Heating anomaly detection method, equipment and medium

Abstract

The invention discloses a heating anomaly detection method and device and a medium, and relates to the technical field of fault detection, a convolutional neural network model is trained according to an obtained data set, and when the heating condition of a target board card needs to be detected, the heating anomaly is detected. And inputting the thermal imaging temperature distribution diagram of the target board card into the trained convolutional neural network model for heating anomaly detection. The thermal imaging temperature distribution diagram and the convolutional neural network model are combined together, detection of the heating condition of the board card and positioning of abnormal heating points are achieved, the thermal imaging temperature distribution diagram of the board card is automatically recognized through the convolutional neural network model, and the detection efficiency is improved; in the training process, consistency processing is further carried out on the thermal imaging temperature distribution diagram in the data set, the model precision of the convolutional neural network model is improved, the model performance is ensured, dependence on manual inspection is greatly reduced through the convolutional neural network model, subjective errors are avoided, and high-precision identification of the heating condition of the board card is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of fault detection, and in particular to a method, device and medium for detecting abnormal heating. Background Art

[0002] Against the backdrop of the rapid development of digital information in society, servers, as important physical carriers and interaction centers for data and information, are deeply integrated with the information society and support the operation of key businesses. As an important component of the server's internal architecture, boards and cards, once an abnormality occurs, may cause the server to experience problems such as slower data processing speeds and service interruptions, which in severe cases can cause huge economic losses. Common board and card damage includes leakage, damage to internal chips resulting in increased power consumption, short-circuit damage to devices, damage to the cooling system, and damage to connector interfaces. These damages can cause abnormal heating of the board and card. Using the abnormal heating detection results to diagnose and locate the fault of the board and card has become an important way to repair the board and card. Therefore, how to detect the heating condition of the board and card has become a technical problem that urgently needs to be solved.

[0003] It can be seen that how to accurately detect the heating condition of the board is a problem that needs to be solved by those skilled in the art. Summary of the Invention

[0004] The purpose of the embodiments of the present invention is to provide a method, device and medium for detecting abnormal heating, which can solve the problem of how to accurately detect the heating condition of a board.

[0005] To solve the above technical problems, an embodiment of the present invention provides a method for detecting abnormal fever, comprising: Acquire a data set; samples in the data set include a thermal imaging temperature distribution map of a normal board card, a thermal imaging temperature distribution map of an abnormal board card, a normal heating condition of the board card, and an abnormal heating condition of the board card; wherein the heating condition of the normal board card is normal heating of the board card, and the heating condition of the abnormal board card includes abnormal heating of the board card and the location of the abnormal heating point; Performing consistency processing on any thermal imaging temperature distribution map in the data set; wherein the processing dimension of the consistency processing includes at least one of spatial resolution, brightness and / or contrast; Train the convolutional neural network model based on the consistent processed dataset; Obtain thermal imaging temperature distribution map of target board; The trained convolutional neural network model is used to process the thermal imaging temperature distribution map of the target board to obtain the heating status of the target board.

[0006] In some embodiments, consistency processing is performed on any thermal imaging temperature distribution map in the data set, including: For any thermal imaging temperature distribution map in the data set, determining the spatial resolution of the thermal imaging temperature distribution map, and adjusting the spatial resolution of the thermal imaging temperature distribution map to a preset pixel range; Obtain brightness and / or contrast of all thermal imaging temperature distribution maps; The pixel value of each pixel in the thermal imaging temperature distribution map is adjusted to adjust the brightness and / or contrast of all the thermal imaging temperature distribution maps to the corresponding brightness target value and / or contrast target value.

[0007] In some embodiments, before performing consistency processing on any thermal imaging temperature distribution map in the data set, the method further includes: For any thermal imaging temperature distribution map in the data set, a plurality of filtering regions of the thermal imaging temperature distribution map are determined with any pixel in the thermal imaging temperature distribution map as the center; For any filter area, the Gaussian kernel is used to cover the filter area; The temperature values corresponding to all pixels in the filter area are weighted and summed according to the weight of the Gaussian kernel to obtain the filtered temperature value corresponding to the center pixel of the filter area; The filtered temperature value is used to update the temperature value of each pixel in the thermal imaging temperature distribution map to obtain a filtered thermal imaging temperature distribution map.

[0008] In some embodiments, after obtaining the data set, the method further includes: Rotating and / or flipping and / or enlarging and / or reducing the thermal imaging temperature distribution map in the data set to obtain a transformed image; The transformed image and the fever condition corresponding to the transformed image are added to the dataset as new samples.

[0009] In some embodiments, the dataset includes a training set, a validation set, and a test set; The convolutional neural network model is trained based on the consistent dataset, including: Inputting the thermal imaging temperature distribution map in the training set into the convolutional neural network model, and obtaining the detection result of the convolutional neural network model on the thermal imaging temperature distribution map in the training set; Iterate the model parameters of the convolutional neural network model based on the difference between the test results and the fever status corresponding to the training set; the model parameters include the weight of the convolutional layer, the bias of the convolutional layer, the weight of the fully connected layer, and the bias of the fully connected layer; Input the thermal imaging temperature distribution map in the validation set into the convolutional neural network model to obtain the validation results of the validation set; Adjust the hyperparameters of the convolutional neural network model based on the verification results. These hyperparameters include the size of the convolution kernel in the convolution layer, the size of the pooling window in the pooling layer, and the learning rate of the convolutional neural network model. After training the convolutional neural network model based on the consistent dataset, it also includes: Input the thermal imaging temperature distribution map in the test set into the trained convolutional neural network model to obtain the test results of the test set; Determine whether the model performance of the trained convolutional neural network model reaches the target value based on the test results; If not, adjust the model architecture of the convolutional neural network model and jump back to the step of training the convolutional neural network model based on the consistency-processed dataset.

[0010] In some embodiments, the specific process of processing the thermal imaging temperature distribution map of the target board by the trained convolutional neural network model includes: Using a plurality of first convolution kernels to perform a convolution operation on the thermal imaging temperature distribution map obtained through the input layer, so as to extract features of the thermal imaging temperature distribution map to obtain a plurality of first layer feature maps; For any first-layer feature map, a first pooling window is used to perform a pooling operation on the first-layer feature map to obtain a first-layer pooled feature map; A second convolution kernel is used to perform a convolution operation on the first-layer pooled feature map to obtain a second-layer feature map; wherein the size of the second convolution kernel is smaller than the size of the first convolution kernel; The second pooling window is used to perform a pooling operation on the second layer feature map to obtain the second layer pooled feature map; A fully connected layer is used to integrate all the second-layer pooled feature maps into a feature vector; The feature vector is mapped to the classification layer to output the heating status of the target board through the classification layer.

[0011] In some embodiments, a plurality of first convolution kernels are used to perform a convolution operation on the thermal imaging temperature distribution map obtained through the input layer to extract features from the thermal imaging temperature distribution map to obtain a plurality of first layer feature maps, including: A plurality of first convolution kernels are configured for the thermal imaging temperature distribution map obtained through the input layer; wherein the plurality of first convolution kernels extract different feature types; the feature types include edges, extreme values, and textures; For any first convolution kernel, the first convolution kernel is used to cover a local area on the thermal imaging temperature distribution map; Perform a convolution operation on the local area covered by the first convolution kernel to extract the image features of the local area; Controlling the first convolution kernel to slide according to a first preset step size to cover another local area on the thermal imaging temperature distribution map, and jumping to the step of performing a convolution operation on the local area covered by the first convolution kernel until the first convolution kernel covers the thermal imaging temperature distribution map; The image features of all local areas are combined to obtain the first layer feature map generated by the first convolution kernel.

[0012] In some embodiments, a first pooling window is used to perform a pooling operation on the first layer feature map to obtain a first layer pooled feature map, including: Use the first pooling window to cover a pooling area on the first layer feature map; Perform a pooling operation on the pooling area covered by the first pooling window to obtain the eigenvalue corresponding to the pooling area; Controlling the first pooling window to slide according to the second preset step size to cover another pooling area on the first layer feature map, and jumping to the step of performing a pooling operation on the pooling area covered by the first pooling window until the first pooling window covers the first layer feature map; The eigenvalues corresponding to all pooled areas are combined to obtain the first layer of pooled feature map generated by the first pooling window.

[0013] To solve the above technical problems, an embodiment of the present invention further provides an electronic device, including: memory for storing computer programs; The processor is used to execute the computer program to implement the steps of the above-mentioned fever abnormality detection method.

[0014] To solve the above technical problems, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned fever anomaly detection method are implemented.

[0015] As can be seen from the above technical solution, the convolutional neural network model is trained based on the acquired data set. When it is necessary to detect the heating condition of the target board, the thermal imaging temperature distribution map of the target board can be directly input into the trained convolutional neural network model to detect abnormal heating. The beneficial effect of the present invention is that the thermal imaging temperature distribution map is combined with the convolutional neural network model to realize the detection of the heating condition of the board and the location of abnormal heating points. The convolutional neural network model is used to automatically identify the thermal imaging temperature distribution map of the board, thereby improving the detection efficiency. During training, the thermal imaging temperature distribution map in the data set is further processed for consistency to improve the model accuracy of the convolutional neural network model and ensure model performance. The convolutional neural network model greatly reduces the reliance on manual inspection, avoids subjective errors, and achieves high-precision identification of the heating condition of the board. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention, the following is a brief introduction to the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0017] Figure 1 A schematic flow chart of a method for detecting abnormal fever provided by an embodiment of the present invention; Figure 2 A schematic diagram of the structure of a convolutional neural network model provided by an embodiment of the present invention; Figure 3 A schematic flow chart of another method for detecting abnormal fever provided by an embodiment of the present invention; Figure 4 A schematic diagram of the training process of a convolutional neural network model provided by an embodiment of the present invention; Figure 5 A schematic diagram of the processing flow of a convolutional neural network model provided by an embodiment of the present invention; Figure 6 A schematic structural diagram of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0018] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making any creative efforts shall fall within the scope of protection of the present invention.

[0019] The terms "including" and "having," as used in the present description and accompanying drawings, and any variations thereof, are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus comprising a series of steps or elements is not limited to the listed steps or elements and may include steps or elements that are not listed.

[0020] In order to enable those skilled in the art to better understand the present invention, the present invention will be further described in detail below with reference to the accompanying drawings and specific implementation methods.

[0021] Next, a method for detecting abnormal fever provided by an embodiment of the present invention is described in detail. Figure 1 As shown, Figure 1 A schematic flow chart of a method for detecting abnormal fever provided by an embodiment of the present invention; the method for detecting abnormal fever includes: S11: Acquire a data set; samples in the data set include a thermal imaging temperature distribution map of a normal board card, a thermal imaging temperature distribution map of an abnormal board card, a normal heating condition of the board card, and an abnormal heating condition of the board card; wherein the heating condition of the normal board card is normal heating of the board card, and the heating condition of the abnormal board card includes abnormal heating of the board card and the location of the abnormal heating point; It's easy to understand that to detect the heating status of a board, particularly whether its heating is abnormal, this application uses tools such as thermal imagers to scan the board's surface and generate a thermal imaging temperature distribution map. In this thermal imaging temperature distribution map, areas with higher temperatures appear brighter, providing a visual representation of the distribution of hot spots on the board. Therefore, when detecting the heating status of a board, abnormal heating can be determined by analyzing and interpreting the corresponding thermal imaging temperature distribution map. A thermal imaging temperature distribution map is a thermal image generated by scanning the board's surface using a professional tool such as a thermal imager. This helps quickly and comprehensively locate hot areas. The entire detection process does not require physical contact with the board, thus avoiding potential interference caused by contact. Furthermore, to more accurately analyze and interpret the thermal imaging temperature distribution map of the board, this application also specifically constructs and trains a CNN (Convolutional Neural Network) model to directly analyze and output the thermal imaging temperature distribution map of the board to detect the heating status of the board.

[0022] It is understood that in order to ensure the accuracy and reliability of the convolutional neural network model in detecting heating conditions, the convolutional neural network model needs to be trained. Therefore, a dataset must be obtained first for training the convolutional neural network model. The dataset includes thermal imaging temperature distribution maps of normal boards, thermal imaging temperature distribution maps of abnormal boards, and the heating conditions of normal boards and abnormal boards. The dataset includes input samples and the actual results corresponding to the input samples. The thermal imaging temperature distribution maps of normal boards and abnormal boards serve as input samples for model training, and the heating conditions of normal boards and abnormal boards serve as the actual results corresponding to the input samples. Both normal boards and abnormal boards have known heating conditions. Pre-knowledge of the actual heating conditions of normal and abnormal boards provides a benchmark for model training. A normal board is known to have a normal heating condition, while an abnormal board is known to have an abnormal heating condition, and the location of the abnormal heating point causing the abnormal heating is known. The abnormal heating point refers to a local location on the board where the temperature is significantly higher than the normal operating state or the surrounding area, and is the primary detection target for abnormal heating detection on the board. A board refers to the hardware circuit board inside a device such as a server.

[0023] It should be noted that the number of samples in the data set can be set and adjusted according to actual needs. In particular, the number of abnormal boards needs to be set according to different heating conditions when the boards have abnormal heating. For example, when there are N positions on the target board that need to be detected for abnormal heating, it is necessary to obtain the abnormal board with abnormal heating at the first position, the abnormal board with abnormal heating at the second position... and the abnormal board with abnormal heating at the Nth position. These N abnormal boards are used as samples, and then one or more normal boards are obtained as samples. The N+1 real heating conditions of the board are obtained as labels, including normal heating of the board, abnormal heating of the board and the first position as an abnormal heating point, abnormal heating of the board and the second position as an abnormal heating point... and abnormal heating of the board and the Nth position as an abnormal heating point. The convolutional neural network model is trained to learn the mapping relationship between the input thermal imaging temperature distribution map and these N+1 labels, so that the convolutional neural network model can directly output a label as the heating condition of the target board based on the input thermal imaging temperature distribution map; N is a positive integer.

[0024] It should be further explained that the N positions can be N areas or positions where important components are located on the board, or N areas or positions on the board with a relatively high risk of abnormal heating determined by experience or simulation; or the entire board can be divided into N areas as N positions directly according to the actual design of the components and wiring on the board, and the divided areas are used as units for abnormal heating detection; the N positions can also be all positions on the entire board. This application does not make any special restrictions on the specific value of N and the implementation method. The specific number of labels in the data set can be set according to the detection accuracy and detection requirements required for the actual application. The more labels determined when obtaining the data set, the higher the detection accuracy when actually detecting the target board, and the more detailed the classification of the heating condition of the target board.

[0025] S12: performing consistency processing on any thermal imaging temperature distribution map in the data set; wherein the processing dimension of the consistency processing includes at least one of spatial resolution, brightness and / or contrast; It is understandable that in order to improve the accuracy and reliability of the final trained convolutional neural network model, when using a dataset to train the convolutional neural network model, the input samples in the dataset can be pre-processed for consistency to obtain standardized and normalized input samples, thereby unifying the feature scale, eliminating bias, and improving the efficiency, stability, and generalization ability of model training. This application does not specifically limit the specific implementation methods of consistency processing. The image size and / or pixel size can be adjusted by normalizing all thermal imaging temperature distribution maps. The differences in brightness and / or contrast between different thermal imaging temperature distribution maps can also be eliminated by using histogram equalization methods or directly calculating the mean and standard deviation of the image dataset. Other methods can also be used to pre-process the input samples in the dataset to improve the accuracy and reliability of the input samples, thereby improving the training efficiency of the model and ensuring model performance.

[0026] S13: Training the convolutional neural network model based on the consistent dataset; It is not difficult to understand that after the dataset is preprocessed such as consistency processing, the preprocessed dataset is used to train the built convolutional neural network model. Figure 2 As shown, Figure 2 A schematic diagram of the structure of a convolutional neural network model provided in an embodiment of the present invention. The convolutional neural network model specifically includes an input layer, a convolution layer, a pooling layer, a fully connected layer, and a classification layer, which are connected in sequence. The input layer obtains the input thermal imaging temperature distribution map. The convolution layer and the pooling layer gradually extract features from the thermal imaging temperature distribution map. The fully connected layer integrates the extracted features and outputs them to the classification layer, which serves as the output layer. The classification layer determines the labels corresponding to the features based on the extracted features and outputs the corresponding labels as the detection results. The thermal imaging temperature distribution map is a two-dimensional image. Its core feature is local spatial correlation, including temperature similarity between adjacent regions, temperature gradient changes between adjacent regions, and the relationship between temperature and position. By combining the CNN model with the thermal imaging temperature distribution map of the board, the CNN model's layer structure can gradually extract the relationship between these temperatures and between temperature and position. Starting from image features such as edges, textures, and temperature regional patterns, it can detect heating conditions at different locations on the board and locate abnormal heating points on the board. The temperature regional pattern refers to the spatial distribution characteristics of temperature.

[0027] S14: Obtain a thermal imaging temperature distribution map of the target board; S15: Using the trained convolutional neural network model to process the thermal imaging temperature distribution map of the target board to obtain the heating condition of the target board.

[0028] It can be understood that after the CNN model training is completed, it can be applied to the abnormal heating detection of the target board and card. The thermal imaging temperature distribution map of the target board and card is directly input into the CNN model. The trained CNN model has a correspondence between the thermal imaging temperature distribution map and the label. The CNN model will output the label corresponding to the target board and card based on the input thermal imaging temperature distribution map, thereby determining the heating condition of the target board and card. Specifically, if the temperature values in the thermal imaging temperature distribution map of the target board and card are normal, the detection result output by the CNN model is that the heating of the board and card is normal. If the temperature values in the thermal imaging temperature distribution map of the target board and card exceed the threshold, the detection result output by the CNN model is that the heating of the board and card is abnormal. In the case where the detection result is that the heating of the board and card is abnormal, the CNN model will further output the position of the abnormal heating point of the target board and card based on the analysis results. For example, the output detection result is that the heating of the target board and card is abnormal and the abnormal heating point is the first position.

[0029] This application proposes a method for analyzing the thermal imaging temperature distribution map of a board using the image recognition function of a convolutional neural network, thereby detecting abnormal heating of the board and locating abnormal heating points. The thermal imaging temperature distribution map of the board obtained by a thermal imager no longer requires professional interpretation. At the same time, it avoids the problem of misjudgment when the abnormal heating temperature is slightly higher than the normal temperature when manually interpreting the thermal imaging temperature distribution map, thereby reducing the reliance on manual labor in the process of troubleshooting abnormal heating points.

[0030] Understandably, see Figure 3 As shown, Figure 3 A flow chart of another method for detecting abnormal heating provided by an embodiment of the present invention; first, thermal imaging temperature distribution maps of normal boards and abnormal boards are obtained as a data set, and then the data set is divided into a training set, a validation set, and a test set, and the thermal imaging temperature distribution maps in the data set are preprocessed; after obtaining the data set, a CNN model can be built based on the layer structure described above, the layer structure of the CNN model can be preliminarily defined, the initial model parameters and activation function can be set, and then the CNN model can be trained using the training set in the data set, and the model performance can be evaluated using the validation set in the data set, and the model parameters can be adjusted according to the evaluation results. Finally, the model performance of the CNN model can be finally evaluated using the test set after the training is completed. After completing the training and parameter tuning of the CNN model through the data set, the final CNN model is obtained, and then the final CNN model is used to accurately locate the abnormal heating points of the board and assist R&D personnel in quickly troubleshooting board failures.

[0031] Furthermore, new data generated in the actual application process can also be regularly input into the CNN model. The new data includes the input thermal imaging temperature distribution map of the target board and the detection results of the target board output by the CNN model. They are used as new samples to dynamically adjust the model parameters and hyperparameters of the CNN model, thereby enhancing the applicability of the CNN model to complex boards and enabling it to always maintain a high resolution and positioning accuracy.

[0032] The present invention provides a method for detecting abnormal heating using a CNN module to identify thermal imaging temperature distribution patterns of abnormally heated boards and cards, enabling high-precision detection and location of heating faults. The CNN model's powerful feature extraction capabilities accurately capture subtle features associated with abnormal heating in the thermal imaging temperature distribution patterns, enabling high-precision identification and significantly reducing false positives and missed detections. Once the thermal imaging temperature distribution patterns are input into a trained CNN model, the model quickly analyzes and identifies the patterns, rapidly locating boards and cards that may be experiencing abnormal heating, significantly improving detection speed and efficiency. Using a CNN model to identify thermal imaging temperature distribution patterns significantly reduces reliance on manual inspection. Once trained, the CNN model can automatically perform inspections, requiring only a small number of personnel to maintain the model and conduct preliminary verification of the results. This frees up significant manpower and reduces reliance on manual inspection. Manual inspection of thermal imaging temperature distribution patterns can result in varying inspection results due to differences in experience, skill level, and subjective judgment criteria. However, the CNN model's recognition process is based on pre-trained algorithms and data, unaffected by subjective factors, providing more objective and consistent fault diagnosis results and avoiding subjective errors. By detecting and locating abnormal heating points on the board using the CNN model, the risk of board hardware failure can be detected and the cause of the failure can be located promptly, ensuring the safety and service life of the server's hardware. This can also prevent the heat generated by a single component from radiating to the surrounding area and damaging surrounding components, ensuring the stable operation of the server system, effectively avoiding slow server response and reduced data processing capacity caused by abnormal heating, and ensuring the smooth operation of the program. It is also beneficial for board troubleshooting. When a board failure occurs, the abnormal heating point can be located, the location of the faulty component can be determined, the cause of the board failure can be preliminarily determined, and a more detailed system analysis can be conducted.

[0033] In some embodiments, consistency processing is performed on any thermal imaging temperature distribution map in the data set, including: For any thermal imaging temperature distribution map in the data set, determining the spatial resolution of the thermal imaging temperature distribution map, and adjusting the spatial resolution of the thermal imaging temperature distribution map to a preset pixel range; Obtain brightness and / or contrast of all thermal imaging temperature distribution maps; The pixel value of each pixel in the thermal imaging temperature distribution map is adjusted to adjust the brightness and / or contrast of all the thermal imaging temperature distribution maps to the corresponding brightness target value and / or contrast target value.

[0034] It is not difficult to understand that consistency processing specifically includes one or more combinations of consistency processing of spatial resolution, consistency processing of brightness and consistency processing of contrast. In order to facilitate the CNN model to process the thermal imaging temperature distribution map, the image size and pixel size can be adjusted by adjusting the spatial resolution of the thermal imaging temperature distribution map, and all thermal imaging temperature distribution maps are normalized to adjust the image size, so that the pixel size of the thermal imaging temperature distribution map is unified within a specific preset pixel range. When adjusting the spatial resolution, you can first select a thermal imaging temperature distribution map as a reference image, and adjust the spatial resolution of other thermal imaging temperature distribution maps based on the reference image. If the spatial resolution of a thermal imaging temperature distribution map is greater than the reference image, you can use downsampling to reduce the number of pixels. If the spatial resolution of a thermal imaging temperature distribution map is less than the reference image, you can use upsampling to reduce the number of pixels. This application does not specifically limit the specific value of the preset pixel range and the implementation method. At the same time, histogram equalization is used or the mean and standard deviation of the brightness and / or contrast of the thermal imaging temperature distribution map is directly calculated. These values are then applied to each thermal imaging temperature distribution map. The calculated mean and standard deviation serve as the corresponding target brightness and / or contrast values. This eliminates differences in brightness and contrast between different thermal imaging temperature distribution maps, maintaining consistency across all thermal imaging temperature distribution maps, such as brightness and contrast, to facilitate model learning and feature extraction comparison. Brightness adjustment can be achieved by shifting pixel values overall, while contrast adjustment can be achieved by expanding or narrowing the distribution range of pixel values.

[0035] Specifically, by performing consistency processing on the spatial resolution, brightness and contrast of each thermal imaging temperature distribution map in the data set, the interference and influence of spatial scale differences and brightness / contrast fluctuations on the CNN model feature extraction process can be eliminated, and the training difficulty of the model can be reduced. After the input samples are standardized, the model can focus on the characteristics of the input samples themselves (such as shape and texture), with higher training efficiency and more stable performance when applied, thereby improving the reliability of the final trained model.

[0036] In some embodiments, before performing consistency processing on any thermal imaging temperature distribution map in the data set, the method further includes: For any thermal imaging temperature distribution map in the data set, a plurality of filtering regions of the thermal imaging temperature distribution map are determined with any pixel in the thermal imaging temperature distribution map as the center; For any filter area, the Gaussian kernel is used to cover the filter area; The temperature values corresponding to all pixels in the filter area are weighted and summed according to the weight of the Gaussian kernel to obtain the filtered temperature value corresponding to the center pixel of the filter area; The filtered temperature value is used to update the temperature value of each pixel in the thermal imaging temperature distribution map to obtain a filtered thermal imaging temperature distribution map.

[0037] It is understood that in addition to performing consistency processing on the thermal imaging temperature distribution map, considering that the thermal imager itself contains a certain amount of noise when acquiring the thermal imaging temperature distribution map using a thermal imager, filtering algorithms can be used to remove interference such as Gaussian noise from the thermal imaging temperature distribution map, thereby performing further preprocessing and improving image quality. Specifically, Gaussian filtering and other methods can be used to filter the thermal imaging temperature distribution map. A Gaussian kernel is used to perform weighted averaging of image pixels to achieve smoothing and noise reduction while preserving image edges and details as much as possible. For pixels at the edges or corners of the image, the area outside the image can be filled using methods such as zero padding, copy padding, or mirror padding to process the pixels at the edge or corner of the image into center pixels, thereby determining the filtered temperature value corresponding to the pixels at the edge or corner of the image. The center pixel refers to the pixel point in the center of the filtering area. The filtering area generally matches the size of the Gaussian kernel, for example, a 3×3 or 5×5 matrix area. Other filtering algorithms can also be used to preprocess the thermal imaging temperature distribution map, and this is not specifically limited in this application.

[0038] Specifically, by using a filtering algorithm to further preprocess the thermal imaging temperature distribution map, the image quality is improved, noise is suppressed, the data quality of the thermal imaging temperature distribution map is improved, the feature robustness is enhanced, the risk of overfitting is reduced, and the model is helped to learn the essential laws of the data in the thermal imaging temperature distribution map more stably and efficiently, ultimately improving the model's generalization ability and detection accuracy.

[0039] In some embodiments, after obtaining the data set, the method further includes: Rotating and / or flipping and / or enlarging and / or reducing the thermal imaging temperature distribution map in the data set to obtain a transformed image; The transformed image and the fever condition corresponding to the transformed image are added to the dataset as new samples.

[0040] It is not difficult to understand that in order to expand the sample, data enhancement can also be achieved by rotating, flipping, zooming in and out, and other operations on the thermal imaging temperature distribution map in the acquired data set, thereby expanding the universal applicability of the data set. The transformed image obtained after these operations is placed in the data set as an independent new sample, specifically in the training set of the data set, to expand the coverage of the CNN model training set and provide more training data to assist in CNN model training. This application does not specifically limit the specific method of obtaining the transformed image, and is not limited to the rotation and / or flipping and / or zooming and / or zooming operations of this embodiment. The transformed image can also be obtained by cropping and other methods.

[0041] Specifically, data enhancement of the dataset is achieved by transforming images, which increases the amount of training data during model training, alleviates the overfitting of the CNN model, improves the generalization ability of the model, and makes full use of the limited data in the dataset to ensure the accuracy of model training.

[0042] See also Figure 4 As shown, Figure 4 A schematic diagram of a training process for a convolutional neural network model provided in an embodiment of the present invention; in some embodiments, the data set includes a training set, a validation set, and a test set; The convolutional neural network model is trained based on the consistent dataset, including: Inputting the thermal imaging temperature distribution map in the training set into the convolutional neural network model, and obtaining the detection result of the convolutional neural network model on the thermal imaging temperature distribution map in the training set; Iterate the model parameters of the convolutional neural network model based on the difference between the test results and the fever status corresponding to the training set; the model parameters include the weight of the convolutional layer, the bias of the convolutional layer, the weight of the fully connected layer, and the bias of the fully connected layer; Input the thermal imaging temperature distribution map in the validation set into the convolutional neural network model to obtain the validation results of the validation set; Adjust the hyperparameters of the convolutional neural network model based on the verification results. These hyperparameters include the size of the convolution kernel in the convolution layer, the size of the pooling window in the pooling layer, and the learning rate of the convolutional neural network model. After training the convolutional neural network model based on the consistent dataset, it also includes: Input the thermal imaging temperature distribution map in the test set into the trained convolutional neural network model to obtain the test results of the test set; Determine whether the model performance of the trained convolutional neural network model reaches the target value based on the test results; If not, adjust the model architecture of the convolutional neural network model and jump back to the step of training the convolutional neural network model based on the consistency-processed dataset.

[0043] Understandably, to achieve an accurate and reliable CNN model, the dataset must be divided into training, validation, and test sets. The CNN model first learns using the training set, then iteratively optimizes the model using the validation set. Finally, the test set is used to evaluate the model's performance. This ensures that the CNN model not only performs well on the training data in the dataset, but also operates stably on new data in real-world applications. First, a sufficient number of thermal image temperature distribution images of normal boards and known faulty boards are captured using a thermal imager as the dataset. These images include those of boards with localized overheating due to capacitor failures and those of boards experiencing high chip loads. All captured thermal image temperature distribution images of both normal and faulty boards are then preprocessed, including filtering, consistency checks, and data augmentation. After acquiring the dataset, the initial CNN model is built using the ResNet architecture, which boasts high accuracy, strong generalization capabilities, and ease of adjustment. The model is constructed using a predefined layer structure, predefined model parameters, and predefined hyperparameters. After the model is built, the preprocessed thermal imaging temperature distribution map is divided into training set, validation set and test set. The preprocessing mainly processes the parameters such as the size and brightness of the thermal imaging temperature distribution map to match the input requirements and number of channels of the CNN model.

[0044] After the initial CNN model is constructed, it is first trained and its parameters adjusted using the training set. During training, the thermal image temperature distribution maps from the training set are continuously fed into the CNN model. Using a backpropagation algorithm, the CNN model continuously adjusts model parameters in real time based on the discrepancies between the model output and the true image labels (i.e., the true representation of the card status by the CNN model). Specifically, the weights and biases of the convolutional layers and the weights and biases of the fully connected layers in the CNN model are adjusted in real time, so that the output of the fully connected layers gradually approaches the feature values corresponding to the true image labels. The true image labels are the true representation of the card status by the thermal image temperature distribution maps. These labels describe the card status, such as whether the card heating is normal and the location of abnormal heating points, and are fed into the classification layer of the CNN model. The validation set is then used to evaluate model performance in real time during training. Hyperparameters, including network architecture parameters (convolution kernel size, pooling window size), optimizer parameters (learning rate, weight decay), and training parameters (batch size, number of training epochs), are adjusted based on the validation results to prevent overfitting or underfitting. For example, the convolution kernel size parameter determines the convolution layer's ability to extract input features. Iteratively optimizing its size improves feature extraction efficiency. Adjusting the offset of the pooling window in the pooling layer helps obtain a streamlined image that accurately reflects feature information. The validation set is used to evaluate model training results and adjust hyperparameters such as the learning rate. After training is complete, the trained model is fully evaluated using the test set. Test results include key metrics such as accuracy and recall, which measure the CNN model's ability to recognize the board's thermal image temperature distribution. If model performance does not meet expectations, further optimization can be performed. For example, the model structure can be adjusted by increasing or decreasing the number of convolutional and fully connected layers before retraining. Alternatively, the preprocessing of the thermal image temperature distribution can be improved, and different data augmentation strategies can be tried. Regularization techniques can also be used to constrain the learning rate, weight parameters, the percentage of discarded unimportant information, and the number of iterations of the convolutional and fully connected layers to improve model generalization. Finally, the trained and optimized CNN model is deployed in a real-world board monitoring system. When a new target board's thermal imaging temperature distribution map is input into the board monitoring system, the CNN model rapidly processes and analyzes it according to the established convolution, pooling, and full-connection process. It then outputs the board's heating condition category corresponding to the thermal imaging temperature distribution map. This clearly identifies whether the board is experiencing abnormal heating and the location of the heating point, which serves as the possible abnormality type for the board, providing strong support for timely hardware maintenance and troubleshooting. Furthermore, new abnormal heating conditions of target boards are regularly input into the model, continuously optimizing its structure and parameters to make it more adaptable to diverse situations and complex boards.

[0045] Specifically, the CNN model is trained, optimized, and evaluated by dividing the model into training sets, validation sets, and test sets, thereby achieving comprehensive and effective training of the CNN model, improving the detection accuracy of the final trained CNN model, and ensuring that the CNN model can stably and reliably detect the heating conditions of different target boards.

[0046] In some embodiments, the specific process of processing the thermal imaging temperature distribution map of the target board by the trained convolutional neural network model includes: Using a plurality of first convolution kernels to perform a convolution operation on the thermal imaging temperature distribution map obtained through the input layer, so as to extract features of the thermal imaging temperature distribution map to obtain a plurality of first layer feature maps; For any first-layer feature map, a first pooling window is used to perform a pooling operation on the first-layer feature map to obtain a first-layer pooled feature map; A second convolution kernel is used to perform a convolution operation on the first-layer pooled feature map to obtain a second-layer feature map; wherein the size of the second convolution kernel is smaller than the size of the first convolution kernel; The second pooling window is used to perform a pooling operation on the second layer feature map to obtain the second layer pooled feature map; A fully connected layer is used to integrate all the second-layer pooled feature maps into a feature vector; The feature vector is mapped to the classification layer to output the heating status of the target board through the classification layer.

[0047] It's easy to understand that the input layer in the CNN model is used to input the preprocessed thermal imaging temperature distribution map into the CNN model. The target board's thermal imaging temperature distribution map can be preprocessed by referring to the dataset's preprocessing process before inputting it into the input layer. The input layer defines the dimensions of the input data. The input layer then sends the input thermal imaging temperature distribution map to the convolutional layer, which convolves the thermal imaging temperature distribution map with a convolution kernel. Specifically, the convolutional layer uses different convolution kernels to capture different features of the thermal imaging temperature distribution map, extracting the features of the thermal imaging temperature distribution map and forming a new feature image containing the main feature information of the thermal imaging temperature distribution map, namely the first and second layer feature maps. The pooling layer, which is generally connected after the convolutional layer, can reduce the model's computational complexity by reducing the resolution of the feature image output by the convolutional layer, while retaining the key features of the thermal imaging temperature distribution map and preventing overfitting. In this embodiment, two convolutional layers and two corresponding pooling layers are used to extract features from the thermal imaging temperature distribution map. To improve the accuracy and efficiency of feature extraction, the convolution kernel used in the second convolutional layer is smaller in size. The feature image output by the first convolutional layer is the first-layer feature image. After further processing by the first pooling layer, the first-layer pooled feature image that retains key features is obtained. The feature image output by the second convolutional layer is the second-layer feature image. After further processing by the second pooling layer, the second-layer pooled feature image that retains key features is obtained. The fully connected layer then integrates the feature information output by the pooling layer to obtain a feature vector for the thermal imaging temperature distribution map. Finally, the classification layer maps the feature vector to the category space determined during training. The classification layer classifies abnormal hot spots, with one label corresponding to one category space. The category space is output as the final detection result to locate the abnormal points.

[0048] Specifically, by setting up multiple convolutional layers and corresponding multi-pooling layers, the features in the thermal imaging temperature distribution map can be extracted layer by layer, thereby achieving accurate recognition and interpretation of the thermal imaging temperature distribution map, and improving the accuracy and reliability of the CNN model in detecting the heating condition of the board.

[0049] See also Figure 5 As shown, Figure 5 A schematic diagram of a processing flow of a convolutional neural network model provided in an embodiment of the present invention; in some embodiments, a plurality of first convolution kernels are used to perform a convolution operation on a thermal imaging temperature distribution map obtained through an input layer to extract features from the thermal imaging temperature distribution map to obtain a plurality of first-layer feature maps, including: A plurality of first convolution kernels are configured for the thermal imaging temperature distribution map obtained through the input layer; wherein the plurality of first convolution kernels extract different feature types; the feature types include edges, extreme values, and textures; For any first convolution kernel, the first convolution kernel is used to cover a local area on the thermal imaging temperature distribution map; Perform a convolution operation on the local area covered by the first convolution kernel to extract the image features of the local area; Controlling the first convolution kernel to slide according to a first preset step size to cover another local area on the thermal imaging temperature distribution map, and jumping to the step of performing a convolution operation on the local area covered by the first convolution kernel until the first convolution kernel covers the thermal imaging temperature distribution map; The image features of all local areas are combined to obtain the first layer feature map generated by the first convolution kernel.

[0050] It is understandable that in order to achieve a comprehensive interpretation of the thermal imaging temperature distribution map, the present application sets multiple different first convolution kernels in the convolution layer. Each first convolution kernel extracts different types of features. The first convolution kernel specifically includes a convolution kernel for detecting edges (temperature mutation boundaries in the thermal imaging temperature distribution map), a convolution kernel for detecting gradients (temperature gradient patterns in the thermal imaging temperature distribution map), a convolution kernel for detecting extreme values (thermoelectric sums with the highest temperature in the thermal imaging temperature distribution map or cold spots with the lowest temperature in the thermal imaging temperature distribution map), a convolution kernel for detecting textures in uniform temperature zones (temperature distribution shapes in the thermal imaging temperature distribution map), etc. The convolution kernel can be designed based on the spatial features of the thermal imaging temperature distribution map (including edges, gradients, shapes) and temperature value distribution patterns (sudden changes, gradual changes, aggregations), etc. The convolution kernel specifically captures and extracts image features such as the position distribution information, gradient changes, and local structures of the board temperature by sliding and calculating the weighted sum of local areas, ultimately obtaining a feature image formed by the combination of multiple areas. The convolution kernel is generally a 3×3 or 5×5 area matrix, each time covering a local area of corresponding size in the temperature distribution map. The convolution kernel determines the importance of the temperature at different positions in the local area through different weight values to improve the credibility of the feature information. At the same time, it can also adapt to the changes in feature information caused by the temperature distribution differences at different positions of the board by changing the weight value of the convolution kernel. This application does not make any special restrictions on the specific configuration of the first convolution kernel and the specific implementation methods such as the first preset step size in the feature extraction process, which can be set and adjusted according to actual application conditions. The configuration of the second convolution kernel and the specific feature extraction process can refer to the implementation method of the first convolution kernel in this embodiment, and this application will not go into details.

[0051] Furthermore, when configuring the convolution kernel, it can also be achieved by dividing the thermal imaging temperature distribution map into areas. Specifically, the board can be divided into several detection areas according to the actual components and wiring design of the board, such as the area where the processor is located, the area where the capacitor is located, etc. For each detection area, a corresponding convolution kernel is configured to perform feature extraction. The convolution kernel can also be configured by comprehensively considering the differences in the divided areas and feature types. For example, after the board is divided into several detection areas, for any detection area, several convolution kernels with different extraction feature types are configured. For example, for i detection areas, each detection area is configured with j convolution kernels with different extraction feature types. Finally, i×j convolution kernels are configured in one convolution layer to realize i×j feature extraction channels. The extraction feature type of the convolution kernel of each detection area can be adjusted according to the actual heating situation. Different detection areas select convolution kernels with different extraction feature types, which is more flexible and has a wider range of applications.

[0052] Specifically, the convolution kernel is configured in the convolution layer to realize the automatic extraction of image features in the thermal imaging temperature distribution map, and then multiple convolution kernels are used to realize the extraction of different types of features in the thermal imaging temperature distribution map, thereby realizing the effective recognition and interpretation of the temperature distribution in the thermal imaging temperature distribution map.

[0053] In some embodiments, a first pooling window is used to perform a pooling operation on the first layer feature map to obtain a first layer pooled feature map, including: Use the first pooling window to cover a pooling area on the first layer feature map; Perform a pooling operation on the pooling area covered by the first pooling window to obtain the eigenvalue corresponding to the pooling area; Controlling the first pooling window to slide according to the second preset step size to cover another pooling area on the first layer feature map, and jumping to the step of performing a pooling operation on the pooling area covered by the first pooling window until the first pooling window covers the first layer feature map; The eigenvalues corresponding to all pooled areas are combined to obtain the first layer of pooled feature map generated by the first pooling window.

[0054] It is not difficult to understand that the pooling operation of the pooling layer is to calculate the pixels in each pooling window by sliding a fixed-size pooling window on the input feature map, such as extracting the maximum value or average value in the pooling window as the feature value of the pooling area, thereby generating a smaller-sized pooled feature map, and controlling the pooling window to move a specific interval with a second preset step size, thereby reducing the spatial resolution of the feature map, retaining the key feature information of the image, and increasing the efficiency of the model processing information. The second preset step size and the size of the pooling window can be set according to actual conditions, and this application does not make any special restrictions here. The feature extraction process of the second pooling layer can refer to the implementation method of the first pooling layer in this embodiment, and this application will not go into details.

[0055] Specifically, the pooling process allows the model to discard information lacking characteristic parameters in certain areas during training, improving its ability to distinguish characteristic information, thereby alleviating overfitting and improving model training efficiency. When applied, it can compress and filter the feature maps output by the convolutional layer, enhancing the robustness of the model.

[0056] As a specific embodiment, taking the input of 64×64 thermal imaging temperature distribution map samples as the input of the CNN model as an example, the processing flow of the CNN model is as follows: Figure 5 As shown in the figure, the 64×64 thermal imaging temperature distribution map first undergoes feature extraction using a 5×5 convolution kernel in the first convolutional layer C1. It is then pooled using a 2×2 pooling window in the first pooling layer P1. It then undergoes feature extraction using a 3×3 convolution kernel in the second convolutional layer C2. It is then pooled using a 2×2 pooling window in the second pooling layer P2. Finally, the map is output to a fully connected layer with 128 neurons for feature integration. The integrated feature vector is then subjected to dropout (a regularization technique) and output to a fully connected layer with 3 neurons. The softmax function in the classification layer then converts the thermal imaging temperature distribution map into a probability distribution corresponding to each category space. The category with the highest probability is then output as the detection result.

[0057] See also Figure 6 As shown, Figure 6 This is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. To solve the above technical problems, an embodiment of the present invention further provides an electronic device, including: Memory 60, for storing computer programs; The processor 61 is configured to execute a computer program to implement the steps of the aforementioned abnormal fever detection method.

[0058] The electronic device provided in this embodiment may include but is not limited to a smart phone, a tablet computer, a laptop computer, or a desktop computer.

[0059] The processor 61 may include one or more processing cores, such as a quad-core processor or an octa-core processor. The processor 61 may be implemented using at least one of the following hardware forms: a digital signal processing (DSP), a field-programmable gate array (FPGA), or a programmable logic array (PLA). The processor 61 may also include a main processor and a coprocessor. The main processor is a processor for processing data in the awake state, also known as a central processing unit (CPU); the coprocessor is a low-power processor for processing data in the standby state. In some embodiments, the processor 61 may be integrated with a graphics processing unit (GPU), which is responsible for rendering and drawing content required to be displayed on the display screen. In some embodiments, the processor 61 may also include an artificial intelligence (AI) processor for handling computational operations related to machine learning.

[0060] The memory 60 may include one or more computer-readable storage media, which may be non-transitory. The memory 60 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In this embodiment, the memory 60 is at least used to store the following computer program 601, wherein, after the computer program is loaded and executed by the processor 61, it can implement the relevant steps of the fever anomaly detection method disclosed in any of the aforementioned embodiments. In addition, the resources stored in the memory 60 may also include an operating system 602 and data 603, etc., and the storage method may be temporary storage or permanent storage. Among them, the operating system 602 may include Windows, Unix, Linux, etc. The data 603 may include but is not limited to data in the fever anomaly detection method, etc.

[0061] In some embodiments, the electronic device may further include a display screen 62 , an input / output interface 63 , a communication interface 64 , a power supply 65 , and a communication bus 66 .

[0062] Those skilled in the art will understand that Figure 6 The structure shown in the figure does not constitute a limitation of the electronic device, and may include more or fewer components than shown in the figure.

[0063] For descriptions of features in the electronic device provided by the embodiments of the present invention, reference may be made to the relevant descriptions of the embodiments of the method for detecting abnormal heating, which will not be described in detail here.

[0064] It is understandable that if the fever anomaly detection method in the above embodiment 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 invention, or the part that contributes to the current technology, 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 and executes all or part of the steps of the methods of each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), electrically erasable programmable ROM, register, hard disk, removable disk, CD-ROM, magnetic disk or optical disk, etc. Various media that can store program code.

[0065] To solve the above technical problems, an embodiment of the present invention further provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the above-mentioned fever anomaly detection method are implemented.

[0066] For descriptions of features in the computer-readable storage medium provided in the embodiments of the present invention, reference may be made to the relevant descriptions of the embodiments of the fever anomaly detection method, which will not be described in detail here.

[0067] An embodiment of the present invention further provides a device for detecting abnormal heating, including:

[0068] a data acquisition unit configured to acquire a data set; samples in the data set include a thermal imaging temperature distribution map of a normal board card, a thermal imaging temperature distribution map of an abnormal board card, a heating condition of the normal board card, and a heating condition of the abnormal board card; wherein the heating condition of the normal board card indicates normal heating of the board card, and the heating condition of the abnormal board card includes abnormal heating of the board card and the location of the abnormal heating point;

[0069] a pre-processing unit, configured to perform consistency processing on any thermal imaging temperature distribution map in the data set; wherein the processing dimension of the consistency processing includes at least one of spatial resolution, brightness and / or contrast;

[0070] A training unit, used to train a convolutional neural network model based on the consistency-processed dataset;

[0071] A target acquisition unit, used to acquire a thermal imaging temperature distribution map of a target board;

[0072] The detection unit is used to process the thermal imaging temperature distribution map of the target board using the trained convolutional neural network model to obtain the heating condition of the target board.

[0073] For the description of the features of the abnormal heating detection device provided in the embodiment of the present invention, please refer to the relevant description of the embodiment of the abnormal heating detection method, which will not be repeated here.

[0074] An embodiment of the present invention further provides a computer program product, including a computer program / instruction, which, when executed by a processor, implements the steps of the fever anomaly detection method of the above embodiment.

[0075] For descriptions of features in the computer program product provided by the embodiments of the present invention, reference may be made to the relevant descriptions of the embodiments of the fever anomaly detection method, which will not be described in detail here.

[0076] The above describes in detail the method, device, and medium for detecting abnormal heating provided by the embodiments of the present invention. The various embodiments are described in a progressive manner throughout this specification, with each embodiment focusing on the differences from the other embodiments. For similar or identical parts between the various embodiments, reference can be made to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple. For relevant details, refer to the method description.

[0077] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the above description has generally described the components and steps of each example according to their functions. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present invention.

[0078] The above is a detailed introduction to the method, device and medium for detecting abnormal fever provided by the present invention. Specific examples are used herein to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method and core ideas of the present invention. It should be pointed out that for ordinary technicians in this technical field, without departing from the principles of the present invention, the present invention can also be improved and modified in several ways, and these improvements and modifications also fall within the scope of protection of the present invention.

Claims

1. A method for detecting abnormal fever, characterized in that: include: Acquire a data set; samples in the data set include a thermal imaging temperature distribution map of a normal board card, a thermal imaging temperature distribution map of an abnormal board card, a heating condition of the normal board card, and a heating condition of the abnormal board card; wherein the heating condition of the normal board card is normal heating of the board card, and the heating condition of the abnormal board card includes abnormal heating of the board card and the location of the abnormal heating point; Performing consistency processing on any thermal imaging temperature distribution map in the data set; wherein the processing dimension of the consistency processing includes at least one of spatial resolution, brightness and / or contrast; Train the convolutional neural network model based on the consistent processed dataset; Obtain thermal imaging temperature distribution map of target board; The trained convolutional neural network model is used to process the thermal imaging temperature distribution map of the target board to obtain the heating condition of the target board.

2. The method for detecting abnormal fever according to claim 1, wherein: Consistency processing is performed on any thermal imaging temperature distribution map in the data set, including: For any thermal imaging temperature distribution map in the data set, determining the spatial resolution of the thermal imaging temperature distribution map, and adjusting the spatial resolution of the thermal imaging temperature distribution map to a preset pixel range; Obtaining brightness and / or contrast of all the thermal imaging temperature distribution maps; The pixel value of each pixel in the thermal imaging temperature distribution map is adjusted to adjust the brightness and / or contrast of all the thermal imaging temperature distribution maps to the corresponding brightness target value and / or contrast target value.

3. The method for detecting abnormal fever according to claim 1, wherein: Before performing consistency processing on any thermal imaging temperature distribution map in the data set, the method further includes: For any thermal imaging temperature distribution map in the data set, determining a plurality of filtering regions of the thermal imaging temperature distribution map with any pixel in the thermal imaging temperature distribution map as the center; For any filter area, a Gaussian kernel is used to cover the filter area; Performing weighted summation on the temperature values corresponding to all pixels in the filtering area according to the weight of the Gaussian kernel to obtain a filtered temperature value corresponding to the central pixel of the filtering area; The filtered temperature value is used to update the temperature value of each pixel in the thermal imaging temperature distribution map to obtain a filtered thermal imaging temperature distribution map.

4. The method for detecting abnormal fever according to claim 1, wherein: After obtaining the dataset, it also includes: Rotating and / or flipping and / or enlarging and / or reducing the thermal imaging temperature distribution map in the data set to obtain a transformed image; The transformed image and the fever condition corresponding to the transformed image are added to the data set as new samples.

5. The method for detecting abnormal fever according to claim 1, wherein: The data set includes a training set, a validation set and a test set; The convolutional neural network model is trained based on the consistent dataset, including: Inputting the thermal imaging temperature distribution map in the training set into the convolutional neural network model to obtain a detection result of the convolutional neural network model on the thermal imaging temperature distribution map in the training set; Iterating model parameters of the convolutional neural network model according to the difference between the detection result and the fever condition corresponding to the training set; wherein the model parameters include weights of the convolutional layer, biases of the convolutional layer, weights of the fully connected layer, and biases of the fully connected layer; Inputting the thermal imaging temperature distribution map in the validation set into the convolutional neural network model to obtain a validation result of the validation set; Adjusting the hyperparameters of the convolutional neural network model according to the verification results; wherein the hyperparameters include the size of the convolution kernel in the convolution layer, the size of the pooling window in the pooling layer, and the learning rate of the convolutional neural network model; After training the convolutional neural network model based on the consistent dataset, it also includes: Input the thermal imaging temperature distribution map in the test set into the trained convolutional neural network model to obtain the test results of the test set; Determining whether the model performance of the trained convolutional neural network model reaches the target value based on the test results; If not, the model architecture of the convolutional neural network model is adjusted, and the process jumps back to the step of training the convolutional neural network model based on the consistency-processed data set.

6. The method for detecting abnormal fever according to any one of claims 1 to 5, characterized in that: The specific process of processing the thermal imaging temperature distribution map of the target board by the trained convolutional neural network model includes: Using a plurality of first convolution kernels to perform a convolution operation on the thermal imaging temperature distribution map obtained through the input layer, so as to extract features from the thermal imaging temperature distribution map to obtain a plurality of first layer feature maps; For any first-layer feature map, a first pooling window is used to perform a pooling operation on the first-layer feature map to obtain a first-layer pooled feature map; Performing a convolution operation on the first-layer pooled feature map using a second convolution kernel to obtain a second-layer feature map; wherein the size of the second convolution kernel is smaller than the size of the first convolution kernel; Using the second pooling window to perform a pooling operation on the second layer feature map to obtain a second layer pooled feature map; Use a fully connected layer to integrate all the second-layer pooled feature maps into a feature vector; The feature vector is mapped to a classification layer, so as to output the heating condition of the target board through the classification layer.

7. The method for detecting abnormal fever according to claim 6, wherein: A plurality of first convolution kernels are used to perform a convolution operation on the thermal imaging temperature distribution map obtained through the input layer to extract features of the thermal imaging temperature distribution map to obtain a plurality of first layer feature maps, including: Configuring a plurality of first convolution kernels for the thermal imaging temperature distribution map obtained through the input layer; wherein the plurality of first convolution kernels extract different types of features; the feature types include edges, extreme values, and textures; For any first convolution kernel, using the first convolution kernel to cover a local area on the thermal imaging temperature distribution map; Performing a convolution operation on the local area covered by the first convolution kernel to extract image features of the local area; controlling the first convolution kernel to slide according to a first preset step size to cover another local area on the thermal imaging temperature distribution map, and jumping to the step of performing a convolution operation on the local area covered by the first convolution kernel until the first convolution kernel covers the thermal imaging temperature distribution map; The image features of all the local areas are combined to obtain a first layer feature map generated by the first convolution kernel.

8. The method for detecting abnormal fever according to claim 6, wherein: The first layer feature map is pooled using the first pooling window to obtain the first layer pooled feature map, including: Using a first pooling window to cover a pooling area on the first layer feature map; Performing a pooling operation on the pooling area covered by the first pooling window to obtain a eigenvalue corresponding to the pooling area; Controlling the first pooling window to slide according to a second preset step size to cover another pooling area on the first layer feature map, and jumping to the step of performing a pooling operation on the pooling area covered by the first pooling window until the first pooling window covers the first layer feature map; The eigenvalues corresponding to all pooled areas are combined to obtain a first-layer pooled feature map generated by the first pooling window.

9. An electronic device, characterized in that: include: memory for storing computer programs; A processor is configured to execute the computer program to implement the steps of the fever abnormality detection method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps of the fever abnormality detection method according to any one of claims 1 to 8 are implemented.

Images