A power detection apparatus and method

Abstract

The application provides a kind of power detection equipment and method, the distribution difference of temperature gradient between each segmentation sub-block in infrared image is determined to inhibit coefficient of hot spot interference in infrared image;Infrared image is carried out multi-scale feature decomposition, and then extract high-frequency feature under thermal response and low-frequency feature under thermal background of power equipment, inhibit hot spot interference in high-frequency feature by inhibition coefficient, obtain high-frequency suppression feature;Further determine the thermal distribution of regional energy between low-frequency feature and high-frequency suppression feature, and then according to thermal distribution, low-frequency feature and high-frequency suppression feature are carried out feature level fusion, and obtain the fusion feature map of power equipment thermal fault detection;Further, the measured value of power equipment thermal fault is determined by fusion feature map, and then the thermal fault state of power equipment is determined based on measured value.The scheme of the application can realize the confidence detection of power equipment thermal fault based on the high and low frequency features of infrared image.

Description

Technical Field

[0001] This application relates to the field of power technology, and more specifically, to a power detection device and method. Background Technology

[0002] Power equipment is subjected to high voltage and high current environments for extended periods during operation, making it highly susceptible to localized overheating due to poor conductivity, component aging, or external environmental factors. This can lead to thermal faults, a common hidden danger in power systems. Failure to identify and address thermal faults in a timely manner can result in serious consequences such as equipment damage, power outages, or even fires. Compared to traditional manual inspections, infrared imaging-based thermal fault detection technology enables non-contact, real-time monitoring of surface temperature distribution on equipment, effectively identifying early signs of faults. This is a key means to improve the intelligence level of equipment operation and maintenance and ensure the safe and stable operation of the power system.

[0003] In existing technologies, the core technical bottleneck of current thermal fault detection methods lies in the lack of multi-domain feature collaborative sensing capabilities. Specifically, existing technologies fail to effectively achieve cross-domain collaborative fusion of frequency domain feature complementarity and spatial domain anti-interference capabilities. At the frequency domain feature extraction level, the high- and low-frequency mixed characteristics of thermal response signals are not fully explored. Detection models often sever the coupling relationship between high-frequency texture features and low-frequency background features, resulting in a single feature representation dimension and insufficient detection confidence. Furthermore, at the spatial domain anti-interference level, pseudo-hotspot noise generated by complex environments overlaps with real thermal fault signals in spatial distribution. Existing methods lack cross-domain feature verification mechanisms and cannot achieve dynamic suppression of interference signals through joint frequency-spatial domain analysis. This defect of multi-domain feature decoupling makes it difficult for the detection system to fully extract the essential features of thermal faults and effectively resist environmental noise interference, ultimately leading to a significant decrease in detection accuracy under complex operating conditions. Therefore, how to achieve confident detection of thermal faults in power equipment based on the high- and low-frequency features of infrared images has become a challenge for the industry. Summary of the Invention

[0004] This application provides a power detection device and method that can realize the confidence detection of thermal faults in power equipment based on the high and low frequency characteristics of infrared images.

[0005] In a first aspect, this application provides a thermal fault detection method for detecting thermal faults in power equipment, comprising the following steps: Initiate thermal fault detection of power equipment and acquire infrared images of the power equipment; Information on hot spot regions caused by thermal reflection from power equipment is extracted from the infrared image. The infrared image is then segmented into multiple sub-blocks based on the hot spot region information. The suppression coefficient of hot spot interference in the infrared image is determined by the difference in temperature gradient distribution between the sub-blocks. The infrared image is subjected to multi-scale feature decomposition to extract the high-frequency features of the power equipment under thermal response and the low-frequency features under thermal background. The hot spot interference in the high-frequency features is suppressed by the suppression coefficient to obtain the high-frequency suppressed features. The thermal distribution of regional energy between the low-frequency feature and the high-frequency suppression feature is determined, and then the low-frequency feature and the high-frequency suppression feature are fused at the feature level based on the thermal distribution to obtain a fused feature map for thermal fault detection of power equipment. The measured values ​​of thermal faults in power equipment are determined by the fused feature map, and then the thermal fault status of the power equipment is determined based on the measured values.

[0006] Preferably, extracting information about hotspot areas caused by thermal reflection from power equipment in the infrared image specifically includes: Temperature semantic analysis was performed on the infrared image to extract the temperature semantics of thermal reflection from the power equipment. Temperature anomaly areas are filtered out from the infrared image by setting a temperature threshold; Based on the temperature semantics and the temperature anomaly region, information on the hot spot region caused by thermal reflection from power equipment is determined.

[0007] Preferably, the infrared image is segmented into multiple sub-blocks based on the hot spot region information, specifically including: Based on the hot spot region information, the segmented regions in the infrared image are initially delineated; Perform region clustering on the pixels in the segmented region to obtain multiple clustered regions; The infrared image is segmented based on all clustered regions to obtain multiple segmented sub-blocks.

[0008] Preferably, determining the suppression coefficient of hot spot interference in the infrared image based on the distribution difference of temperature gradients between the various segmented sub-blocks specifically includes: Determine the temperature gradient for each segmented sub-block; The interference score of hot spots on thermal fault detection in each segmented sub-block is determined based on the difference in the distribution of temperature gradient between each segmented sub-block and its adjacent segmented sub-blocks; The suppression coefficient of hot spot interference in the infrared image is determined by all interference scores.

[0009] Preferably, performing multi-scale feature decomposition on the infrared image to extract high-frequency features of the power equipment under thermal response and low-frequency features under thermal background specifically includes: The infrared image is subjected to wavelet transform to decompose the infrared image into feature information at multiple spatial resolution levels; Frequency domain filtering separates all feature information by frequency, yielding the high-frequency characteristics of the power equipment under thermal response and the low-frequency characteristics under thermal background.

[0010] Preferably, the high-frequency suppression feature obtained by suppressing hotspot interference in the high-frequency feature using the suppression coefficient specifically includes: Extract hot spot interference features from the high-frequency features; The hot spot interference feature is suppressed and adjusted by the suppression coefficient to obtain the suppressed hot spot interference feature; The high-frequency suppression characteristics are determined based on the suppressed hot spot interference characteristics and the high-frequency characteristics.

[0011] Preferably, determining the thermal distribution of regional energy between the low-frequency feature and the high-frequency suppression feature specifically includes: The infrared image of the power equipment is divided into grids to obtain multiple grid regions. Determine the energy difference between the low-frequency feature and the high-frequency suppression feature in each grid region; The thermal distribution of regional energy between the low-frequency feature and the high-frequency suppression feature is determined by all energy differences.

[0012] Preferably, determining the measured values ​​of thermal faults in power equipment through the fused feature map specifically includes: Pre-train a machine model for thermal fault detection in power equipment; The fused feature map is used as the input parameter of the machine model; The machine model outputs the measured values ​​of thermal faults in the power equipment.

[0013] Preferably, infrared images are acquired using an infrared imaging device.

[0014] Secondly, this application provides a power detection device, which includes a thermal fault detection unit, the thermal fault detection unit comprising: The acquisition module is used to acquire infrared images of the power equipment after the thermal fault detection of the power equipment is initiated. The processing module is used to extract hot spot area information caused by thermal reflection of power equipment in the infrared image, segment the infrared image into multiple sub-blocks based on the hot spot area information, and determine the suppression coefficient of hot spot interference in the infrared image by the distribution difference of temperature gradient between the sub-blocks. The processing module is also used to perform multi-scale feature decomposition on the infrared image, thereby extracting the high-frequency features of the power equipment under thermal response and the low-frequency features under thermal background, and suppressing hot spot interference in the high-frequency features through the suppression coefficient to obtain high-frequency suppressed features; The processing module is further configured to determine the thermal distribution of regional energy between the low-frequency feature and the high-frequency suppression feature, and then perform feature-level fusion of the low-frequency feature and the high-frequency suppression feature based on the thermal distribution to obtain a fused feature map for thermal fault detection of power equipment. The execution module is used to determine the measured value of thermal fault of power equipment through the fused feature map, and then determine the thermal fault state of power equipment based on the measured value.

[0015] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects: In this embodiment, after initiating thermal fault detection of the power equipment, an infrared image of the power equipment is acquired; hot spot region information caused by thermal reflection of the power equipment is extracted from the infrared image; the infrared image is segmented into multiple sub-blocks based on the hot spot region information; the suppression coefficient of hot spot interference in the infrared image is determined by the distribution difference of temperature gradient between the sub-blocks; multi-scale feature decomposition is performed on the infrared image to extract high-frequency features of the power equipment under thermal response and low-frequency features under thermal background; hot spot interference in the high-frequency features is suppressed by the suppression coefficient to obtain high-frequency suppressed features; the thermal distribution of regional energy between the low-frequency features and the high-frequency suppressed features is determined; and the low-frequency features and the high-frequency suppressed features are then fused at the feature level based on the thermal distribution to obtain a fused feature map for thermal fault detection of the power equipment; the measured value of the thermal fault of the power equipment is determined by the fused feature map; and the thermal fault state of the power equipment is determined based on the measured value.

[0016] Therefore, this application achieves feature-level fusion of low-frequency features and high-frequency suppression features through thermal distribution to obtain a fused feature map for detecting thermal faults in power equipment, and then determines the thermal fault state of the power equipment based on the fused feature map. First, by identifying non-real heat source areas in the infrared image caused by thermal reflection, preliminary isolation of hot spot interference is achieved. Furthermore, by combining block segmentation and temperature gradient analysis to obtain the thermal distribution differences between different sub-blocks, a suppression coefficient reflecting the intensity of hot spot interference is quantified and constructed. This facilitates dynamic adjustment of the interference suppression degree in subsequent feature processing, reducing the possibility of misjudging thermal faults due to thermal reflection from the source and improving the discrimination accuracy and reliability of the entire detection process. Second, by using multi-scale feature decomposition technology, high-frequency and low-frequency features are extracted from the infrared image, fully considering the multi-dimensional characteristics of thermal response and avoiding the high-frequency and low-frequency feature segmentation in traditional methods. This paper addresses the limitations of traditional infrared imaging by suppressing noise interference from hot spots in high-frequency features through a suppression coefficient. This results in more accurate representation of high-frequency features, improving the resolution of infrared images and the confidence level of thermal fault detection, thus ensuring effective fusion of high-frequency textures and low-frequency background. Next, a thermal distribution map is constructed by analyzing the energy differences between low-frequency and high-frequency suppressed features within the grid area, reflecting the thermal contribution of each region. This is then combined with an attention mechanism to achieve feature-level fusion, dynamically weighting and combining the two types of features to construct a more representative fused feature map. This process enhances the expressive power of thermal fault-sensitive areas and reduces interference from irrelevant areas, making the final output feature map more discriminative and accurate in representing real thermal faults. Finally, based on the fused feature map, the thermal fault state of the power equipment is accurately determined, improving not only the sensitivity of fault identification but also enhancing the model's adaptability to complex environments. In summary, this application's solution can achieve confident detection of thermal faults in power equipment based on the high and low frequency features of infrared images. Attached Figure Description

[0017] Figure 1 This is an exemplary flowchart of a thermal fault detection method according to some embodiments of this application; Figure 2 This is a flowchart illustrating the determination of high-frequency suppression features according to some embodiments of this application; Figure 3 This is a schematic flowchart illustrating the determination of heat distribution according to some embodiments of this application; Figure 4 This is a schematic diagram of the structure of a thermal fault detection unit according to some embodiments of this application; Figure 5 This is a schematic diagram of the structure of a computer device implementing a thermal fault detection method according to some embodiments of this application. Detailed Implementation

[0018] To better understand the technical solution of this application, the technical solution of this application will be described in detail below with reference to the accompanying drawings and specific embodiments.

[0019] refer to Figure 1 The figure is an exemplary flowchart of a thermal fault detection method according to some embodiments of this application. The thermal fault detection method 100 mainly includes the following steps: In step 101, thermal fault detection of the power equipment is initiated, and infrared images of the power equipment are acquired.

[0020] It should be noted that thermal fault detection of power equipment refers to the identification and analysis of abnormal heat distribution generated by power equipment during operation using infrared imaging technology, thereby discovering local overheating phenomena caused by poor contact, overload, or insulation aging, and realizing early warning and condition assessment of potential faults.

[0021] In practice, infrared imaging equipment can be deployed to collect thermal radiation from operating power equipment from a distance without contact, and the collected images can be used as infrared images of the power equipment.

[0022] In step 102, hot spot region information caused by thermal reflection of power equipment in the infrared image is extracted. The infrared image is then segmented into multiple sub-blocks based on the hot spot region information. The suppression coefficient of hot spot interference in the infrared image is determined by the distribution difference of temperature gradient between the sub-blocks.

[0023] In some embodiments, extracting hot spot region information caused by thermal reflection from power equipment in the infrared image can be achieved using the following steps: Temperature semantic analysis was performed on the infrared image to extract the temperature semantics of thermal reflection from the power equipment. Temperature anomaly areas are filtered out from the infrared image by setting a temperature threshold; Based on the temperature semantics and the temperature anomaly region, information on the hot spot region caused by thermal reflection from power equipment is determined.

[0024] It should be noted that, in this application, thermal reflection refers to the phenomenon where infrared energy radiated by an external heat source is reflected by the surface of the power equipment and captured by an infrared imaging device, thereby generating a high-temperature area in the infrared image that is not its own heat source; it should also be noted that, in this application, temperature semantics refers to the semantic features corresponding to the temperature changes on the surface of the power equipment caused by the reflection of radiation from an external heat source in the infrared image; and hot spot area information in this application refers to the temperature anomaly area in the infrared image caused by thermal reflection, which is usually manifested as a hot spot area with a higher temperature.

[0025] In specific implementation, temperature semantic analysis of the infrared image is performed to extract the temperature semantics of heat reflection from the power equipment. This can be achieved in the following way: A deep learning-based infrared image semantic segmentation model (such as U-Net or DeepLabV3+) can be used to identify temperature patterns in different parts of the infrared image. Combined with labels representing the operating characteristics of the power equipment, the infrared image semantic segmentation model extracts the semantics of regions that may have heat reflection features, such as reflective areas on metal surfaces with high brightness but no heat source support. The extracted regional semantics are then used as the temperature semantics of the power equipment's heat reflection. It should be noted that, in this application, the labels representing the operating characteristics of the power equipment refer to introducing the functional attributes and typical heating behaviors of power equipment components as auxiliary labels when training the semantic segmentation model. For example, busbars and joints are usually real heat sources, while metal casings are prone to heat reflection. These labels guide the semantic segmentation model in establishing the relationship between regional semantics and actual heat reflection during the learning process. The correspondence between source types is used to improve the accuracy of semantic segmentation models in identifying heat reflection areas. It should also be noted that the infrared image semantic segmentation model in this application refers to the use of deep learning to classify each pixel in the infrared image, thereby achieving accurate division of semantic regions such as the device body, thermal fault area, and heat reflection area. Its technical principle is based on the convolutional neural network structure. The encoder extracts multi-scale spatial and temperature features in the infrared image, and the decoder restores these features into pixel-level label maps with the same size as the original image. Representative models include U-Net and DeepLabV3+. The former uses symmetric skip connections to enhance the ability to locate details, while the latter introduces dilated convolution to obtain multi-scale contextual information. During the training phase, these models learn the mapping relationship between temperature patterns and spatial structures by inputting a large number of semantically labeled infrared images. During the inference phase, they can achieve automatic identification and extraction of specific areas such as heat reflection in the infrared image.

[0026] Additionally, it should be noted that in this embodiment, pixel values ​​in the infrared image are pre-mapped to temperature, meaning each pixel value corresponds to a temperature value. Specifically, filtering temperature anomaly regions from the infrared image using a set temperature threshold can be achieved as follows: High-temperature regions are extracted from the infrared image by setting a dynamic temperature threshold (e.g., mean ± 3 standard deviation method), and these extracted high-temperature regions are used as temperature anomaly regions. Determining hotspot region information caused by thermal reflection from power equipment based on the temperature semantics and the temperature anomaly regions can be achieved as follows: The temperature semantics are converted into a semantic mask of thermal reflection from power equipment, and the temperature anomaly regions are converted into a temperature anomaly mask. Then, a logical intersection operation is performed on the semantic mask and the temperature anomaly mask, retaining only the region information that simultaneously possesses thermal reflection semantic features and temperature anomaly response. This retained region information is used as the hotspot region information caused by thermal reflection from power equipment. Further explanation is needed: Converting the temperature semantics into a semantic mask of thermal reflection from power equipment can be achieved as follows: After completing pixel-level temperature semantic annotation of the infrared image using a semantic segmentation model (such as U-Net), the model can be... The output is converted into a semantic mask, which is a two-dimensional matrix with the same size as the original image. The value of each pixel represents its semantic category. To identify the heat reflection area of ​​power equipment, a specific "heat reflection" category label needs to be assigned to the semantic segmentation model during training. During the inference phase, the model output is post-processed, retaining only pixels identified as "heat reflection" and setting other categories to zero or background. The conversion process includes: first, selecting the channel corresponding to the heat reflection category based on the probability map output by the model; second, setting a threshold (e.g., the heat reflection category in the softmax output). The probability is greater than 0.5) to generate a binary mask; third, the mask image is used as the spatial representation result of the thermal reflection semantic region for subsequent logical intersection operations and other operations; in addition, it should be noted that the specific implementation of converting the temperature anomaly region into a temperature anomaly mask image is as follows: first, map each pixel value in the infrared image to the corresponding temperature value, then calculate the mean and standard deviation of the temperature in the entire infrared image, set a dynamic temperature threshold (such as the mean plus or minus 3 times the standard deviation), perform threshold segmentation on the temperature image, set the pixels higher than the upper threshold to 1, and set the rest to 0, and the generated binary image is the temperature anomaly mask image.

[0027] In some embodiments, segmenting the infrared image into multiple sub-blocks based on the hot spot region information can be achieved using the following steps: Based on the hot spot region information, the segmented regions in the infrared image are initially delineated; Perform region clustering on the pixels in the segmented region to obtain multiple clustered regions; The infrared image is segmented based on all clustered regions to obtain multiple segmented sub-blocks.

[0028] It should be noted that the segmented sub-blocks in this application refer to image blocks segmented within the hot spot region.

[0029] In specific implementation, the preliminary delineation of the segmented regions in the infrared image based on the hotspot region information can be achieved in the following way: the hotspot region in the hotspot region information can be used as the segmented region in the infrared image. It should be noted that the hotspot region in this application is the region after edge contour smoothing. The pixel clustering in the segmented region to obtain multiple clustered regions can be achieved in the following way: the K-means clustering algorithm can be used to assign the pixels in the segmented region to multiple clustered regions. In the K-means clustering process, clustering is performed based on the gradient and spatial proximity between pixels. The segmentation of the infrared image based on all clustered regions to obtain multiple segmented sub-blocks can be achieved in the following way: the infrared image can be segmented according to the edges of each clustered region, and the segmented blocks can be used as segmented sub-blocks. Each segmented sub-block contains pixels with similar thermal distribution characteristics.

[0030] In some embodiments, determining the suppression coefficient of hot spot interference in the infrared image based on the distribution difference of temperature gradients between the various segmented sub-blocks can be achieved using the following steps: Determine the temperature gradient for each segmented sub-block; The interference score of hot spots on thermal fault detection in each segmented sub-block is determined based on the difference in the distribution of temperature gradient between each segmented sub-block and its adjacent segmented sub-blocks; The suppression coefficient of hot spot interference in the infrared image is determined by all interference scores.

[0031] It should be noted that the temperature gradient in this application is an indicator for measuring the spatial variation characteristics of heat distribution in the segmented sub-blocks; the interference score in this application refers to a numerical indicator used to quantitatively evaluate the degree of interference caused by a certain region in the image to the thermal fault detection results; and the suppression coefficient in this application refers to an adjustment factor used to weaken hot spot interference during the thermal fault feature extraction process, reflecting its influence on the final detection results.

[0032] In specific implementation, the temperature gradient of each segmented sub-block can be determined as follows: For each segmented sub-block, an image gradient operator (such as Sobel or Laplacian) can be used to extract the gradient change matrix of the pixels in each segmented sub-block, and this gradient change matrix can then be used as the temperature gradient of each segmented sub-block. The interference score of hot spots in each segmented sub-block for thermal fault detection can be determined based on the difference in temperature gradient distribution between each segmented sub-block and its adjacent segments as follows: A temperature gradient comparison method between adjacent segmented sub-blocks can be used to assess whether there are abrupt changes or abnormal deviations in each segmented sub-block. If the gradient of a certain segmented sub-block is significantly higher than that of the surrounding area and lacks heat source support, it can be preliminarily determined that the segmented sub-block is a thermal reflection interference sub-block, and an interference score can be assigned accordingly. Statistical indicators such as Z-score can be introduced when performing interference scoring. Quantization is performed specifically by introducing Z-score. First, the average level and range of temperature gradient for each segmented sub-block are statistically analyzed. Then, the temperature gradient of each segmented sub-block is compared with the average value to measure its deviation. The greater the deviation, the more significant the difference between the temperature gradient of the sub-block and the surrounding area. It is more likely to be an abnormal area caused by non-thermal fault factors such as thermal reflection. Therefore, a higher interference score is assigned to achieve quantitative identification of hot spot interference. The suppression coefficient of hot spot interference in the infrared image can be determined by using all interference scores. This can be achieved by normalizing all interference scores using a maximum value normalization strategy, and then forming a matrix of the processed interference scores according to the positional relationship between the segmented sub-blocks. This matrix is ​​then used as the coefficient matrix for hot spot interference suppression, and the suppression coefficient of hot spot interference in the infrared image is described by this coefficient matrix.

[0033] In step 103, the infrared image is decomposed into multiple scale features to extract the high-frequency features of the power equipment under thermal response and the low-frequency features under thermal background. The hot spot interference in the high-frequency features is suppressed by the suppression coefficient to obtain the high-frequency suppressed features.

[0034] In some embodiments, performing multi-scale feature decomposition on the infrared image to extract high-frequency features of the power equipment under thermal response and low-frequency features under thermal background can be achieved by the following steps: The infrared image is subjected to wavelet transform to decompose the infrared image into feature information at multiple spatial resolution levels; Frequency domain filtering separates all feature information by frequency, yielding the high-frequency characteristics of the power equipment under thermal response and the low-frequency characteristics under thermal background.

[0035] It should be noted that the high-frequency features under thermal response in this application refer to the image detail features formed by temperature abrupt changes in local overheated areas; the low-frequency features under thermal background in this application refer to the background thermal field features that change gradually and have a continuous structure in the overall temperature distribution of the power equipment.

[0036] In specific implementation, wavelet transform is performed on the infrared image to decompose it into feature information at multiple spatial resolution levels. This can be achieved as follows: A two-dimensional discrete wavelet transform (DWT) can be used to decompose the infrared image. Utilizing its local analysis capabilities in multi-scale space, the infrared image is divided into sub-band images at different resolution levels, each corresponding to image features in different frequency ranges. Specifically, wavelet basis functions with good edge-preserving capabilities, such as Daubechies (db4) or Symlets (sym5), can be selected. Filtering and downsampling are performed according to the row and column directions of the infrared image, progressively decomposing the infrared image into a low-frequency sub-image (LL) and three high-frequency sub-images (LH, HL, HH), corresponding to edge detail features in the horizontal, vertical, and diagonal directions. Frequency domain filtering is then used to separate all feature information by frequency, obtaining the high-frequency features of the power equipment under thermal response and the high-frequency features under thermal response. Low-frequency features under thermal background can be achieved through the following steps: bandpass filtering is applied to the high-frequency sub-image to retain information about regions with significant abrupt changes, highlighting the local thermal response features of the power equipment. The features obtained after bandpass filtering are used as the high-frequency features of the power equipment under thermal response. Low-frequency filtering is applied to the low-frequency sub-image to extract large-scale, uniformly varying background thermal distribution features in the image. The features obtained after low-pass filtering are used as the low-frequency features of the power equipment under thermal background. It should be further noted that high-frequency features reflect the detailed changes of potential thermal fault points on the surface of the power equipment, while low-frequency features reflect the overall temperature background and thermal field equilibrium state of the equipment. Both can provide a high-quality, hierarchical image feature foundation for subsequent hot spot interference identification and multi-feature fusion.

[0037] In some embodiments, reference Figure 2 As shown in the figure, this is a flowchart illustrating the determination of high-frequency suppression features in some embodiments of this application. In this embodiment, the hot spot interference in the high-frequency features is suppressed by the suppression coefficient, and the high-frequency suppression features can be obtained by the following steps: In step 1031, hot spot interference features are extracted from the high-frequency features; In step 1032, the hot spot interference feature is suppressed and adjusted by the suppression coefficient to obtain the suppressed hot spot interference feature; In step 1033, high-frequency suppression features are determined based on the suppressed hot spot interference features and the high-frequency features.

[0038] It should be noted that the hot spot interference feature in this application refers to the abnormal region feature in the infrared image that has a local temperature change but no actual heat source support, caused by non-thermal factors such as thermal reflection; the high frequency suppression feature in this application refers to the feature obtained after interference suppression.

[0039] In specific implementation, hot spot interference features can be extracted from the high-frequency features as follows: Local sliding window analysis can be used to divide the high-frequency features (in the form of feature maps) into regions. Within each sliding window, local temperature gradient values ​​and grayscale statistical features are calculated. By comparing the temperature gradient differences between each sliding window region and its surrounding neighborhood, window regions with significant abrupt changes are identified. Further edge detection algorithms (such as Sobel and LoG) are used to refine the boundaries of these abrupt change regions, thus determining the shape and extent of potential hot spot regions. To further determine whether these regions represent actual heat source responses, the stability of grayscale fluctuations within the regions or their correlation with low-frequency background features can be detected to identify interference regions whose temperature responses lack heat source support, which are then used as hot spot interference features. Output; The suppression of the hot spot interference feature by the suppression coefficient can be achieved in the following way: the suppression coefficient can be used as the suppression guidance parameter of the hot spot interference feature. The feature of each segmented sub-block in the interference region can be adjusted by the expression form of the suppression coefficient (i.e., coefficient matrix). The adjustment method can be weighted. The adjusted hot spot interference feature is used as the suppressed hot spot interference feature. The high-frequency suppression feature can be determined based on the suppressed hot spot interference feature and the high-frequency feature in the following way: the hot spot interference feature in the high-frequency feature can be replaced by the suppressed hot spot interference feature. The replaced feature is then smoothed and the smoothed feature is used as the high-frequency suppression feature.

[0040] In step 104, the thermal distribution of regional energy between the low-frequency feature and the high-frequency suppression feature is determined, and then the low-frequency feature and the high-frequency suppression feature are fused at the feature level according to the thermal distribution to obtain a fused feature map for thermal fault detection of power equipment.

[0041] In some embodiments, reference Figure 3 As shown in the figure, this is a schematic flowchart of determining the thermal distribution in some embodiments of this application. In this embodiment, determining the thermal distribution of energy in the region between the low-frequency feature and the high-frequency suppression feature can be achieved by the following steps: The infrared image of the power equipment is divided into grids to obtain multiple grid regions. Determine the energy difference between the low-frequency feature and the high-frequency suppression feature in each grid region; The thermal distribution of regional energy between the low-frequency feature and the high-frequency suppression feature is determined by all energy differences.

[0042] It should be noted that the thermal distribution in this application is a characteristic indicator that measures the spatial variation of temperature energy in different regions of an infrared image.

[0043] In practical implementation, dividing the infrared image of the power equipment into grid regions can be achieved in the following way: the infrared image can be divided into several grid regions according to a preset size, with each grid region serving as an independent analysis unit to ensure the comparability of local thermal features; determining the energy difference between the low-frequency feature and the high-frequency suppression feature in each grid region can be achieved in the following way: within each grid region, the sum of the gray values ​​of the low-frequency feature and the high-frequency suppression feature is calculated, and the sum of the gray values ​​of the low-frequency feature is used as the characterization value of the low-frequency energy in each grid region, and the sum of the gray values ​​of the high-frequency suppression feature is used as the characterization value of the high-frequency energy in each grid region. The absolute difference between the characterization values ​​of the low-frequency energy and the high-frequency energy is then used as the energy difference in each grid region; the low-frequency feature and the high-frequency suppression feature are determined through all energy differences. The thermal distribution of energy between regions can be achieved by spatially mapping the energy differences of all grid regions in the form of a heatmap, and using the spatially mapped distribution map as the thermal distribution of energy between low-frequency features and high-frequency suppression features. It should be further explained that spatial mapping refers to the process of reorganizing and visualizing the numerical information of energy differences of all grid regions according to their corresponding spatial positions in the image. In this process, these energy differences can be filled back into their respective grid positions and restored to the image space according to the two-dimensional coordinate relationship of the image, finally forming a heatmap. In terms of implementation, matrix transformation, interpolation or image reconstruction algorithms can be used to ensure that the spatial mapping result maintains both numerical authenticity and intuitive spatial readability. This heatmap can clearly show the spatial distribution of energy differences in different regions, which is helpful for subsequent feature fusion or abnormal region identification.

[0044] In some embodiments, the fusion of the low-frequency features and the high-frequency suppression features based on the thermal distribution to obtain a fused feature map for thermal fault detection of power equipment can be achieved through the following steps: The thermal contribution of each grid region in the infrared image is determined based on the thermal distribution. The attention weights of the low-frequency feature and the high-frequency suppression feature in each grid region are determined based on all thermal contributions. The low-frequency features and the high-frequency suppression features are dynamically fused according to the grid region position by using all attention weights to obtain a fused feature map for thermal fault detection of power equipment.

[0045] It should be noted that the thermal contribution in this application is a metric that measures the importance of the raster region to the feature representation in the overall temperature and energy distribution of the image; the fusion feature map in this application refers to the comprehensive representation map generated by spatially weighting and fusing low-frequency features and high-frequency suppression features according to thermal distribution weights.

[0046] In specific implementation, determining the thermal contribution of each grid region in the infrared image based on the thermal distribution can be achieved in the following way: The energy difference value between low-frequency features and high-frequency features after hotspot suppression in each grid region can be obtained from the thermal distribution, and the energy difference value can be normalized to its maximum value. The normalized energy difference value can then be used as the thermal contribution of each grid region. Determining the attention weights of the low-frequency features and the high-frequency suppression features in each grid region based on all thermal contributions can be achieved in the following way: A grid region attention weight allocation mechanism can be constructed based on the thermal contribution, assigning dynamic weights to low-frequency features and high-frequency suppression features respectively. This allocation mechanism can be combined with classic attention modules (such as CBAM and SE modules) to construct a fusion attention network. By introducing the thermal contribution of the grid region as a guiding signal, the neural network can be guided to automatically learn which type of feature is more representative in each grid region. That is, when a certain grid region… If a region has a high thermal contribution and significant high-frequency suppression features, then the high-frequency suppression features are given a greater attention weight; conversely, the expression ratio of low-frequency features in the grid region is increased, thereby achieving adaptive allocation of spatially heterogeneous features. The low-frequency features and the high-frequency suppression features are dynamically fused according to the grid region position using all attention weights to obtain the fused feature map for power equipment thermal fault detection. This can be achieved in the following way: based on the above attention weights, the low-frequency features and the high-frequency suppression features are weighted and fused in each grid region. It should be further noted that the fusion method can be to use grid-by-grid channel weighted superposition or to construct a fusion convolution kernel to achieve local feature fusion. The final fused feature map is used as the fused feature map for power equipment thermal fault detection. This fused feature map retains the low-frequency temperature field background information and the high-frequency local thermal fault response features, while suppressing hot spot interference, which can greatly improve the ability to distinguish the real thermal anomaly areas of power equipment.

[0047] It should be noted that this scheme, by introducing a thermal distribution guidance mechanism, can fully utilize the overall thermal field background of the equipment reflected by low-frequency features and the prominent local anomaly information in high-frequency suppression features to achieve weighted fusion based on regional thermal contribution, thereby improving the spatial accuracy of feature expression and the sensitivity of fault perception. By dynamically allocating attention weights in each grid area, the fused features are more focused on areas where faults may exist, enhancing the expression intensity of thermal fault signals and effectively suppressing background redundancy interference. This significantly improves the intelligent identification performance and robustness of infrared images for thermal faults in power equipment under complex thermal environments.

[0048] In step 105, the measured value of thermal fault of power equipment is determined by the fused feature map, and then the thermal fault state of power equipment is determined based on the measured value.

[0049] In some embodiments, determining the measured values ​​of thermal faults in power equipment using the fused feature map can be achieved through the following steps: Pre-train a machine model for thermal fault detection in power equipment; The fused feature map is used as the input parameter of the machine model; The machine model outputs measured values ​​of thermal faults in power equipment.

[0050] In practical implementation, a deep learning model (such as ResNet or Transformer architecture) suitable for thermal fault detection in power equipment can be pre-trained. Using a fused feature map as input, the model learns discriminative patterns reflecting thermal faults within the fused feature map. During pre-training, using labeled infrared image samples, the model optimizes its parameters through supervised learning, enabling it to accurately identify different types of thermal fault regions. During inference, the fused feature map obtained using the aforementioned attention mechanism is input into the model. The model extracts features and performs classification predictions for each region, ultimately outputting the measured value of the power equipment's thermal fault, such as the thermal fault level, location coordinates, or confidence probability, thus forming a quantitative diagnostic result. It should be noted that the machine model in this application is a detection model built on a deep learning architecture. Its core principle is to learn the nonlinear mapping relationship between input features and target results by continuously adjusting the parameters of the neural network during the training process of a large number of labeled samples. In the detection of thermal faults in power equipment, the model takes a fused feature map as input. This feature map integrates the high-frequency response features of the power equipment and the low-frequency features of the background, and contains rich spatial thermal distribution and local anomaly information. Through convolutional layers to extract multi-scale features, attention mechanisms to focus on key regions, and fully connected layers to achieve feature compression and classification, the model can effectively determine whether there is a thermal fault in the image, the fault level and its location, and achieve accurate quantitative output of thermal faults.

[0051] It should be noted that, in this application, determining the thermal fault status of power equipment based on the measured values ​​means using the numerical or probabilistic results output by the machine model to classify and judge whether the power equipment currently has a thermal fault and the severity of the fault, thereby giving a specific status identification conclusion of normal, warning or fault.

[0052] On the other hand, in some embodiments, this application provides a power detection device, which includes a thermal fault detection unit, with reference to... Figure 4 The figure is a schematic diagram of the structure of a thermal fault detection unit according to some embodiments of this application. The thermal fault detection unit 400 includes: an acquisition module 401, a processing module 402, and an execution module 403, which are described below: The acquisition module 401 in this application is mainly used to acquire infrared images of the power equipment after the thermal fault detection of the power equipment is started. Processing module 402, in this application, is used to extract hot spot area information caused by thermal reflection of power equipment in the infrared image, segment the infrared image into blocks according to the hot spot area information, obtain multiple segmented sub-blocks, and determine the suppression coefficient of hot spot interference in the infrared image by the distribution difference of temperature gradient between each segmented sub-block. In this application, the processing module 402 is also used to perform multi-scale feature decomposition on the infrared image, thereby extracting the high-frequency features of the power equipment under thermal response and the low-frequency features under thermal background, and suppressing the hot spot interference in the high-frequency features through the suppression coefficient to obtain high-frequency suppressed features; In this application, the processing module 402 is also used to determine the thermal distribution of regional energy between the low-frequency feature and the high-frequency suppression feature, and then perform feature-level fusion of the low-frequency feature and the high-frequency suppression feature based on the thermal distribution to obtain a fused feature map for thermal fault detection of power equipment. The execution module 403 in this application is mainly used to determine the measured value of thermal fault of power equipment through the fused feature map, and then determine the thermal fault state of power equipment based on the measured value.

[0053] In addition, this application also provides a computer device, the computer device including a memory and a processor, the memory storing code, and the processor being configured to acquire the code and execute the above-described thermal fault detection method.

[0054] In some embodiments, reference Figure 5 The figure is a schematic diagram of the structure of a computer device implementing a thermal fault detection method according to some embodiments of this application. The thermal fault detection method in the above embodiments can... Figure 5The computer device shown is used to implement this, and the computer device 500 includes at least one processor 501, a communication bus 502, a memory 503, and at least one communication interface 504.

[0055] Processor 501 can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0056] The communication bus 502 can be used to transmit information between the aforementioned components.

[0057] Memory 503 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital versatile optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto. Memory 503 may exist independently and be connected to processor 501 via communication bus 502. Memory 503 may also be integrated with processor 501.

[0058] The memory 503 stores program code that executes the solution of this application, and its execution is controlled by the processor 501. The processor 501 executes the program code stored in the memory 503. The program code may include one or more software modules. In the above embodiments, the thermal fault detection method can be implemented by the processor 501 and one or more software modules in the program code in the memory 503.

[0059] Communication interface 504 uses any transceiver-like device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0060] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0061] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This application does not limit the type of computer device.

[0062] In addition, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described thermal fault detection method.

[0063] Although preferred embodiments of this application have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments as well as all changes and modifications falling within the scope of this application.

[0064] Obviously, those skilled in the art can make various modifications and variations to this application without departing from the spirit and scope of this application. Therefore, if such modifications and variations fall within the scope of the claims of this application and their equivalents, this application also intends to include such modifications and variations.

Claims

1. A thermal fault detection method for detecting thermal faults in power equipment, characterized in that, Includes the following steps: Initiate thermal fault detection of power equipment and acquire infrared images of the power equipment; Information on hot spot regions caused by thermal reflection from power equipment is extracted from the infrared image. The infrared image is then segmented into multiple sub-blocks based on the hot spot region information. The suppression coefficient of hot spot interference in the infrared image is determined by the difference in temperature gradient distribution between the sub-blocks. The infrared image is subjected to multi-scale feature decomposition to extract the high-frequency features of the power equipment under thermal response and the low-frequency features under thermal background. The hot spot interference in the high-frequency features is suppressed by the suppression coefficient to obtain the high-frequency suppressed features. The thermal distribution of regional energy between the low-frequency feature and the high-frequency suppression feature is determined, and then the low-frequency feature and the high-frequency suppression feature are fused at the feature level based on the thermal distribution to obtain a fused feature map for thermal fault detection of power equipment. The measured values ​​of thermal faults in power equipment are determined by the fused feature map, and then the thermal fault status of the power equipment is determined based on the measured values.

2. The method as described in claim 1, characterized in that, Extracting information about hotspot areas caused by thermal reflection from power equipment in the infrared image specifically includes: Temperature semantic analysis was performed on the infrared image to extract the temperature semantics of thermal reflection from the power equipment. Temperature anomaly areas are filtered out from the infrared image by setting a temperature threshold; Based on the temperature semantics and the temperature anomaly region, information on the hot spot region caused by thermal reflection from power equipment is determined.

3. The method as described in claim 1, characterized in that, The infrared image is segmented into multiple sub-blocks based on the hot spot region information, specifically including: Based on the hot spot region information, the segmented regions in the infrared image are initially delineated; Perform region clustering on the pixels in the segmented region to obtain multiple clustered regions; The infrared image is segmented based on all clustered regions to obtain multiple segmented sub-blocks.

4. The method as described in claim 1, characterized in that, Determining the suppression coefficient of hot spot interference in the infrared image by the distribution difference of temperature gradient between each segmented sub-block specifically includes: Determine the temperature gradient for each segmented sub-block; The interference score of hot spots on thermal fault detection in each segmented sub-block is determined based on the difference in the distribution of temperature gradient between each segmented sub-block and its adjacent segmented sub-blocks; The suppression coefficient of hot spot interference in the infrared image is determined by all interference scores.

5. The method as described in claim 1, characterized in that, Performing multi-scale feature decomposition on the infrared image to extract high-frequency features of the power equipment under thermal response and low-frequency features under thermal background specifically includes: The infrared image is subjected to wavelet transform to decompose the infrared image into feature information at multiple spatial resolution levels; Frequency domain filtering separates all feature information by frequency, yielding the high-frequency characteristics of the power equipment under thermal response and the low-frequency characteristics under thermal background.

6. The method as described in claim 1, characterized in that, The high-frequency suppression features obtained by suppressing hotspot interference in the high-frequency features using the suppression coefficient specifically include: Extract hot spot interference features from the high-frequency features; The hot spot interference feature is suppressed and adjusted by the suppression coefficient to obtain the suppressed hot spot interference feature; The high-frequency suppression characteristics are determined based on the suppressed hot spot interference characteristics and the high-frequency characteristics.

7. The method as described in claim 1, characterized in that, Determining the thermal distribution of energy in the region between the low-frequency feature and the high-frequency suppression feature specifically includes: The infrared image of the power equipment is divided into grids to obtain multiple grid regions. Determine the energy difference between the low-frequency feature and the high-frequency suppression feature in each grid region; The thermal distribution of regional energy between the low-frequency feature and the high-frequency suppression feature is determined by all energy differences.

8. The method as described in claim 1, characterized in that, Determining the measured values ​​of thermal faults in power equipment through the fused feature map specifically includes: Pre-train a machine model for thermal fault detection in power equipment; The fused feature map is used as the input parameter of the machine model; The machine model outputs the measured values ​​of thermal faults in the power equipment.

9. The method as described in claim 1, characterized in that, Infrared images are acquired using infrared imaging equipment.

10. A power detection device, comprising a thermal fault detection unit, characterized in that, The thermal fault detection unit includes: The acquisition module is used to acquire infrared images of the power equipment after the thermal fault detection of the power equipment is initiated. The processing module is used to extract hot spot area information caused by thermal reflection of power equipment in the infrared image, segment the infrared image into multiple sub-blocks based on the hot spot area information, and determine the suppression coefficient of hot spot interference in the infrared image by the distribution difference of temperature gradient between the sub-blocks. The processing module is also used to perform multi-scale feature decomposition on the infrared image, thereby extracting the high-frequency features of the power equipment under thermal response and the low-frequency features under thermal background, and suppressing hot spot interference in the high-frequency features through the suppression coefficient to obtain high-frequency suppressed features; The processing module is further configured to determine the thermal distribution of regional energy between the low-frequency feature and the high-frequency suppression feature, and then perform feature-level fusion of the low-frequency feature and the high-frequency suppression feature based on the thermal distribution to obtain a fused feature map for thermal fault detection of power equipment. The execution module is used to determine the measured value of thermal fault of power equipment through the fused feature map, and then determine the thermal fault state of power equipment based on the measured value.

Images