer room;

[0079] Step 220: pre-process the original image data to obtain first image data;

[0080] Step 230: performing image enhancement processing on the first image data by using an image enhancement algorithm to obtain target image data;

[0081] Step 240: input the target image data into an improved YOLOv8 model, and detect whether there is any dangerous behavior of the computer room operation and maintenance personnel in the target image data through the improved YOLOv8 model to obtain a dangerous behavior detection result; wherein the dangerous behavior detection result is used to characterize the category of the dangerous behavior of the computer room operation and maintenance personnel in the target image data or the absence of the dangerous behavior of the computer room operation and maintenance personnel in the target image data; the improved YOLOv8 model includes a multi-scale dynamic spatiotemporal attention module.

[0082] In an embodiment of the present application, a method for detecting dangerous behaviors of computer room operation and maintenance personnel is provided to achieve efficient dangerous behavior detection.

[0083] Specifically, in an embodiment of the present application, before performing the dangerous behavior detection operation of the computer room operation and maintenance personnel, you can first select a suitable camera device to ensure that it can still capture clear images under low light conditions. For example, you can use a Sony STARVIS series camera, or select other high-sensitivity cameras, such as FLIR or Axis cameras. This application does not limit this. After completing the selection of the camera device, you can select a suitable installation location in the computer room to set up the camera device to ensure coverage of all key areas, such as server racks, power distribution cabinets, and entrances and exits. Determine the installation location of the camera device to ensure coverage without blind spots and improve the monitoring effect. In an embodiment of the present application, after the installation location is determined, the camera device is installed on the ceiling or wall to ensure that it is stable and fixed, with an appropriate viewing angle to avoid obstruction. Of course, in some other embodiments, a mobile camera can also be used to cover the area.

[0084] In an embodiment of the present application, after the installation of the camera equipment is completed, the dangerous behavior detection of the computer room operation and maintenance personnel can be carried out. Specifically, the original image data in the computer room can be collected by the installed camera equipment. After the original image data collection is completed, the collected original image data is transmitted to the processing server or storage device using a network transmission protocol (such as FTP or HTTP) for further processing. When processing the original image data, first, it can be pre-processed, and the image data obtained after the pre-processing is recorded as the first image data. Exemplarily, in an embodiment of the present application, when the original image data is pre-processed, some of the following processing methods may be included but are not limited to: denoising processing, cropping processing and normalization processing. The image tools that can be used here can be OpenCV library or Matlab, etc. By pre-processing the original image data, the image quality can be improved and noise and irrelevant information can be reduced.

[0085] After the first image data is obtained by preprocessing, in an embodiment of the present application, the first image data is subjected to image enhancement processing by an image enhancement algorithm to obtain target image data. Exemplarily, in an embodiment of the present application, an improved convolutional neural network (CGCNN) can be used for image enhancement. The input image is subjected to a series of convolutional layers and pooling layers to extract preliminary features, the feature map is enhanced using a self-attention mechanism, the image resolution is improved through a deconvolution layer, and finally the brightness and contrast of the image are improved and the details are enhanced through an activation function (such as ReLU). In some embodiments, other image enhancement algorithms (such as Retinex) may also be used, and this application does not limit this. After obtaining the target image data, it can be stored in a specified folder or database, and file naming rules (such as timestamps) can be used to facilitate subsequent target detection.

[0086] In the embodiment of the present application, when detecting the target image data, it can be input into the improved YOLOv8 model, in which a multi-scale dynamic spatiotemporal attention module is included. In the embodiment of the present application, its optimization in multi-scale spatiotemporal feature extraction can be utilized to improve the detection effect of potentially dangerous behaviors of operation and maintenance personnel. By enhancing the feature extraction capability of different scales and time points, it is possible to more accurately identify and locate dangerous behaviors that may occur during the operation and maintenance process, thereby improving the overall detection accuracy and real-time response capability.

[0087] In an embodiment of the present application, the dangerous behavior of the computer room operation and maintenance personnel in the target image data is detected by an improved YOLOv8 model to obtain a dangerous behavior detection result. The dangerous behavior detection result here can be used to characterize the category of dangerous behavior of the computer room operation and maintenance personnel in the target image data. For example, in some embodiments, the category of dangerous behavior of the computer room operation and maintenance personnel can include 8 types: accidentally touching sharp edges, improper use of tools, contact with running equipment, high-altitude operations without wearing safety belts, electrical operations without power off, wrong posture for heavy objects, incomplete wearing of protective equipment, and unauthorized entry into dangerous areas. Of course, the specific categories can be flexibly set according to actual needs, and this application is not limited to this. When the behavior of the computer room operation and maintenance personnel in the target image data is compliant and does not belong to any of the set dangerous behavior categories, the dangerous behavior detection result can also be used to characterize the dangerous behavior of the computer room operation and maintenance personnel in the target image data. In an embodiment of the present application, the corresponding relationship between the specific data format of the dangerous behavior detection result and the represented meaning is not limited by this application.

[0088] It is understandable that in an embodiment of the present application, a method for detecting dangerous behaviors of computer room operation and maintenance personnel is provided, by deploying multiple cameras in the computer room to capture and analyze the behavior images of the operation and maintenance personnel in real time. Using deep learning algorithms for data matching and recognition, this system can detect potential dangerous behaviors with high accuracy while minimizing interference with normal work activities. In addition, in order to address the problem that traditional machine vision systems are difficult to be widely used in real-time detection due to their high computational complexity, this application is optimized based on image enhancement technology and an improved YOLOv8 model, and is adjusted for the typical low-light environment and limited computing resources of the computer room, significantly improving the recognition accuracy and detection speed under these conditions. Through these technical improvements, not only the efficiency of security management is improved, but also a safer working environment is provided for computer room operation and maintenance personnel.

[0089] Specifically, in some embodiments, collecting raw image data in the computer room includes:

[0090] Query the historical time node when the original image data was last collected;

[0091] If the time length between the historical time node and the current time node reaches a predetermined time interval, new original image data is acquired;

[0092] Alternatively, when it is determined that a predetermined event is triggered in the machine room, the original image data is acquired.

[0093] In the embodiments of the present application, when collecting the original image data in the computer room, in some embodiments, it can be collected in a timed manner. For example, new original image data is collected every certain period of time. Specifically, after each collection, the historical time node of the last collection of the original image data can be queried in real time, and then the length of time from the historical time node to the current time node is compared to see if it reaches the predetermined time interval. If it reaches it, a collection task is executed; if not, continue to wait. The predetermined time interval here can be flexibly set as needed, and this application does not limit it.

[0094] In other embodiments, a task event that triggers the acquisition of raw image data can be pre-set. If these predetermined events are triggered in the computer room, a collection task can be executed to obtain the corresponding raw image data. The specific circumstances of the predetermined events can be flexibly set, and this application is also not limited.

[0095] Specifically, in some embodiments, performing image enhancement processing on the first image data by using an image enhancement algorithm to obtain target image data includes:

[0096] Inputting the first image data into a trained convolutional neural network, and performing multi-scale feature fusion and adaptive batch normalization processing on the first image data through the convolutional neural network to obtain second image data;

[0097] By using an improved color visibility restoration module, color intensity averaging and histogram equalization are performed on each color channel of the second image data to obtain third image data;

[0098] The third image data is input into a trained generative adversarial network, and the third image data is optimized and enhanced by the generative adversarial network to obtain the target image data.

[0099] In an embodiment of the present application, when performing image enhancement processing on the first image data, a variety of strategies can be combined. Specifically, the present application can use a convolutional neural network, an improved color visibility restoration (ICVR) module, and a generative adversarial network to implement image enhancement processing. First, the first image data can be processed using a trained convolutional neural network, and the brightness and contrast of the image can be enhanced through multi-scale feature fusion and adaptive batch normalization to obtain second image data. Then, an improved color visibility restoration (ICVR) module can be used to perform color intensity averaging and histogram equalization on each color channel of the second image data to improve the visual quality of the image and obtain third image data. Next, a generative adversarial network is introduced to further optimize the enhanced image, enhance details and noise suppression effects, and obtain target image data. The following is a detailed introduction and explanation of these three parts.

[0100] Convolutional Neural Networks

[0101] Please refer to Figure 3 , Figure 3 A schematic diagram of the structure of a convolutional neural network provided in an embodiment of the present application is shown. In the embodiment of the present application, the functions of each layer of the convolutional neural network are as follows:

[0102] Input layer: input image.

[0103] Convolutional Layers: Multiple convolutional layers are stacked in sequence. Each convolutional layer uses several convolution kernels (filters) to perform convolution operations to extract local features of the image. The size of the convolution kernel is generally 3×3×3. After the convolution operation, the nonlinear activation function ReLU is used to increase the nonlinear expression ability of the model.

[0104] Batch Normalization Layers: Add batch normalization layers after some convolutional layers to standardize the data distribution of each mini-batch, speed up the training process and improve the stability of the model.

[0105] Pooling Layers: Common pooling operations include Max Pooling and Average Pooling, which are used to downsample feature maps, reduce data dimensions, and retain important features.

[0106] Multi-scale Feature Fusion: Capture the multi-scale information of the image by fusing feature maps of different scales. Feature maps of different levels can be combined to enhance the feature expression ability of the model.

[0107] Adaptive Batch Normalization: During the training process, the batch normalization parameters are adaptively adjusted to make the model better adaptable to different types of low-light images. This method can improve the model's generalization ability and ability to handle diverse data.

[0108] Upsampling Layers: Use deconvolution or interpolation methods to restore the low-resolution feature map to the original image resolution. This step is used to generate an enhanced high-resolution image.

[0109] Output layer: The last layer is usually a convolutional layer that transforms the feature map into the final enhanced image. The output image is the same size as the input image, but the brightness and contrast are enhanced.

[0110] Loss Function: Use a suitable loss function (such as mean square error (MSE) or perceptual loss) to measure the difference between the generated image and the target image to guide the training of the model.

[0111] Through the design and training of the above levels, convolutional neural networks can effectively enhance the brightness and contrast of low-light images and improve the visual quality of images.

[0112] Improved color visibility restoration module

[0113] In the embodiment of the present application, an adaptive method is used to adjust the balance of each color channel to better respond to color deviations under different lighting conditions. Specifically, in the embodiment of the present application, the adaptive color balancing method can be used to calculate the color channel average value, and the improved color difference adjustment method can be used to enhance the color contrast:

[0114]

[0115] μ c '=αμ c +(1-α)μ y

[0116] In the formula, μ c Represents the average value of each color channel, μ y is a grayscale image, c represents the color channel, J represents the processed image; M and N represent the width and height of the image respectively. i and j represent the width and height coordinates within the image. α is an adaptive weight factor that depends on the overall brightness of the image and can balance the video conditions of rooms with different brightness.

[0117] In the embodiment of the present application, the improved color difference adjustment can also be used to calculate the color difference image dc of the red, green and blue channels. σ is an enhancement factor used to adjust the contrast of the color difference image, which is flexibly adjusted according to the random computer room environment. In the image processing process of the ICVR module, multiple key parameters work together to achieve efficient color restoration and contrast enhancement. The adaptive weight factor α, the enhancement factor σ, and the cdf 25％ and cdf 75％ The dynamic adjustment of these four parameters under different lighting conditions has a decisive influence on the final image quality.

[0118] First, the determination of the adaptive weight factor α needs to consider the balancing effect of the overall brightness of the image. By adjusting the α value, the system can automatically adapt to different lighting conditions, thereby optimizing the brightness performance of the image. In order to ensure that the selection of α does not introduce too much noise or distortion while improving image quality, the embodiment of the present application can be optimized in combination with image enhancement performance indicators such as SSIM (structural similarity index) and PSNR (peak signal-to-noise ratio). Through a large number of experiments, the system will screen out a set of candidate α values ​​that meet these performance indicators. These values ​​can meet or even exceed the preset SSIM and PSNR thresholds within a running time of no more than 10ms. The enhancement factor σ is mainly used to adjust the contrast of the color difference image. Contrast enhancement is crucial to the detail performance and color separation effect of the image. The selection of the σ value also needs to be rigorously tested to ensure that it can achieve the expected effect under different brightness and color difference conditions. The system will establish a candidate pool of enhancement factors σ based on the experimental results, and the candidate values ​​can provide the best image enhancement effect within a processing time of no more than 10ms. In addition, cdf 25％ and cdf 75％ These two parameters play an important role in the color difference adjustment process. They represent the values ​​of the cumulative distribution function of the color difference image dc at the 25% and 75% quantiles, respectively. They are mainly used to measure the concentration of the image brightness distribution, thereby guiding the contrast enhancement. By dynamically adjusting these two parameters, the system can automatically detect and adapt to the brightness and color difference distribution in the image, so as to adjust the image contrast and detail performance in real time under different environmental conditions. Similar to the determination method of the adaptive weight factor α and the enhancement factor σ, the system will optimize cdf25 and cdf75 according to image enhancement performance indicators such as SSIM and PSNR, and establish a candidate parameter pool whose values ​​can provide stable and high-quality image output within a running time of no more than 10ms.

[0119] The ICVR module provides two modes for parameter selection: precise mode and fast mode. In precise mode, the system gives priority to the parameter combination in the candidate pool that can maximize the image quality, so as to give full play to the performance indicators such as SSIM and PSNR while ensuring the best visual effect. In fast mode, the system will give priority to the parameters with the fastest calculation speed to ensure efficient image processing in real-time applications.

[0120] d c =σ(μ y -μ c )

[0121] Then the histogram of the equalized image is calculated by the advanced histogram equalization method.

[0122]

[0123] In the formula, cdf 25％ and cdf 75％ They are d c The cumulative distribution function values ​​of the 25% and 75% quantiles. L represents d c The intensity level of the image, J eh Represents the equalized image.

[0124]

[0125] J c Represents the image after color restoration. Representative J eh The maximum and minimum values ​​of the image pixels.

[0126] It can be understood that in the embodiments of the present application, under the application of these improvements, ICVR not only improves the visual effect of the image, but also adapts to the needs of working under different environmental conditions, especially in low light and high dynamic range conditions, and can provide better color restoration.

[0127] Generative Adversarial Networks

[0128] In the embodiment of the present application, a generative adversarial network (IGAN) is also introduced to further optimize the enhanced image, enhance the details and noise suppression effect.

[0129] Specifically, in the embodiment of the present application, in order to make the generator better adapt to the dim environment, a light level variable can be introduced so that the generator can adjust the generation strategy according to the specific conditions of the dim environment when generating images. For example, for the third image data, when optimizing and enhancing it, the light intensity in the room can be detected, and the light parameters can be determined according to the light intensity. Then, the light parameters and the third image data are input into the trained generative adversarial network together, so as to optimize and enhance it.

[0130] The generative adversarial network in the embodiment of the present application may include a generator and a discriminator. Among them, the generator can be used to generate an optimized image, and the discriminator is used to determine whether the image is a real image or a generated image. When training the generative adversarial network, a batch of sample images can be obtained, the sample images are input into the generator to generate the corresponding generated images, and then the discriminator is used to perform discrimination to obtain the discrimination result. During the training process, the generator and the discriminator can be trained alternately. For the generator, the goal of the training process generator is to maximize the probability of the discriminator's incorrect judgment of the generated image, that is, to minimize the following loss function:

[0131]

[0132] In the formula, l G is the loss value corresponding to the generator, G(x|θ) is the output of the generator, z represents the generated image, and θ is the light intensity variable.

[0133] The goal of the discriminator is to maximize the probability of correct judgment of real images and minimize the probability of incorrect judgment of generated images, that is, to minimize the following loss function:

[0134]

[0135] In the formula, l D is the loss value corresponding to the discriminator, G(x|θ) is the output of the generator, x represents the real image (sample image), z represents the generated image, and θ is the light intensity variable.

[0136] Reference Figure 4 , Figure 4 A schematic diagram of the principle of image enhancement provided in the embodiment of the present application is shown. It can be understood that in the embodiment of the present application, through the above-mentioned improvement of image enhancement, while reducing the computational cost and time complexity, the low-light image enhancement effect and the adaptability of the model are significantly improved, ensuring the high recognition accuracy and detection speed of the system under low-light conditions.

[0137] In the embodiment of the present application, the model used in the detection is an improved YOLOv8 model. Figure 5 and Figure 6 , Figure 5 A schematic diagram of the structure of YOLOv8 provided in the related art is shown. Figure 6 A schematic diagram of the structure of an improved YOLOv8 provided in an embodiment of the present application is shown. In the YOLOv8 model, the Backbone structure is responsible for extracting feature information at different levels, the Neck structure is responsible for fusing features of different scales output by the Backbone, and the detection head Head is a decoupling head responsible for classification and regression tasks. In an embodiment of the present application, in order to improve the performance of the YOLOv8 model in monitoring dangerous behaviors of computer room operation and maintenance personnel, a multi-scale dynamic spatiotemporal attention module is introduced.

[0138] The multi-scale dynamic spatiotemporal attention module can be used for multi-scale spatiotemporal feature extraction and spatiotemporal attention dynamic adjustment. This module uses 3D convolution to extract spatiotemporal features at multiple scales and dynamically adjusts the attention weights in combination with the self-attention mechanism to accurately capture key behaviors, significantly enhancing the model's ability to analyze human behavior and improving detection accuracy and robustness. For details, please refer to Figure 7 , Figure 7 A schematic diagram of the processing flow of a multi-scale dynamic spatiotemporal attention module provided in an embodiment of the present application is shown.

[0139] In the embodiment of the present application, the multi-scale dynamic spatiotemporal attention module can be used for:

[0140] 1. Extract spatiotemporal features:

[0141] F u =ω(W u *X+b u )

[0142] In the formula, F u represents the spatiotemporal feature map of the u-th scale; ω represents the activation function; W u is the weight of the u-th 3D convolutional layer; X represents the input feature map (or image data); b u is the bias term.

[0143] 2. Dynamically adjust spatiotemporal attention:

[0144] Spatiotemporal attention dynamic adjustment is mainly to dynamically adjust the spatiotemporal attention weights so that the model can focus more on the key information and behaviors in the video content. This process usually involves evaluating the importance of specific areas or features in the video frame and allocating more attention accordingly.

[0145] A ijklmn =softmax(V×tanh(W a ×F+U a ×P))

[0146] In the formula, A ijklmn represents the elements of the spatiotemporal attention weight matrix; softmax is the normalization function; V is the weight vector of the attention mechanism; W a and U a is a learnable weight matrix; F is the feature map extracted by multi-scale spatiotemporal feature extraction; P is the additional feature introduced by the external environment.

[0147] 3. Apply dynamically adjusted spatiotemporal attention:

[0148] F att =A·F

[0149] In the formula, F att represents the feature map after applying spatiotemporal attention; A represents the spatiotemporal attention weight matrix.

[0150] 4. Feature Fusion

[0151] F out =conv1x1(F att +F)

[0152] In the formula, F out Represents the fused output feature map; conv1x1 represents a 1x1 convolution operation, which is used to fuse features.

[0153] It is understandable that in the embodiment of the present application, by improving the image enhancement method and optimizing the YOLOv8 model, the problems of insufficient lighting in the computer room and large computing resource requirements of the machine vision model are solved. Specifically, by using the improved image enhancement method, the brightness and contrast of the image under low light conditions are improved, and the image details are enhanced; by lightweight processing and optimization of the YOLOv8 model, the number of parameters and computational complexity of the model are reduced, and the computational requirements are further reduced through quantization and pruning technology, and the detection speed is improved; finally, the image enhancement module is integrated with the improved YOLOv8 model to form a complete dangerous behavior detection system, and tested and optimized in the actual computer room environment to ensure the high recognition accuracy and detection speed of the system under low light conditions, thereby ensuring the reliability and stability of the system in practical applications.

[0154] In order to verify the effectiveness and application value of this application, the collected picture data set was used for experimental testing. The experiment used 500 on-site photos to test the model proposed in this application, and selected the most common dangerous behaviors of operation and maintenance personnel in the computer room for testing. The data set is divided into training set, verification set and test set in a ratio of 6:1:1. The experiment selected accuracy as the evaluation index. In the embodiment of this application, the H.265 video compression standard is also used. Compared with the traditional H.264 standard, it can reduce the amount of data transmission by about 50% without sacrificing video quality, effectively alleviating bandwidth pressure. At the same time, the technology also integrates programmable hardware logic blocks and connected FPGAs. FPGA technology has significant advantages in processing low-light images because it can use its highly parallel hardware architecture for real-time image enhancement and noise reduction, which may take longer processing time on traditional CPUs or GPUs. Through programmable logic blocks, FPGAs can efficiently execute complex image processing algorithms, thereby improving the brightness and contrast of images while retaining image detail information.

[0155] The dangerous behavior dataset of data center operation and maintenance personnel includes the following behavior samples: accidentally touching sharp edges, improper use of tools, touching running equipment, not wearing safety belts when working at heights, not turning off the power when doing electrical work, incorrect posture when carrying heavy objects, not wearing protective equipment completely, and entering dangerous areas without authorization, as shown in Table 1.

[0156] Table 1

[0157] Dangerous physical behavior Sample labeling Accidental touch on sharp edges Sharp Improper use of tools ToolMisuse Contact Operation Equipment Behavior Contact Not wearing a safety belt when working at height Unbelted Electrical work without powering off LiveWork Wrong posture when carrying heavy objects Lifting Failure to wear protective equipment NoPPE Unauthorized entry into dangerous areas NoEntry

[0158] Experiments have shown that the accuracy of the improved dangerous behavior detection algorithm in this application can be improved by 10%-20% compared to the single YOLOv8 model in the detection of operation and maintenance personnel, and the accuracy in the detection of other types of dangerous behaviors has also been significantly improved. While reducing the amount of calculation, it also reduces the probability of false detection. This improvement will continue to expand as the amount of training data in the computer room increases. The detailed comparison results are shown in Table 2.

[0159] Table 2

[0160]

[0161] It can be seen that compared with the existing detection methods, this application has superior accuracy. Aiming at the problem of detecting dangerous behaviors of computer room operation and maintenance personnel, the YOLOv8 model improved by image enhancement algorithm and multi-scale dynamic spatiotemporal attention module (MDSTAC) is adopted to make the model lighter, more accurate, and more suitable for the application scenarios of computer room operation and maintenance.

[0162] Specifically, the present application significantly outperforms other models in terms of parameter quantity and computational complexity on a large-scale video action detection dataset. Computational complexity: The present application has the lowest GFlops (105), which is lower in computational complexity than other models. Spatial action detection accuracy (F-mAP): The present application has the highest F-mAP (88.5%), which is better than other models. Spatiotemporal action detection accuracy (V-mAP): The present application has the highest V-mAP (85.2%). These data show that the present application not only has an advantage in computational complexity, but also performs best in accuracy, demonstrating its ability to maintain efficient computing while improving detection accuracy, significantly improving the overall performance of the model.

[0163] Compared with models of the same magnitude, this application achieves the best accuracy performance and effectively reduces the difficulty of deploying image enhancement and behavior detection models to edge devices. Through the further design of the lightweight model of this application, combined with the improved image enhancement algorithm and YOLOv8's multi-scale spatiotemporal feature extraction and spatiotemporal attention dynamic adjustment, the best balance between accuracy and speed is achieved. The improved algorithm designed in this application significantly improves the accuracy of dangerous behavior detection of computer room personnel and the overall efficiency of the system while retaining the lightweight of the model.

[0164] Reference Figure 8 , Figure 8 The following is a schematic diagram showing the implementation principle of a method for detecting dangerous behaviors of computer room operation and maintenance personnel provided in an embodiment of the present application. The method in the embodiment of the present application has at least the following technical effects:

[0165] 1. Improved image enhancement algorithm

[0166] By introducing a CGCNN algorithm for low-light image enhancement, this technology significantly improves the brightness and contrast of the image. The algorithm includes the following four main steps: first, design and train the CGCNN model to focus on the low-light image enhancement task; second, process low-light images from different sources through multi-scale feature fusion and adaptive batch normalization to improve the brightness and contrast of the image; then, propose the ICVR technology to improve the computing efficiency in resource-constrained environments. In addition, this technology adopts the H.265 video compression standard to effectively alleviate bandwidth pressure. The technology also integrates programmable hardware logic blocks and connected FPGAs, allowing the system to efficiently perform parallel data processing for image and video processing. Finally, by introducing an improved generative adversarial network (IGAN), the generator and discriminator are trained in dim environments using light level variables as adjustment factors. This method makes the generated images closer to the low-light conditions of real scenes, thereby significantly enhancing the image details and noise suppression effects. These comprehensive technical improvements significantly improve the processing speed and efficiency of the system, especially for computer room scenarios.

[0167] 2. Improved YOLOv8 algorithm

[0168] In order to improve the performance of YOLOv8 in monitoring dangerous behaviors of maintenance personnel in computer rooms, the Multiscale Dynamic Spatiotemporal Attention Module (MDSTAC) was introduced. The main improvements of this module include multiscale spatiotemporal feature extraction and dynamic adjustment of spatiotemporal attention: by applying multiscale spatiotemporal feature extraction at multiple scales, rich spatiotemporal features are extracted, and the attention weights are dynamically adjusted in combination with the self-attention mechanism to accurately capture key behaviors. In addition, the 1×1 convolutional layer is used to efficiently fuse features, effectively reducing feature redundancy and improving learning efficiency. These innovations significantly enhance the model's ability to identify dangerous behaviors of maintenance personnel and significantly improve the accuracy and robustness of detection.

[0169] 3. Method for detecting dangerous behaviors of personnel in combined computer room

[0170] By combining the improved CGCNN image enhancement algorithm and the optimized YOLOv8 algorithm, the effect of monitoring dangerous behaviors of maintenance personnel in the computer room is significantly improved. The CGCNN algorithm combines ICVR technology and IGAN technology with the CNN method to significantly enhance the brightness, contrast and visual quality of images under low light conditions, while effectively reducing the model size and computational complexity. The YOLOv8 algorithm introduces the multi-scale dynamic spatiotemporal attention module (MDSTAC), which extracts spatiotemporal features in the video frame sequence through scale spatiotemporal feature extraction and dynamic adjustment of spatiotemporal attention, and combines 1×1 convolutional layers for efficient feature fusion, significantly improving the YOLOv8 algorithm's ability to identify dangerous behaviors of maintenance personnel. This method effectively combines image enhancement and spatiotemporal feature extraction technology, while reducing computational cost and time complexity, greatly improving the image enhancement effect and behavior detection accuracy under low light conditions, and ensuring the high recognition accuracy and real-time performance of the system in complex environments.

[0171] Reference Fig. 9 In an embodiment of the present application, a dangerous behavior detection system for computer room operation and maintenance personnel is also provided, including:

[0172] The acquisition unit 910 is used to acquire the original image data in the computer room;

[0173] A processing unit 920 is used to pre-process the original image data to obtain first image data;

[0174] An enhancement unit 930 is configured to perform image enhancement processing on the first image data by using an image enhancement algorithm to obtain target image data;

[0175] The prediction unit 940 is used to input the target image data into the improved YOLOv8 model, and detect the dangerous behavior of the computer room operation and maintenance personnel in the target image data through the improved YOLOv8 model to obtain a dangerous behavior detection result; wherein the dangerous behavior detection result is used to characterize the category of the dangerous behavior of the computer room operation and maintenance personnel in the target image data or the absence of the dangerous behavior of the computer room operation and maintenance personnel in the target image data; the improved YOLOv8 model includes a multi-scale dynamic spatiotemporal attention module.

[0176] It can be understood that the contents of the above method embodiments are all applicable to the present system embodiments, the functions specifically implemented by the present system embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0177] Reference Fig.10 , an embodiment of the present application provides an electronic device, including:

[0178] at least one processor 1010;

[0179] At least one memory 1020, used to store at least one program;

[0180] When at least one program is executed by at least one processor 1010, at least one processor 1010 implements the above-mentioned dangerous behavior detection method for computer room operation and maintenance personnel.

[0181] Similarly, the contents of the above method embodiments are all applicable to the electronic device embodiments. The functions specifically implemented by the electronic device embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0182] The embodiment of the present application also provides a computer-readable storage medium, which stores a program executable by the processor 1010. The program executable by the processor 1010 is used to execute the above-mentioned dangerous behavior detection method for computer room operation and maintenance personnel when executed by the processor 1010.

[0183] Similarly, the contents of the above method embodiments are all applicable to the computer-readable storage medium embodiments. The functions specifically implemented by the computer-readable storage medium embodiments are the same as those of the above method embodiments, and the beneficial effects achieved are also the same as those achieved by the above method embodiments.

[0184] In some selectable embodiments, the function / operation mentioned in the block diagram may not occur in the order mentioned in the operation diagram. For example, depending on the function / operation involved, the two boxes shown in succession can actually be executed substantially simultaneously or the boxes can sometimes be executed in reverse order. In addition, the embodiment presented and described in the flow chart of the application is provided by way of example, for the purpose of providing a more comprehensive understanding of technology. The disclosed method is not limited to the operation and logic flow presented herein. Selectable embodiments are expected, wherein the order of various operations is changed and the sub-operation of a part described as a larger operation is performed independently.

[0185] In addition, although the present application is described in the context of functional modules, it should be understood that, unless otherwise specified, one or more of the functions and / or features can be integrated into a single physical device and / or software module, or one or more functions and / or features can be implemented in a separate physical device or software module. It is also understood that a detailed discussion of the actual implementation of each module is unnecessary for understanding the present application. More specifically, in view of the properties, functions, and internal relationships of the various functional modules in the device disclosed herein, the actual implementation of the module will be understood within the conventional techniques of the engineer. Therefore, those skilled in the art can implement the present application set forth in the claims without excessive experimentation using ordinary techniques. It is also understood that the specific concepts disclosed are merely illustrative and are not intended to limit the scope of the present application, which is determined by the full scope of the attached claims and their equivalents.

[0186] If the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for a device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program codes.

[0187] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in conjunction with such instruction execution systems, devices or apparatuses. For the purposes of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in conjunction with such instruction execution systems, devices or apparatuses.

[0188] More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or more wires (electronic device), a portable computer disk case (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be a paper or other suitable medium on which the program is printed, since the program may be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, deciphering or, if necessary, processing in another suitable manner, and then stored in a computer memory.

[0189] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above-mentioned embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0190] In the above description of this specification, the description with reference to the terms "one embodiment / example", "another embodiment / example" or "certain embodiments / examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present application. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0191] Although the embodiments of the present application have been shown and described, those skilled in the art will appreciate that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present application, and that the scope of the present application is defined by the claims and their equivalents.

[0192] The above is a specific description of the preferred implementation of the present application, but the present application is not limited to the embodiments. Technical personnel familiar with the field can make various equivalent modifications or substitutions without violating the spirit of the present application. These equivalent modifications or substitutions are all included in the scope defined by the claims of the present application.

Claims

1. A method for detecting dangerous behaviors of computer room operation and maintenance personnel, characterized in that: The method comprises: Collect the original image data in the computer room; Preprocessing the original image data to obtain first image data; Performing image enhancement processing on the first image data by using an image enhancement algorithm to obtain target image data; The target image data is input into an improved YOLOv8 model, and the dangerous behaviors of the computer room operation and maintenance personnel in the target image data are detected by the improved YOLOv8 model to obtain dangerous behavior detection results; wherein the dangerous behavior detection results are used to characterize the categories of dangerous behaviors of the computer room operation and maintenance personnel in the target image data or the absence of dangerous behaviors of the computer room operation and maintenance personnel in the target image data; the improved YOLOv8 model includes a multi-scale dynamic spatiotemporal attention module.

2. A method for detecting dangerous behaviors of computer room operation and maintenance personnel according to claim 1, characterized in that: The collecting of raw image data in the computer room includes: Query the historical time node when the original image data was last collected; If the time length between the historical time node and the current time node reaches a predetermined time interval, new original image data is acquired; Alternatively, when it is determined that a predetermined event is triggered in the machine room, the original image data is acquired.

3. A method for detecting dangerous behaviors of computer room operation and maintenance personnel according to claim 1, characterized in that: The preprocessing of the original image data comprises: Performing denoising processing on the original image data; Alternatively, the original image data is cropped; Alternatively, the original image data is normalized.

4. A method for detecting dangerous behaviors of computer room operation and maintenance personnel according to any one of claims 1 to 3, characterized in that: The performing image enhancement processing on the first image data by using an image enhancement algorithm to obtain target image data includes: Inputting the first image data into a trained convolutional neural network, and performing multi-scale feature fusion and adaptive batch normalization processing on the first image data through the convolutional neural network to obtain second image data; By using an improved color visibility restoration module, color intensity averaging and histogram equalization are performed on each color channel of the second image data to obtain third image data; The third image data is input into a trained generative adversarial network, and the third image data is optimized and enhanced by the generative adversarial network to obtain the target image data.

5. A method for detecting dangerous behaviors of computer room operation and maintenance personnel according to claim 4, characterized in that: The step of inputting the third image data into a trained generative adversarial network and optimizing and enhancing the third image data through the generative adversarial network includes: Detecting the light intensity in the computer room, and determining the light parameters according to the light intensity; The illumination parameters and the third image data are input into a trained generative adversarial network, and the third image data is optimized and enhanced through the generative adversarial network.

6. A method for detecting dangerous behaviors of computer room operation and maintenance personnel according to claim 4, characterized in that: The generative adversarial network includes a generator and a discriminator, and the generative adversarial network is trained by the following steps: Get a sample image; Inputting the sample image into the generator, and generating a generated image corresponding to the sample image through the generator; Inputting the sample image and the generated image into the discriminator for discrimination to obtain a discrimination result; According to the discrimination result, determining a corresponding loss value; According to the loss value, the parameters of the generative adversarial network are updated.

7. A method for detecting dangerous behaviors of computer room operation and maintenance personnel according to claim 1, characterized in that: The multi-scale dynamic spatiotemporal attention module is used to extract spatiotemporal features and perform feature fusion on the spatiotemporal features through dynamically adjusted spatiotemporal attention.

8. A dangerous behavior detection system for computer room operation and maintenance personnel, characterized in that: The system comprises: A collection unit, used to collect raw image data in the computer room; A processing unit, configured to pre-process the original image data to obtain first image data; an enhancement unit, configured to perform image enhancement processing on the first image data by using an image enhancement algorithm to obtain target image data; A prediction unit is used to input the target image data into an improved YOLOv8 model, and detect the dangerous behavior of the computer room operation and maintenance personnel in the target image data through the improved YOLOv8 model to obtain a dangerous behavior detection result; wherein the dangerous behavior detection result is used to characterize the category of the dangerous behavior of the computer room operation and maintenance personnel in the target image data or the absence of the dangerous behavior of the computer room operation and maintenance personnel in the target image data; the improved YOLOv8 model includes a multi-scale dynamic spatiotemporal attention module.

9. An electronic device, characterized in that: include: at least one processor; at least one memory for storing at least one program; When the at least one program is executed by the at least one processor, the at least one processor implements a method according to any one of claims 1 to 7.

10. A computer-readable storage medium storing a program executable by a processor, characterized in that: The processor-executable program is used to implement a method according to any one of claims 1 to 7 when executed by the processor.

Images