Big data computer room monitoring method, device, equipment and storage medium

Abstract

The present application relates to the field of big data technology, and discloses a big data computer room monitoring method, device, equipment and storage medium. The method comprises: performing infrared image scanning and preprocessing on a server unit through a monitoring device to obtain a plurality of denoised infrared images; performing overlapping area recognition and image splicing on a plurality of denoised infrared images to obtain a spliced ​​infrared image; constructing an edge perception factor and a detail adjustment factor of the spliced ​​infrared image, and generating a corresponding target infrared image; performing pixel temperature extraction on the target infrared image to obtain pixel temperature data, and constructing a surface temperature field of the server unit; performing feature extraction on the surface temperature field to obtain a target temperature feature set; inputting the target temperature feature set into a unit abnormality discrimination model to perform server unit abnormality discrimination, and obtaining a target abnormality discrimination result. The present application improves the accuracy of temperature abnormality monitoring of server units in a big data computer room.

Description

Technical Field

[0001] The present application relates to the field of big data technology, and in particular to a big data computer room monitoring method, device, equipment and storage medium. Background Art

[0002] In data centers, the reliable operation of server units is crucial to the stability of the entire system. However, due to long-term high-load operation of server units, problems such as overheating and local heat accumulation often occur. If these problems are not discovered and handled in time, they may lead to server performance degradation, hardware damage, or even system crashes.

[0003] Traditional temperature monitoring methods mainly rely on temperature sensors installed inside or near the server unit. These sensors can monitor the temperature changes of the server in real time, but their monitoring range and accuracy are limited, and it is difficult to fully reflect the temperature distribution on the server surface. In addition, the failure and error of the sensor itself may also lead to inaccurate temperature monitoring data, increasing the difficulty of fault diagnosis and handling. Summary of the invention

[0004] The present application provides a big data computer room monitoring method, device, equipment and storage medium, which are used to improve the accuracy of temperature anomaly monitoring of server units in a big data computer room.

[0005] In a first aspect, the present application provides a big data computer room monitoring method, the big data computer room monitoring method comprising:

[0006] Scan and pre-process the infrared images of the server unit through the preset monitoring equipment to obtain multiple de-noised infrared images;

[0007] Performing overlapping area recognition and image stitching on the multiple denoised infrared images to obtain a stitched infrared image;

[0008] Constructing an edge perception factor and a detail adjustment factor of the spliced ​​infrared image, and generating a corresponding target infrared image;

[0009] Extracting pixel temperature from the target infrared image to obtain pixel temperature data, and constructing a surface temperature field of the server unit;

[0010] Extracting features of the surface temperature field to obtain a target temperature feature set;

[0011] The target temperature feature set is input into a preset unit abnormality discrimination model to perform server unit abnormality discrimination to obtain a target abnormality discrimination result.

[0012] In a second aspect, the present application provides a big data computer room monitoring device, the big data computer room monitoring device comprising:

[0013] A scanning module is used to scan and pre-process the infrared images of the server unit through a preset monitoring device to obtain multiple de-noised infrared images;

[0014] An identification module, used for performing overlapping area identification and image stitching on the multiple denoised infrared images to obtain a stitched infrared image;

[0015] A construction module, used to construct an edge perception factor and a detail adjustment factor of the spliced ​​infrared image, and generate a corresponding target infrared image;

[0016] A construction module is used to extract pixel temperature from the target infrared image, obtain pixel temperature data, and construct a surface temperature field of the server unit;

[0017] An extraction module, used for extracting features of the surface temperature field to obtain a target temperature feature set;

[0018] The discrimination module is used to input the target temperature feature set into a preset unit abnormality discrimination model to perform server unit abnormality discrimination to obtain a target abnormality discrimination result.

[0019] The third aspect of the present application provides a big data computer room monitoring device, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory so that the big data computer room monitoring device executes the above-mentioned big data computer room monitoring method.

[0020] A fourth aspect of the present application provides a computer-readable storage medium, wherein the computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the above-mentioned big data computer room monitoring method.

[0021] In the technical solution provided by the present application, infrared image scanning and preprocessing are performed by pre-set monitoring equipment, and the initial infrared image is subjected to denoising, median filtering and Gaussian filtering, which can effectively eliminate image noise and improve image quality. When overlapping area recognition and image splicing are performed on the denoised infrared image, high-precision image splicing is achieved. By constructing edge perception factors and detail adjustment factors, and using detail enhancement guided filters for processing, the generated target infrared image is clearer in details and overall structure, and can accurately reflect the temperature distribution on the server surface. By calculating the temperature gradient of the target infrared image, the surface temperature field of the server unit is constructed, which fully reflects the characteristic information of the temperature field. The preset unit abnormality discrimination model is used, and the target temperature feature set is combined with the spatiotemporal graph convolution network and the support vector machine to discriminate the server unit abnormality. By analyzing the multi-model discrimination results through weighted voting, abnormal situations can be accurately identified, and the accuracy of temperature abnormality monitoring of server units in big data computer rooms is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0023] Figure 1 This is a schematic diagram of an embodiment of a big data room monitoring method in an embodiment of the present application;

[0024] Figure 2 This is a schematic diagram of an embodiment of a big data computer room monitoring device in an embodiment of the present application. DETAILED DESCRIPTION

[0025] The embodiments of the present application provide a method, device, equipment and storage medium for monitoring a big data computer room. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0026] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 , an embodiment of the big data room monitoring method in the embodiment of the present application includes:

[0027] Step S101, performing infrared image scanning and preprocessing on a server unit through a preset monitoring device to obtain a plurality of de-noised infrared images;

[0028] It is understandable that the execution subject of the present application may be a big data room monitoring device, or a terminal or a server, which is not specifically limited here. The present application embodiment is described by taking a server as the execution subject as an example.

[0029] Specifically, the operation video of the server unit is recorded by the preset monitoring equipment to obtain the operation video data. The operation video data is matched with the operation task association to identify and extract the video data associated with the operation task. After the monitoring operation task is completed, the associated video data is segmented into a multi-angle infrared image frame sequence, and the infrared image frames of each angle in the video are extracted to obtain multiple initial infrared images. Noise detection is performed on multiple initial infrared images, and the noise area in the image is identified by a specific algorithm, and noise detection data is generated. A 3x3 filter window is used to perform median filtering on the initial infrared image according to the noise detection data to remove salt and pepper noise and other burst noises, and obtain multiple median filtered images. Median filtering can effectively retain the edge information of the image and make the image smoother. Using a Gaussian kernel with a standard deviation of 1, Gaussian filtering is performed on the median filtered image to remove Gaussian noise in the image, and multiple Gaussian filtered images are obtained. Gaussian filtering reduces random noise in the image through smoothing processing and improves the quality of the image. Perform histogram equalization on the Gaussian filter image to enhance the contrast of the image, make the grayscale distribution of the image more uniform, and obtain multiple enhanced infrared images. Perform brightness adjustment on multiple enhanced infrared images, and adjust the brightness of the image to maintain good visual effects under different lighting conditions, and obtain multiple brightness-adjusted images. Use the contrast stretching algorithm to adjust the contrast of the brightness-adjusted image, improve the contrast of the image, make the details of the image clearer, and obtain multiple denoised infrared images. Contrast stretching makes the image's detail information richer and enhances the visual effect by stretching the grayscale range of the image.

[0030] Step S102, performing overlapping area recognition and image stitching on a plurality of denoised infrared images to obtain a stitched infrared image;

[0031] Specifically, the modified scale-invariant feature transformation algorithm is used to detect feature points of multiple denoised infrared images to obtain preliminary feature point data. By detecting key points in the image, the significant features of the image are extracted. The preliminary feature point data is corrected to improve the accuracy of feature point detection and obtain corrected feature point data. The correction process of feature point data includes optimizing the position and number of feature points, so that subsequent calculations are more accurate. The modified normalized cross-correlation algorithm is used to perform normalized cross-correlation calculation on the corrected feature point data to obtain overlapping area data. The normalized cross-correlation algorithm identifies the overlapping area of ​​the image by calculating the similarity between images. Feature points are extracted from the overlapping area data to improve the accuracy of feature points and obtain precise feature point data. The SIFT descriptor algorithm is used to calculate feature descriptors on the precise feature point data to obtain feature descriptor data. Feature descriptors are mathematical representations of feature points, and image matching can be performed more accurately through feature descriptors. The nearest neighbor matching algorithm is used to perform feature matching on the feature descriptor data to obtain initial matching feature point pairs. Compare feature descriptors to find similar feature point pairs, so as to achieve registration between images. Use the RANSAC (Random Sampling Consensus) algorithm to remove mismatches from the initial matching feature point pairs and obtain the final matching feature point pairs. The RANSAC algorithm effectively removes mismatched feature point pairs by iteratively selecting random subsets and evaluating their consistency, thereby improving the accuracy of matching. Perform affine transformation calculation on the final matching feature point pairs to obtain the transformation matrix. By calculating the transformation matrix, the feature points of one image are aligned with the feature points of another image to achieve image registration. According to the transformation matrix, multiple denoised infrared images are registered to obtain multiple registered infrared images. Use the weighted fusion algorithm to stitch multiple registered infrared images to obtain a stitched infrared image. By weighted averaging the pixels in the overlapping area, the edge effects and seam problems that occur during the stitching process are reduced to obtain a complete stitched infrared image.

[0032] Step S103, constructing an edge perception factor and a detail adjustment factor of the spliced ​​infrared image, and generating a corresponding target infrared image;

[0033] Specifically, the detail enhancement guided filter is initialized, and the radius and regularization parameter of the detail enhancement guided filter are set. The Sobel operator is used to perform edge detection on the spliced ​​infrared image to obtain the edge perception factor. The Sobel operator can effectively detect the edge in the image by calculating the gradient of the image and generate the edge perception factor. The edge perception factor is gradient calculated to calculate the gradient amplitude and direction of the edge in the spliced ​​infrared image to obtain the gradient image. The gradient image is constructed by detail adjustment factor, and the detail adjustment factor is obtained by analyzing and processing the gradient image to highlight the detail part. The detail adjustment factor is guided filtered, and the detail adjustment factor is introduced into the cost function of the detail enhancement guided filter to obtain the detail enhanced image. Guided filtering is an image processing technology that effectively enhances the detail part in the image by using the detail adjustment factor, making the image sharper and clearer. The detail layer is extracted from the detail enhanced image to obtain an enhanced detail layer image. At the same time, the base layer is extracted from the detail enhanced image to obtain the base layer image. The purpose of detail layer extraction and base layer extraction is to separate the detail information and base information in the image for subsequent processing and fusion. The enhanced detail layer image and the base layer image are weighted fused to generate the corresponding target infrared image.

[0034] Step S104, extracting pixel temperature from the target infrared image to obtain pixel temperature data, and constructing a surface temperature field of the server unit;

[0035] Specifically, linear mapping is used to extract the temperature of each pixel of the target infrared image to obtain preliminary temperature data. Linear mapping converts the pixel value in the infrared image into the corresponding temperature value through a predetermined formula. The preliminary temperature data is linearly corrected using a correction coefficient to obtain corrected temperature data. The correction coefficient is pre-set according to the specific characteristics of the infrared camera and environmental conditions. The accuracy of the temperature data is improved through the correction coefficient to ensure the accuracy of temperature extraction. Bilinear interpolation is performed on the corrected temperature data to obtain interpolated temperature data. Bilinear interpolation is a commonly used image processing technology. Through interpolation, the resolution of the temperature data is increased to make the temperature field smoother and more detailed. The finite difference method is used to calculate the gradient of the interpolated temperature data to obtain temperature gradient data. The finite difference method extracts the gradient information in the temperature field by calculating the rate of change of the temperature data. The inverse tangent function is used to calculate the gradient direction of the temperature gradient data to obtain the gradient direction data. The gradient direction calculation determines the direction of the temperature gradient through the inverse tangent function, so that the change trend in the temperature field is clearer. The temperature field is constructed for the interpolated temperature data and the temperature gradient data to obtain the surface temperature field of the server unit. By comprehensively interpolating temperature data and gradient data, the temperature distribution is intuitively displayed to form a complete temperature field image.

[0036] Step S105, extracting features of the surface temperature field to obtain a target temperature feature set;

[0037] Specifically, pixel frequency statistics are performed on the surface temperature field to obtain pixel statistics. The frequency distribution of the temperature field is obtained by calculating the number of times each temperature value appears in the entire temperature field. Average temperature calculation is performed on pixel statistics to obtain average temperature data. The average temperature of the entire temperature field is obtained by averaging all pixel temperature values, thereby reflecting the overall temperature level of the temperature field. Maximum temperature calculation is performed on pixel statistics to obtain maximum temperature data, which helps to identify possible hot spots and abnormally high temperature areas. Minimum temperature calculation is performed on pixel statistics to obtain minimum temperature data, which helps to understand the range of temperature distribution. Temperature standard deviation calculation is performed on pixel statistics to obtain temperature standard deviation data. Temperature standard deviation is an important indicator for measuring the uniformity of temperature distribution and reflects the degree of discreteness of temperature values ​​in the temperature field. K-means algorithm is used to perform cluster analysis on pixel statistics to obtain temperature hot spot area data. By clustering pixels with similar temperature values ​​together, temperature hot spot areas are effectively identified. These hot spot areas are areas with higher temperatures and may be potential failure points or areas that require special attention. Density calculation is performed on temperature hot spot area data to obtain hot spot density data. By counting the density of temperature values ​​in each hotspot area, the temperature concentration of the hotspot area is measured, which helps to assess the severity of the hotspot area. The average temperature data, maximum temperature data, minimum temperature data, temperature standard deviation data, and hotspot density data are transformed into a set to obtain the target temperature feature set. The statistical data are integrated together to form a comprehensive temperature feature set.

[0038] Step S106: input the target temperature feature set into a preset unit abnormality discrimination model to perform server unit abnormality discrimination to obtain a target abnormality discrimination result.

[0039] Specifically, the target temperature feature set is input into a preset unit abnormality discrimination model, which includes an embedding layer, a spatiotemporal graph convolutional network, a first support vector machine, and a second support vector machine. The input target temperature feature set is feature encoded by the embedding layer to obtain a temperature feature encoding vector. The high-dimensional input data is mapped to a low-dimensional space, so that the data is more concise and efficient in subsequent processing. The temperature feature encoding vector is extracted by the spatiotemporal graph convolutional network to obtain a temperature spatiotemporal feature vector. The spatiotemporal graph convolutional network is a neural network that processes spatiotemporal data and can simultaneously capture the spatial structure and time variation characteristics of the data. In temperature monitoring, spatiotemporal feature extraction can better reflect the change of temperature over time and space, thereby improving the accuracy of abnormality discrimination. The temperature spatiotemporal feature vector is used by the first support vector machine to discriminate the server unit abnormality, and a first abnormality discrimination result is obtained. The support vector machine is a commonly used classification algorithm that classifies data as normal or abnormal by finding the best separating hyperplane. The first support vector machine analyzes the temperature spatiotemporal features to preliminarily determine whether the server unit is abnormal. At the same time, the second support vector machine is used to identify the abnormality of the server unit on the temperature spatiotemporal feature vector to obtain the second abnormality identification result. The temperature characteristics are analyzed again by an independent model to provide a second identification result. The first abnormality identification result and the second abnormality identification result are subjected to weighted voting analysis to obtain the target abnormality identification result. The final identification result is obtained by combining the identification results of the two independent models and weighting the results according to the preset weights.

[0040] In the embodiment of the present application, infrared image scanning and preprocessing are performed by a preset monitoring device, and the initial infrared image is subjected to denoising, median filtering and Gaussian filtering, which can effectively eliminate image noise and improve image quality. When the denoised infrared image is subjected to overlapping area recognition and image splicing, high-precision image splicing is achieved. By constructing edge perception factors and detail adjustment factors, and using detail enhancement guided filters for processing, the generated target infrared image is clearer in details and overall structure, and can accurately reflect the temperature distribution on the server surface. By calculating the temperature gradient of the target infrared image, the surface temperature field of the server unit is constructed, which fully reflects the characteristic information of the temperature field, and the preset unit abnormality discrimination model is used, combined with the spatiotemporal graph convolution network and the support vector machine to discriminate the server unit abnormality on the target temperature feature set. By analyzing the multi-model discrimination results through weighted voting, abnormal situations can be accurately identified, and the accuracy of temperature abnormality monitoring of server units in big data computer rooms is improved.

[0041] In a specific embodiment, the process of executing step S101 may specifically include the following steps:

[0042] (1) Record the operation video of the server unit through the preset monitoring equipment to obtain the operation video data;

[0043] (2) Performing operation task association matching on the operation video data to obtain associated video data;

[0044] (3) When the monitoring operation task is completed, the associated video data is segmented into a multi-angle infrared image frame sequence to obtain multiple initial infrared images;

[0045] (4) Performing noise detection on multiple initial infrared images to obtain noise detection data;

[0046] (5) Using a 3x3 filter window, median filtering is performed on multiple initial infrared images according to the noise detection data to obtain multiple median filtered images;

[0047] (6) Using a Gaussian kernel with a standard deviation of 1, Gaussian filtering is performed on multiple median filtered images to obtain multiple Gaussian filtered images;

[0048] (7) Performing histogram equalization on multiple Gaussian filter images to obtain multiple enhanced infrared images;

[0049] (8) performing brightness adjustment on the multiple enhanced infrared images to obtain multiple brightness adjusted images;

[0050] (9) Use a contrast stretching algorithm to adjust the contrast of multiple brightness-adjusted images to obtain multiple denoised infrared images.

[0051] Specifically, the operation video of the server unit is recorded by the preset monitoring device to obtain the operation video data. The monitoring device can be a fixed camera or a mobile device installed in the computer room to capture the whole video of the server unit during operation. The operation video data is matched with the operation task to obtain the associated video data. The operator's action is associated with the specific operation task through the video analysis technology, and part of the video data related to the operation task is extracted. This can be achieved through the action recognition algorithm, for example, using a deep learning model to recognize the gestures and actions of the operator, and matching them with the predefined operation task to filter out the associated video data. When the monitoring operation task is completed, the associated video data is segmented into a multi-angle infrared image frame sequence to obtain multiple initial infrared images. The multi-angle infrared image frame sequence segmentation helps to fully capture the thermal state of the device and ensure the richness and integrity of the image data. Noise detection is performed on multiple initial infrared images to obtain noise detection data. Noise detection is to identify the noise area by analyzing the change of pixel values ​​in the image, and a noise detection algorithm such as median absolute deviation or local variance analysis can be used. Using a 3x3 filter window, median filtering is performed on multiple initial infrared images according to the noise detection data to obtain multiple median filtered images. Median filtering is a commonly used image denoising technology. By replacing the value of each pixel with the median of the pixel values ​​in its neighborhood, salt and pepper noise can be effectively removed. The median filtering formula can be expressed as:

[0052] ;

[0053] in, is the image pixel value after median filtering, is the initial image in Pixel values ​​in the neighborhood. Use a Gaussian kernel with a standard deviation of 1 to perform Gaussian filtering on multiple median filtered images to obtain multiple Gaussian filtered images. Gaussian filtering is achieved through convolution operations to smooth the image and reduce high-frequency noise. The Gaussian filtering formula is:

[0054] ;

[0055] in, is the value of the Gaussian kernel, is the standard deviation, which determines the smoothness of the filter. Using a Gaussian kernel with a standard deviation of 1 can effectively smooth noise while retaining image details. Perform histogram equalization on multiple Gaussian filter images to obtain multiple enhanced infrared images. Histogram equalization stretches the gray value distribution of the image to enhance the contrast of the image and make the details clearer. The histogram equalization formula is:

[0056] ;

[0057] in, is the pixel value of the image after equalization, is the number of gray levels, is the total number of image pixels, Is grayscale The pixel frequency of the image is . The brightness of multiple enhanced infrared images is adjusted to obtain multiple brightness adjusted images. By adding and subtracting each pixel value of the image, the overall brightness of the image is adjusted so that it can maintain a good visual effect under different lighting conditions. The contrast of multiple brightness adjusted images is adjusted using the contrast stretching algorithm to obtain multiple denoised infrared images. Contrast stretching expands the grayscale range of the image through linear transformation, making the image contrast higher and the details more obvious. The contrast stretching formula is:

[0058] ;

[0059] in, is the image pixel value after contrast stretching, and are the minimum and maximum pixel values ​​of the original image, respectively. and is the grayscale range of the stretched image.

[0060] In a specific embodiment, the process of executing step S102 may specifically include the following steps:

[0061] (1) Using the modified scale-invariant feature transformation algorithm, feature point detection is performed on multiple denoised infrared images to obtain preliminary feature point data;

[0062] (2) Correcting the preliminary feature point data to obtain corrected feature point data;

[0063] (3) Using the modified normalized cross-correlation algorithm, normalized cross-correlation calculation is performed on the modified feature point data to obtain the overlapping area data;

[0064] (4) Extract feature points from overlapping area data to obtain accurate feature point data;

[0065] (5) Using the SIFT descriptor algorithm, calculate the feature descriptor of the precise feature point data to obtain feature descriptor data;

[0066] (6) Using the nearest neighbor matching algorithm, feature matching is performed on the feature descriptor data to obtain the initial matching feature point pairs;

[0067] (7) Use the RANSAC algorithm to eliminate mismatches in the initial matching feature point pairs and obtain the final matching feature point pairs;

[0068] (8) Perform affine transformation calculation on the final matching feature point pair to obtain the transformation matrix;

[0069] (9) performing image registration on multiple denoised infrared images according to the transformation matrix to obtain multiple registered infrared images;

[0070] (10) Use a weighted fusion algorithm to stitch multiple registered infrared images to obtain a stitched infrared image.

[0071] Specifically, a modified scale-invariant feature transformation algorithm is used to detect feature points of multiple denoised infrared images to obtain preliminary feature point data. The scale-invariant feature transformation algorithm is a commonly used computer vision algorithm that can extract feature points in an image while maintaining scale and rotation invariance. By performing Gaussian blur processing on the image and calculating the differential Gaussian image, the key points in the image can be effectively detected to obtain preliminary feature point data. The preliminary feature point data is corrected to obtain corrected feature point data. Feature point correction includes fine-tuning the position of the feature points to eliminate errors in the detection process. By analyzing the neighborhood pixel values ​​of the feature points, it is ensured that the feature points more accurately reflect the significant features in the image. The modified normalized cross-correlation algorithm is used to perform normalized cross-correlation calculation on the corrected feature point data to obtain overlapping area data. The normalized cross-correlation algorithm identifies the overlapping parts of the image by calculating the similarity between local areas of the image. The formula is as follows:

[0072] ;

[0073] in, and Represent the pixel values ​​at corresponding positions in the two images, and are the average pixel values ​​of the two images, and is the standard deviation of the two images, is the number of pixels in the image. Through normalized cross-correlation calculation, the overlapping area data in the image is identified and extracted. Feature points are extracted from the overlapping area data to obtain precise feature point data. The SIFT descriptor algorithm is used to calculate the feature descriptor for the precise feature point data to obtain feature descriptor data. The SIFT descriptor is a vector used to describe the neighborhood information of a feature point. The feature descriptor is generated by calculating the direction and amplitude of the image gradient. The nearest neighbor matching algorithm is used to perform feature matching on the feature descriptor data to obtain an initial matching feature point pair. The nearest neighbor matching algorithm calculates the Euclidean distance between feature descriptors and finds a pair of feature points with the smallest distance as a matching pair. The formula is as follows:

[0074] ;

[0075] in, Represents two feature descriptors and The Euclidean distance between them is used to preliminarily determine the corresponding feature point pairs in the two images. The Random Sampling Consensus (RANSAC) algorithm is used to eliminate the mismatches of the initial matching feature point pairs to obtain the final matching feature point pairs. The RANSAC algorithm iteratively selects random sample sets and calculates model parameters, evaluates the number of inliers in each sample set, and finally determines the model parameters with the largest inlier set to eliminate the mismatched feature point pairs. The formula is as follows:

[0076] ;

[0077] in, Indicates the error, and Respectively represent the coordinates of the matching point pairs, is the transformation matrix. The RANSAC algorithm can effectively remove mismatched point pairs and improve matching accuracy. The final matching feature point pairs are affine transformed to obtain the transformation matrix. Affine transformation is achieved through a set of linear transformations and translations. The formula is as follows:

[0078] ;

[0079] in, and Represent the coordinates of the feature points in the original image and the target image respectively, is the linear transformation parameter, is the translation parameter. Through affine transformation, multiple images are geometrically aligned to achieve image registration. Multiple denoised infrared images are registered according to the transformation matrix to obtain multiple registered infrared images. Images from different perspectives are aligned so that they are superimposed in the same coordinate system. Using the weighted fusion algorithm, multiple registered infrared images are stitched to obtain a stitched infrared image. The weighted fusion algorithm reduces the seams and edge effects in the stitching process by weighted averaging the pixels in the overlapping area, making the stitching result more natural. The formula is as follows:

[0080] ;

[0081] in, is the pixel value of the stitched image, It is The image is The weight of the position, It is The image is The pixel values ​​at the respective positions are weightedly fused to obtain a high-quality stitched infrared image.

[0082] In a specific embodiment, the process of executing step S103 may specifically include the following steps:

[0083] (1) Initialize the detail enhancement guided filter and set the radius and regularization parameter of the detail enhancement guided filter;

[0084] (2) Use the Sobel operator to perform edge detection on the spliced ​​infrared image and obtain the edge perception factor;

[0085] (3) Performing gradient calculation on the edge perception factor, calculating the gradient amplitude and direction of the edge in the spliced ​​infrared image, and obtaining a gradient image;

[0086] (4) Constructing detail adjustment factors for the gradient image to obtain detail adjustment factors;

[0087] (5) Guide filtering the detail adjustment factor and introduce the detail adjustment factor into the cost function of the detail enhancement guided filter to obtain a detail enhanced image;

[0088] (6) performing detail layer extraction on the detail enhanced image to obtain an enhanced detail layer image, and performing base layer extraction on the detail enhanced image to obtain a base layer image;

[0089] (7) Perform weighted fusion on the enhanced detail layer image and the base layer image to generate the corresponding target infrared image.

[0090] Specifically, initialize the detail enhancement guided filter, and set the radius and regularization parameter of the detail enhancement guided filter. The function of the detail enhancement guided filter is to enhance the details of the image. By setting the appropriate radius and regularization parameter, ensure that the filter removes noise while retaining the image details. The radius of the filter determines the filtering range, while the regularization parameter controls the balance between smoothing and detail enhancement. Use the Sobel operator to perform edge detection on the spliced ​​infrared image and obtain the edge perception factor. The Sobel operator is an edge detection algorithm that detects the edges in the image by calculating the gradient of the image. The Sobel operator calculates the gradients in the horizontal and vertical directions respectively, and the formula is as follows:

[0091] ;

[0092] ;

[0093] in, and are the gradients in the horizontal and vertical directions respectively. Represents the convolution operation. By calculating and , and get the edge perception factor. Perform gradient calculation on the edge perception factor, calculate the gradient amplitude and direction of the edge in the spliced ​​infrared image, and then get the gradient image. Gradient amplitude and the gradient direction The formula is as follows:

[0094] ;

[0095] ;

[0096] in, is the gradient magnitude, which reflects the strength of the edge; is the gradient direction, indicating the direction of the edge. By calculating the gradient amplitude and direction, the edge features in the image are analyzed. The detail adjustment factor is constructed on the gradient image to obtain the detail adjustment factor. By processing the gradient image, the detail information in the image is extracted so that the details can be enhanced in the subsequent processing. The detail adjustment factor is guided filtered, and the detail adjustment factor is introduced into the cost function of the detail enhancement guided filter to obtain a detail enhanced image. Guided filtering is an edge-preserving smoothing technology that retains detail information while smoothing the image through the detail adjustment factor. The formula for guided filtering is as follows:

[0097] ;

[0098] in, is the pixel value of the filtered image, is the filter window, is the regularization parameter, is the detail adjustment factor, which balances smoothing and detail enhancement through the cost function. Detail layer extraction is performed on the detail enhanced image to obtain an enhanced detail layer image. At the same time, base layer extraction is performed on the detail enhanced image to obtain a base layer image. Detail layer extraction is to extract the detail information of the image by analyzing the high-frequency components in the image; base layer extraction is to extract the basic information of the image by analyzing the low-frequency components in the image. The enhanced detail layer image and the base layer image are weightedly fused to generate the corresponding target infrared image. By weighted averaging the detail layer and the base layer, the final image contains rich details and has a good overall effect.

[0099] In a specific embodiment, the process of executing step S104 may specifically include the following steps:

[0100] (1) Use linear mapping to extract the temperature of each pixel in the target infrared image to obtain preliminary temperature data;

[0101] (2) Using the correction coefficient, perform linear correction on the preliminary temperature data to obtain corrected temperature data;

[0102] (3) Perform bilinear interpolation on the corrected temperature data to obtain interpolated temperature data;

[0103] (4) Using the finite difference method, the gradient of the interpolated temperature data is calculated to obtain the temperature gradient data;

[0104] (5) Using the inverse tangent function, calculate the gradient direction of the temperature gradient data to obtain the gradient direction data;

[0105] (6) The temperature field is constructed based on the interpolated temperature data and temperature gradient data to obtain the surface temperature field of the server unit.

[0106] Specifically, linear mapping is used to extract the temperature of each pixel of the target infrared image to obtain preliminary temperature data. The formula of linear mapping is as follows:

[0107] ;

[0108] in, is the preliminary temperature data, is the pixel in the target infrared image The gray value of and are the proportional coefficient and offset coefficient of the linear mapping respectively. These coefficients can be obtained through a pre-calibration experiment. Through linear mapping, the grayscale value in the infrared image is converted into the corresponding temperature value to obtain preliminary temperature data. The preliminary temperature data is linearly corrected using the correction coefficient to obtain the corrected temperature data. The formula for linear correction is as follows:

[0109] ;

[0110] in, is the corrected temperature data, and They are the proportional coefficient and offset coefficient of the correction respectively. The correction coefficient is pre-set according to the specific characteristics of the infrared camera and environmental conditions. Through correction, the accuracy of the temperature data is improved and the accuracy of temperature extraction is ensured. Bilinear interpolation is performed on the corrected temperature data to obtain interpolated temperature data. Bilinear interpolation increases the resolution of the data by weighted averaging of adjacent pixels, making the temperature field smoother and more detailed. The finite difference method is used to calculate the gradient of the interpolated temperature data to obtain the temperature gradient data. The finite difference method extracts the gradient information in the temperature field by calculating the rate of change of the temperature data. The gradient calculation formula is as follows:

[0111] ;

[0112] Among them, the transverse component and the longitudinal component of the gradient are expressed as:

[0113] ;

[0114] ;

[0115] Through these formulas, the temperature gradient data is calculated to reflect the rate of change of temperature in each direction. The inverse tangent function is used to calculate the gradient direction of the temperature gradient data to obtain the gradient direction data. The gradient direction calculation formula is as follows:

[0116] ;

[0117] in, Represents gradient direction data. By calculating the gradient direction, the direction of the temperature gradient is determined, making the change trend in the temperature field clearer. The temperature field is constructed for the interpolated temperature data and temperature gradient data to obtain the surface temperature field of the server unit. By comprehensively interpolating the temperature data and gradient data, the temperature distribution is intuitively displayed to form a complete temperature field image. The temperature field construction formula is as follows:

[0118] ;

[0119] in, is the final temperature field data, Is a balancing factor that controls the combined ratio of interpolated temperature data and gradient data. For example, suppose a server group in a large data center is monitored and temperature data is extracted from images recorded by an infrared camera. The grayscale values ​​in the infrared image are converted into preliminary temperature data by linear mapping. Assume that the mapping coefficient and They are 0.02 and 30 respectively, which means that the temperature corresponding to each gray value is the gray value multiplied by 0.02 plus 30. Use the correction factor and The initial temperature data are corrected to improve the accuracy of the temperature data. The corrected temperature data are bilinearly interpolated to increase the resolution of the data. Assume that at a certain point , the temperature values ​​of the four adjacent pixels are , , the interpolation temperature data of the point can be obtained by interpolation calculation. The gradient of the interpolation temperature data is calculated using the finite difference method. Assume that at point The interpolated temperature values ​​are and , the gradient component is calculated by formula. The inverse tangent function is used to calculate the gradient direction and determine the change trend in the temperature field. By comprehensively interpolating temperature data and gradient data, a complete temperature field is constructed to intuitively display the temperature distribution on the surface of the server unit.

[0120] In a specific embodiment, the process of executing step S105 may specifically include the following steps:

[0121] (1) Perform pixel frequency statistics on the surface temperature field to obtain pixel statistical data;

[0122] (2) Calculate the average temperature of pixel statistical data to obtain average temperature data;

[0123] (3) Calculate the maximum temperature of pixel statistical data to obtain maximum temperature data;

[0124] (4) Calculate the minimum temperature of pixel statistical data to obtain minimum temperature data;

[0125] (5) Calculate the temperature standard deviation of pixel statistical data to obtain temperature standard deviation data;

[0126] (6) Use the K-means algorithm to perform cluster analysis on pixel statistical data to obtain temperature hotspot area data, and perform density calculation on the temperature hotspot area data to obtain hotspot density data;

[0127] (7) Perform set conversion on the average temperature data, maximum temperature data, minimum temperature data, temperature standard deviation data, and hot spot density data to obtain the target temperature feature set.

[0128] Specifically, pixel frequency statistics are performed on the surface temperature field to obtain pixel statistics. Pixel frequency statistics obtain the distribution of the temperature field by calculating the frequency of each temperature value in the entire temperature field. The average temperature is calculated for the pixel statistics to obtain the average temperature data and the overall temperature level of the entire temperature field. The maximum temperature is calculated for the pixel statistics to obtain the maximum temperature data. The maximum temperature reflects the highest temperature point in the temperature field, which is of great significance for identifying potential hot spots. The minimum temperature is calculated for the pixel statistics to obtain the minimum temperature data. The minimum temperature reflects the lowest temperature point in the temperature field, which helps to fully understand the range of temperature distribution. The temperature standard deviation is calculated for the pixel statistics to obtain the temperature standard deviation data. The temperature standard deviation measures the degree of discreteness of the temperature distribution and reflects the fluctuation of temperature changes. The K-means algorithm is used to perform cluster analysis on the pixel statistics to obtain the temperature hot spot area data. The K-means algorithm achieves data clustering by iteratively assigning data points to the nearest cluster center and updating the position of the cluster center. The objective function of the Kmeans algorithm is as follows:

[0129] ;

[0130] in, is the objective function, is the number of clusters, It belongs to The ith temperature value of the cluster, It is The centers of the clusters are minimized by , effectively clustering pixel statistics to identify temperature hotspots. In cluster analysis, density calculation is performed on the temperature hotspot area data to obtain hotspot density data. Hotspot density calculation measures the temperature concentration of hotspot areas by counting the density of temperature values ​​in each hotspot area. The average temperature data, maximum temperature data, minimum temperature data, temperature standard deviation data, and hotspot density data are transformed into a set to obtain the target temperature feature set. Set transformation is to integrate various statistical data together to form a comprehensive temperature feature set. For example, suppose a server unit in a large data center is monitored. The temperature field data on the surface of the server unit is obtained through an infrared camera Perform pixel frequency statistics on the temperature field. Assuming that the temperature value ranges from 30°C to 70°C, construct a temperature frequency histogram by counting the frequency of occurrence of each temperature value. Calculate the average temperature of the temperature field. Assuming that the temperature data is [35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45], the average temperature is calculated by the formula °C. Calculate the maximum and minimum temperatures of the temperature field, assuming the maximum temperature The minimum temperature is 45°C is 35°C. Calculate the temperature standard deviation of the temperature field, assuming that the temperature data is [35, 36, 37, 38, 39, 40, 41, 42, 43, 44, 45], the temperature standard deviation is calculated by the formula .32°C. Use K-means algorithm to cluster the temperature data. Assume that the temperature data is divided into two clusters, representing the normal area and the hot spot area. The K-means algorithm can effectively identify the temperature hot spot area. Perform density calculation on the temperature hot spot area. Assume that the number of pixels in a hot spot area is 100, area For 50 square pixels, the hot spot density is calculated by the formula All statistical data are combined to form a target temperature feature set.

[0131] In a specific embodiment, the process of executing step S106 may specifically include the following steps:

[0132] (1) Inputting the target temperature feature set into a preset unit abnormality discrimination model, the unit abnormality discrimination model includes an embedding layer, a spatiotemporal graph convolutional network, a first support vector machine, and a second support vector machine;

[0133] (2) In the embedding layer, the target temperature feature set is feature encoded to obtain a temperature feature encoding vector;

[0134] (3) Extract the spatiotemporal features of the temperature feature encoding vector through the spatiotemporal graph convolutional network to obtain the temperature spatiotemporal feature vector;

[0135] (4) Using a first support vector machine to perform abnormality determination on the temperature spatiotemporal feature vector of the server unit, thereby obtaining a first abnormality determination result;

[0136] (5) Using a second support vector machine to perform abnormality identification on the server unit based on the temperature spatiotemporal feature vector, a second abnormality identification result is obtained;

[0137] (6) Perform weighted voting analysis on the first anomaly discrimination result and the second anomaly discrimination result to obtain the target anomaly discrimination result.

[0138] Specifically, the target temperature feature set is input into the preset unit abnormality discrimination model. The model includes an embedding layer, a spatiotemporal graph convolutional network, a first support vector machine, and a second support vector machine. The target temperature feature set contains feature data such as average temperature, maximum temperature, minimum temperature, temperature standard deviation, and hotspot density, which can fully reflect the temperature distribution and changes of the server unit. In the embedding layer, the target temperature feature set is feature encoded to obtain a temperature feature encoding vector. The embedding layer maps the high-dimensional input feature set to a low-dimensional embedding space, making the data more compact and efficient. The feature encoding formula is as follows:

[0139] ;

[0140] in, represents the temperature feature encoding vector, is the weight matrix of the embedding layer, is the target temperature feature set, is the bias vector. Through this formula, the embedding layer can convert the input high-dimensional feature data into a low-dimensional encoding vector The spatiotemporal feature vector of the temperature feature encoding vector is extracted through the spatiotemporal graph convolution network to obtain the temperature spatiotemporal feature vector. The spatiotemporal graph convolution network can capture the spatial structure and temporal variation characteristics of the data at the same time. The formula is as follows:

[0141] ;

[0142] in, Indicates The output feature matrix of the layer, is the activation function, It is adjacency matrix, is the node degree matrix, It is The input feature matrix of the layer, It is The weight matrix of the convolution kernel. Through this formula, the spatiotemporal graph convolution network can extract the complex relationship between temperature features in time and space and obtain the spatiotemporal feature vector. The first support vector machine is used to identify the abnormality of the server unit on the temperature spatiotemporal feature vector to obtain the first abnormality identification result. Support vector machine is a commonly used classification algorithm that classifies data as normal or abnormal by finding the best separating hyperplane. The decision function of SVM is as follows:

[0143] ;

[0144] in, represents the discriminant function, is the Lagrange multiplier, is the sample label, is the kernel function, is the bias sgn sign function. Through this formula, the support vector machine can classify the temperature spatiotemporal features and output the first abnormality discrimination result. At the same time, the second support vector machine is used to discriminate the abnormality of the server unit on the temperature spatiotemporal feature vector to obtain the second abnormality discrimination result. The working principle of the second support vector machine is the same as that of the first support vector machine, but its training process may use different kernel functions or parameter configurations to provide independent discrimination results. A weighted voting analysis is performed on the first abnormality discrimination result and the second abnormality discrimination result to obtain the target abnormality discrimination result. The weighted voting analysis combines the discrimination results of two independent models and weights the results according to the preset weights to obtain the final discrimination result.

[0145] The above describes the big data room monitoring method in the embodiment of the present application. The following describes the big data room monitoring device in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the big data room monitoring device includes:

[0146] The scanning module 201 is used to scan and pre-process the infrared image of the server group through a preset monitoring device to obtain a plurality of de-noised infrared images;

[0147] The recognition module 202 is used to identify overlapping areas and stitch multiple denoised infrared images to obtain a stitched infrared image;

[0148] A construction module 203 is used to construct an edge perception factor and a detail adjustment factor of the spliced ​​infrared image and generate a corresponding target infrared image;

[0149] A construction module 204 is used to extract pixel temperature from the target infrared image, obtain pixel temperature data, and construct a surface temperature field of the server unit;

[0150] An extraction module 205 is used to extract features of the surface temperature field to obtain a target temperature feature set;

[0151] The identification module 206 is used to input the target temperature feature set into a preset unit abnormality identification model to perform server unit abnormality identification to obtain a target abnormality identification result.

[0152] Through the cooperation of the above components, infrared image scanning and preprocessing are performed through the preset monitoring equipment, and the initial infrared image is denoised, median filtered and Gaussian filtered, which can effectively eliminate image noise and improve image quality. When the denoised infrared image is overlapped and stitched, high-precision image stitching is achieved. By constructing edge perception factors and detail adjustment factors and processing with detail enhancement guided filters, the generated target infrared image is clearer in details and overall structure, and can accurately reflect the temperature distribution on the server surface. By calculating the temperature gradient of the target infrared image, the surface temperature field of the server unit is constructed to fully reflect the characteristic information of the temperature field. The preset unit abnormality discrimination model is used, combined with the spatiotemporal graph convolutional network and support vector machine to discriminate the server unit abnormality on the target temperature feature set. By analyzing the multi-model discrimination results through weighted voting, the abnormal situation can be accurately identified, which improves the accuracy of temperature abnormality monitoring of the server unit in the big data computer room.

[0153] The present application also provides a big data computer room monitoring device, which includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the big data computer room monitoring method in the above-mentioned embodiments.

[0154] The present application also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. Instructions are stored in the computer-readable storage medium. When the instructions are executed on a computer, the computer executes the steps of the big data computer room monitoring method.

[0155] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0156] If the integrated unit 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 is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the method described in each embodiment of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk and other media that can store program codes.

[0157] As described above, the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A big data computer room monitoring method, characterized in that: The big data computer room monitoring method comprises: The server group is subjected to infrared image scanning and preprocessing by a preset monitoring device to obtain a plurality of denoised infrared images; specifically comprising: recording the operation video of the server group by a preset monitoring device to obtain operation video data; performing operation task association matching on the operation video data to obtain associated video data; when the monitoring operation task is completed, performing multi-angle infrared image frame sequence segmentation on the associated video data to obtain a plurality of initial infrared images; performing noise detection on the plurality of initial infrared images to obtain noise detection data; using a 3x3 filter window, median filtering is performed on the plurality of initial infrared images according to the noise detection data to obtain a plurality of median filtered images; using a Gaussian kernel with a standard deviation of 1, Gaussian filtering is performed on the plurality of median filtered images to obtain a plurality of Gaussian filtered images; performing histogram equalization on the plurality of Gaussian filtered images to obtain a plurality of enhanced infrared images; performing brightness adjustment on the plurality of enhanced infrared images to obtain a plurality of brightness adjusted images; performing contrast adjustment on the plurality of brightness adjusted images using a contrast stretching algorithm to obtain a plurality of denoised infrared images; The method comprises the following steps: performing overlapping area identification and image stitching on the multiple denoised infrared images to obtain a stitched infrared image; specifically comprising: using a modified scale-invariant feature transformation algorithm to perform feature point detection on the multiple denoised infrared images to obtain preliminary feature point data; correcting the preliminary feature point data to obtain corrected feature point data; using a modified normalized cross-correlation algorithm to perform normalized cross-correlation calculation on the corrected feature point data to obtain overlapping area data; performing feature point extraction on the overlapping area data to obtain precise feature point data; using a SIFT descriptor algorithm to extract the precise feature point data. According to the feature descriptor calculation, feature descriptor data is obtained; the feature descriptor data is feature matched by using the nearest neighbor matching algorithm to obtain an initial matching feature point pair; the initial matching feature point pair is mismatched and eliminated by using the RANSAC algorithm to obtain a final matching feature point pair; the final matching feature point pair is affine transformed and a transformation matrix is ​​obtained; the multiple denoised infrared images are image registered according to the transformation matrix to obtain a multiple registered infrared images; the multiple registered infrared images are image stitched by using the weighted fusion algorithm to obtain a stitched infrared image; Constructing the edge perception factor and detail adjustment factor of the spliced ​​infrared image, and generating the corresponding target infrared image; specifically comprising: initializing the detail enhancement guided filter, setting the radius and regularization parameter of the detail enhancement guided filter; using the Sobel operator to perform edge detection on the spliced ​​infrared image to obtain the edge perception factor; performing gradient calculation on the edge perception factor, calculating the gradient amplitude and direction of the edge in the spliced ​​infrared image, and obtaining a gradient image; constructing the detail adjustment factor on the gradient image to obtain the detail adjustment factor; performing guided filtering on the detail adjustment factor, and introducing the detail adjustment factor into the cost function of the detail enhancement guided filter to obtain a detail enhanced image; performing detail layer extraction on the detail enhanced image to obtain an enhanced detail layer image, and performing base layer extraction on the detail enhanced image to obtain a base layer image; performing weighted fusion on the enhanced detail layer image and the base layer image to generate the corresponding target infrared image; Extracting pixel temperature from the target infrared image to obtain pixel temperature data, and constructing the surface temperature field of the server unit; specifically comprising: using linear mapping to extract temperature from each pixel of the target infrared image to obtain preliminary temperature data; using a correction coefficient to linearly correct the preliminary temperature data to obtain corrected temperature data; performing bilinear interpolation on the corrected temperature data to obtain interpolated temperature data; using a finite difference method to perform gradient calculation on the interpolated temperature data to obtain temperature gradient data; using an inverse tangent function to calculate the gradient direction of the temperature gradient data to obtain gradient direction data; constructing a temperature field for the interpolated temperature data and the temperature gradient data to obtain the surface temperature field of the server unit; Performing feature extraction on the surface temperature field to obtain a target temperature feature set; specifically comprising: performing pixel frequency statistics on the surface temperature field to obtain pixel statistical data; performing average temperature calculation on the pixel statistical data to obtain average temperature data; performing maximum temperature calculation on the pixel statistical data to obtain maximum temperature data; performing minimum temperature calculation on the pixel statistical data to obtain minimum temperature data; performing temperature standard deviation calculation on the pixel statistical data to obtain temperature standard deviation data; using the K-means algorithm to perform cluster analysis on the pixel statistical data to obtain temperature hot spot area data, and performing density calculation on the temperature hot spot area data to obtain hot spot density data; performing set conversion on the average temperature data, the maximum temperature data, the minimum temperature data, the temperature standard deviation data and the hot spot density data to obtain a target temperature feature set; The target temperature feature set is input into a preset unit abnormality discrimination model to perform server unit abnormality discrimination to obtain a target abnormality discrimination result; specifically comprising: inputting the target temperature feature set into a preset unit abnormality discrimination model, the unit abnormality discrimination model comprising an embedding layer, a spatiotemporal graph convolutional network, a first support vector machine and a second support vector machine; in the embedding layer, feature encoding is performed on the target temperature feature set to obtain a temperature feature encoding vector; spatiotemporal feature extraction is performed on the temperature feature encoding vector through the spatiotemporal graph convolutional network to obtain a temperature spatiotemporal feature vector; server unit abnormality discrimination is performed on the temperature spatiotemporal feature vector through the first support vector machine to obtain a first abnormality discrimination result; server unit abnormality discrimination is performed on the temperature spatiotemporal feature vector through the second support vector machine to obtain a second abnormality discrimination result; weighted voting analysis is performed on the first abnormality discrimination result and the second abnormality discrimination result to obtain a target abnormality discrimination result.

2. A big data computer room monitoring device, characterized in that: The big data room monitoring device comprises: The scanning module is used to perform infrared image scanning and preprocessing on the server unit through a preset monitoring device to obtain multiple denoised infrared images; specifically includes: recording the operation video of the server unit through the preset monitoring device to obtain operation video data; performing operation task association matching on the operation video data to obtain associated video data; when the monitoring operation task is completed, performing multi-angle infrared image frame sequence segmentation on the associated video data to obtain multiple initial infrared images; performing noise detection on the multiple initial infrared images to obtain noise detection data; using a 3x3 filter window, median filtering the multiple initial infrared images according to the noise detection data to obtain multiple median filtered images; using a Gaussian kernel with a standard deviation of 1, Gaussian filtering the multiple median filtered images to obtain multiple Gaussian filtered images; performing histogram equalization on the multiple Gaussian filtered images to obtain multiple enhanced infrared images; performing brightness adjustment on the multiple enhanced infrared images to obtain multiple brightness adjusted images; using a contrast stretching algorithm to adjust the contrast of the multiple brightness adjusted images to obtain multiple denoised infrared images; The recognition module is used to identify overlapping areas and stitch the multiple denoised infrared images to obtain a stitched infrared image; specifically, the module includes: using a modified scale-invariant feature transformation algorithm to detect feature points on the multiple denoised infrared images to obtain preliminary feature point data; correcting the preliminary feature point data to obtain corrected feature point data; using a modified normalized cross-correlation algorithm to perform normalized cross-correlation calculation on the corrected feature point data to obtain overlapping area data; extracting feature points from the overlapping area data to obtain precise feature point data; using a SIFT descriptor algorithm to extract the precise feature point data. The feature point data is subjected to feature descriptor calculation to obtain feature descriptor data; a nearest neighbor matching algorithm is used to perform feature matching on the feature descriptor data to obtain an initial matching feature point pair; a RANSAC algorithm is used to eliminate mismatches on the initial matching feature point pair to obtain a final matching feature point pair; an affine transformation is performed on the final matching feature point pair to obtain a transformation matrix; image registration is performed on the multiple denoised infrared images according to the transformation matrix to obtain multiple registered infrared images; a weighted fusion algorithm is used to perform image stitching on the multiple registered infrared images to obtain a stitched infrared image; A construction module is used to construct an edge perception factor and a detail adjustment factor of the spliced ​​infrared image and generate a corresponding target infrared image; specifically comprising: initializing a detail enhancement guided filter and setting the radius and regularization parameter of the detail enhancement guided filter; using a Sobel operator to perform edge detection on the spliced ​​infrared image to obtain an edge perception factor; performing gradient calculation on the edge perception factor to calculate the gradient amplitude and direction of the edge in the spliced ​​infrared image to obtain a gradient image; performing detail adjustment factor construction on the gradient image to obtain a detail adjustment factor; performing guided filtering on the detail adjustment factor and introducing the detail adjustment factor into the cost function of the detail enhancement guided filter to obtain a detail enhanced image; performing detail layer extraction on the detail enhanced image to obtain an enhanced detail layer image, and performing base layer extraction on the detail enhanced image to obtain a base layer image; performing weighted fusion on the enhanced detail layer image and the base layer image to generate a corresponding target infrared image; A construction module is used to extract pixel temperature from the target infrared image to obtain pixel temperature data and construct the surface temperature field of the server unit; specifically comprising: using linear mapping to extract temperature from each pixel of the target infrared image to obtain preliminary temperature data; using a correction coefficient to linearly correct the preliminary temperature data to obtain corrected temperature data; performing bilinear interpolation on the corrected temperature data to obtain interpolated temperature data; using a finite difference method to perform gradient calculation on the interpolated temperature data to obtain temperature gradient data; using an inverse tangent function to perform gradient direction calculation on the temperature gradient data to obtain gradient direction data; constructing a temperature field on the interpolated temperature data and the temperature gradient data to obtain the surface temperature field of the server unit; An extraction module is used to extract features from the surface temperature field to obtain a target temperature feature set; specifically comprising: performing pixel frequency statistics on the surface temperature field to obtain pixel statistical data; performing average temperature calculation on the pixel statistical data to obtain average temperature data; performing maximum temperature calculation on the pixel statistical data to obtain maximum temperature data; performing minimum temperature calculation on the pixel statistical data to obtain minimum temperature data; performing temperature standard deviation calculation on the pixel statistical data to obtain temperature standard deviation data; using a K-means algorithm to perform cluster analysis on the pixel statistical data to obtain temperature hotspot area data, and performing density calculation on the temperature hotspot area data to obtain hotspot density data; performing set conversion on the average temperature data, the maximum temperature data, the minimum temperature data, the temperature standard deviation data and the hotspot density data to obtain a target temperature feature set; A discrimination module is used to input the target temperature feature set into a preset unit abnormality discrimination model to perform server unit abnormality discrimination to obtain a target abnormality discrimination result; specifically comprising: inputting the target temperature feature set into a preset unit abnormality discrimination model, the unit abnormality discrimination model comprising an embedding layer, a spatiotemporal graph convolutional network, a first support vector machine and a second support vector machine; in the embedding layer, feature encoding is performed on the target temperature feature set to obtain a temperature feature encoding vector; spatiotemporal feature extraction is performed on the temperature feature encoding vector through the spatiotemporal graph convolutional network to obtain a temperature spatiotemporal feature vector; server unit abnormality discrimination is performed on the temperature spatiotemporal feature vector through the first support vector machine to obtain a first abnormality discrimination result; server unit abnormality discrimination is performed on the temperature spatiotemporal feature vector through the second support vector machine to obtain a second abnormality discrimination result; weighted voting analysis is performed on the first abnormality discrimination result and the second abnormality discrimination result to obtain a target abnormality discrimination result.

3. A big data computer room monitoring device, characterized in that: The big data room monitoring device comprises: a memory and at least one processor, wherein instructions are stored in the memory; The at least one processor calls the instruction in the memory so that the big data computer room monitoring device executes the big data computer room monitoring method as claimed in claim 1.

4. A computer-readable storage medium having instructions stored thereon, characterized in that: When the instructions are executed by the processor, the big data computer room monitoring method as described in claim 1 is implemented.

Images