Electrical fire hazard perception method and system based on infrared thermal imaging

Abstract

The application relates to the technical field of image processing, and discloses an electrical fire hazard perception method and system based on infrared thermal imaging. The method comprises the following steps: an infrared thermal imager collects an infrared image of electrical equipment and obtains a pretreatment image through adaptive filtering, a material distribution map is obtained through a material recognition network, a correction temperature data is obtained by applying a dynamic emissivity correction algorithm, equipment features are extracted for clustering and grouping, and a temperature standardization deviation value is calculated to determine a fire hazard. The application solves the problems of temperature measurement error caused by the emissivity difference of different materials in traditional infrared thermal imaging electrical fire monitoring and the problems of false positives and false negatives caused by the fixed threshold method, and improves the accuracy and intelligent level of electrical fire hazard perception.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and in particular to a method and system for sensing electrical fire hazards based on infrared thermal imaging. Background Technology

[0002] Currently, electrical fires are occurring frequently, ranking first among all types of fires. The main causes of electrical fires are short circuits, leakage, arcing, and overloads caused by faults in electrical wiring and equipment. Existing infrared thermal imaging technology has been widely used in the field of electrical equipment temperature monitoring. It identifies potential fire hazards by non-contactly detecting the surface temperature distribution of electrical equipment. This is particularly important for fire departments in various large-scale public events and major security tasks, where pre-event detection and in-event monitoring of electrical fire hazards are necessary.

[0003] Traditional infrared thermal imaging methods for electrical fire monitoring have significant technical shortcomings, primarily due to their reliance on fixed alarm temperatures, which easily leads to false alarms and missed alarms. Because of the complex and varied temperature environments at the scene, and the significant differences in the normal operating temperatures of different devices, using a uniform fixed temperature threshold cannot accommodate the temperature characteristics of different electrical equipment. For example, if device A's normal operating temperature is 200℃ while device B's is 60℃, setting the alarm temperature above 200℃ to avoid false alarms for device A will result in device B's temperature rising abnormally to 120℃ without triggering the alarm, leading to a missed alarm. Furthermore, traditional infrared thermal imaging technology has poor anomaly detection capabilities for similar devices, lacking intelligent identification methods to detect abnormal temperature increases in individual devices within the same category. Summary of the Invention

[0004] This application provides a method and system for detecting electrical fire hazards based on infrared thermal imaging, which solves the temperature measurement error caused by the difference in emissivity of different materials and the false alarm and missed alarm problems caused by the fixed threshold method in traditional infrared thermal imaging electrical fire monitoring, thereby improving the accuracy and intelligence level of electrical fire hazard detection.

[0005] In a first aspect, this application provides a method for detecting electrical fire hazards based on infrared thermal imaging, the method comprising:

[0006] Step S1: The infrared thermal imager acquires infrared image data of the monitoring area of ​​the electrical equipment, and the image noise is eliminated by adaptive median filtering to obtain a preprocessed infrared image;

[0007] Step S2: Input the preprocessed infrared image into the electrical equipment material recognition network, identify the material type labels of the electrical equipment in the image, and obtain the electrical equipment material distribution map;

[0008] Step S3: Based on the material type label in the electrical equipment material distribution map, apply the dynamic emissivity correction algorithm to calculate the corrected emissivity parameters of each pixel and obtain emissivity correction temperature data;

[0009] Step S4: Extract the geometric and temperature features of the electrical equipment based on the emissivity-corrected temperature data, and use a similarity metric to perform equipment feature matching to obtain clustering results for similar electrical equipment;

[0010] Step S5: Calculate the temperature standardization deviation of each device in the clustering results of the same type of electrical equipment. When the deviation value exceeds the preset abnormal threshold, it is determined as a fire hazard, and the electrical fire hazard perception result is obtained.

[0011] Secondly, this application provides an electrical fire hazard detection system based on infrared thermal imaging, the electrical fire hazard detection system based on infrared thermal imaging comprising:

[0012] The filtering module is used to collect infrared image data of the monitoring area of ​​electrical equipment by the infrared thermal imager, and to eliminate image noise through adaptive median filtering to obtain a pre-processed infrared image.

[0013] The input module is used to input the preprocessed infrared image into the electrical equipment material recognition network, identify the material type labels of the electrical equipment in the image, and obtain the material distribution map of the electrical equipment.

[0014] The calculation module is used to calculate the corrected emissivity parameters of each pixel based on the material type label in the material distribution map of the electrical equipment, and to obtain emissivity correction temperature data.

[0015] The matching module is used to extract the geometric and temperature features of electrical equipment based on the emissivity-corrected temperature data, perform equipment feature matching using similarity measures, and obtain clustering results of similar electrical equipment.

[0016] The judgment module is used to calculate the temperature standardization deviation value of each device in the clustering results of the same type of electrical equipment. When the deviation value exceeds the preset abnormal threshold, it is judged as a fire hazard, and the electrical fire hazard perception result is obtained.

[0017] Thirdly, an electrical fire hazard sensing device based on infrared thermal imaging is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor invokes the instructions in the memory to cause the electrical fire hazard sensing device based on infrared thermal imaging to execute the aforementioned electrical fire hazard sensing method based on infrared thermal imaging.

[0018] Fourthly, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, which, when executed on a computer, cause the computer to perform the above-described method for detecting electrical fire hazards based on infrared thermal imaging.

[0019] The technical solution provided in this application acquires infrared image data of the monitoring area of ​​electrical equipment using an infrared thermal imager and processes it through adaptive median filtering. The size of the filtering window is dynamically adjusted according to the variance of the pixel neighborhood, effectively eliminating image noise while maintaining the integrity of the edge details of the electrical equipment, thus laying a high-quality image foundation for accurate identification of electrical equipment made of different materials. The preprocessed infrared image is input into an electrical equipment material recognition network. Through depth feature extraction of five convolutional blocks and material category discrimination processing of fully connected classification layers, it can accurately identify six material types: metal shell, plastic insulation material, ceramic components, rubber seals, glass insulators, and composite material shell. The obtained electrical equipment material distribution map provides key material information support for solving the problem of different emissivity differences of different materials in traditional infrared thermal imaging technology. Based on the material type labels in the material distribution map of electrical equipment, a dynamic emissivity correction algorithm is applied. By querying the preset material emissivity parameter database, the reference emissivity value and temperature coefficient parameter corresponding to each material type are obtained. An emissivity correction function model is constructed to calculate the corrected emissivity parameter of each pixel. Combined with the Stefan-Boltzmann law, the true temperature value is solved in reverse. This fundamentally solves the temperature measurement error problem caused by the difference in material emissivity in traditional methods, and significantly improves the accuracy and reliability of infrared thermal imaging temperature measurement.

[0020] Based on emissivity-corrected temperature data, geometric and temperature features of electrical equipment are extracted. Tanimoto similarity measurement is used for equipment feature matching, and the similarity between equipment is quantified by the intersection and union ratio. The resulting clustering of similar electrical equipment overcomes the technical limitations of traditional methods in intelligently identifying similar equipment. In the specific application area of ​​electrical fire hazard detection, the core contribution of the dynamic emissivity correction algorithm lies in reducing the temperature measurement accuracy from ±5℃ using traditional fixed emissivity methods to ±1.2℃, providing a reliable data foundation for subsequent anomaly detection. The key role of the similarity measurement algorithm is its ability to automatically identify groups of electrical equipment with similar geometric and temperature features, establishing a scientific comparison standard for relative temperature anomaly analysis. The method calculates the standardized temperature deviation of each device in the clustering results of similar electrical equipment. When the deviation exceeds the preset abnormal threshold, it is judged as a fire hazard. This anomaly detection method based on relative temperature analysis completely gets rid of the limitations of the traditional fixed threshold method. By calculating the standardized deviation, it automatically adapts to the temperature characteristics of different types of electrical equipment, effectively solving the problem of false alarms and missed alarms caused by different equipment temperature references in the traditional method. It realizes intelligent perception of electrical fire hazards and significantly improves the accuracy and practicality of electrical fire prevention. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of one embodiment of the electrical fire hazard perception method based on infrared thermal imaging in this application.

[0023] Figure 2 This is a schematic diagram of one embodiment of the electrical fire hazard perception system based on infrared thermal imaging in this application.

[0024] Figure 3 This is a schematic block diagram of the electrical fire hazard sensing device based on infrared thermal imaging in an embodiment of the present invention. Detailed Implementation

[0025] This application provides a method and system for detecting electrical fire hazards based on infrared thermal imaging. The terms "first," "second," "third," "fourth," etc. (if present) in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments described herein can be implemented in a sequence other than that illustrated or described herein. Furthermore, the terms "comprising" or "having" and any variations thereof are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0026] For ease of understanding, the specific process of the embodiments of this application is described below. Please refer to [link / reference]. Figure 1 One embodiment of the electrical fire hazard perception method based on infrared thermal imaging in this application includes:

[0027] Step S1: The infrared thermal imager acquires infrared image data of the monitoring area of ​​the electrical equipment, and the image noise is eliminated by adaptive median filtering to obtain a preprocessed infrared image;

[0028] Step S2: Input the preprocessed infrared image into the electrical equipment material recognition network to identify the material type labels of the electrical equipment in the image and obtain the material distribution map of the electrical equipment;

[0029] Step S3: Based on the material type labels in the electrical equipment material distribution map, apply the dynamic emissivity correction algorithm to calculate the corrected emissivity parameters of each pixel and obtain emissivity correction temperature data;

[0030] Step S4: Extract the geometric and temperature features of electrical equipment based on emissivity-corrected temperature data, and use similarity measures to perform equipment feature matching to obtain clustering results of similar electrical equipment;

[0031] Step S5: Calculate the temperature standardization deviation of each device in the clustering results of similar electrical equipment. When the deviation exceeds the preset abnormal threshold, it is judged as a fire hazard, and the electrical fire hazard perception result is obtained.

[0032] It is understood that the executing entity of this application can be an electrical fire hazard detection system based on infrared thermal imaging, or it can be a terminal or a server; the specific implementation is not limited here. This application's embodiment uses a server as an example for illustration.

[0033] Specifically, continuous scanning and acquisition are performed on the monitoring area of ​​electrical equipment. A long-wave infrared sensor captures infrared radiation signals in the 8-14 micrometer band and converts them into digital image data. The original infrared image sequence contains radiation intensity information and preliminary temperature values ​​for each pixel. Adaptive median filtering dynamically determines the filtering window size by calculating the variance value within the neighborhood of each pixel. When the neighborhood variance is large, it indicates that there is more noise in the area, and the filtering window size is increased accordingly; conversely, the window size is reduced to preserve image details. The median filtering algorithm sorts all pixel values ​​within the window by numerical value and selects the median as the output value for the current pixel, effectively eliminating random noise and isolated noise points. Ambient temperature compensation processing is based on real-time ambient temperature data collected by a temperature sensor. The difference between the ambient temperature and the standard reference temperature is calculated, and then linear compensation is applied to the temperature reading of each pixel to eliminate the impact of ambient temperature changes on the accuracy of infrared temperature measurement.

[0034] Preprocessed infrared images are input into an electrical equipment material recognition network for deep feature extraction and classification. The input layer of the convolutional neural network receives 640×480 pixel three-channel infrared image data. Data format conversion normalizes the original pixel values ​​to the range of 0-1, ensuring the stability of network training. The five convolutional blocks employ a progressive feature extraction strategy. The first convolutional layer has a kernel size of 7×7 and a stride of 2, extracting low-level edge and texture features. Subsequent convolutional layers use 3×3 small convolutional kernels to extract more complex semantic features layer by layer. Each convolutional block contains convolution operations, batch normalization, and ReLU activation function processing. Batch normalization standardizes the feature distribution by calculating the mean and variance of each feature map, and the ReLU activation function sets negative values ​​to zero, enhancing the non-linear expressive power of the network. The fully connected classification layer maps the extracted deep feature vectors to six material categories, outputting the probability distribution of metal shells, plastic insulation materials, ceramic components, rubber seals, glass insulators, and composite material shells. The material classification probability results are normalized using the softmax function to ensure that the sum of the probabilities of all categories is 1, and the material type with the highest probability value is selected as the material label for that pixel region.

[0035] A dynamic emissivity correction algorithm is implemented based on the material distribution map of electrical equipment to address temperature measurement errors caused by differences in emissivity between different materials. A material emissivity parameter database pre-stores reference emissivity values ​​and temperature coefficient parameters for six electrical equipment materials. The reference emissivity for metal casings is 0.2, the primary temperature coefficient is 0.0008 degrees Celsius, and the secondary temperature coefficient is 1.2 × 10⁻⁶ degrees Celsius squared. The emissivity correction function model adopts a quadratic polynomial form. The initial temperature reading and material type label of each pixel are substituted into the correction function for calculation to obtain the dynamic emissivity value of that pixel at the current temperature. The emissivity compensation calculation is based on the Stefan-Boltzmann law. The original infrared radiance value is divided by the corrected emissivity value, multiplied by the standard emissivity coefficient, and finally the fourth root is taken to solve for the true temperature value. After correction, the pixel temperature matrix is ​​integrated according to the equipment area boundaries, and the average temperature, maximum temperature, and temperature distribution characteristic parameters of all pixels within each equipment area are calculated.

[0036] Features of electrical equipment are extracted and clustered based on emissivity-corrected temperature data. Geometric feature extraction utilizes equipment contour detection algorithms to obtain equipment boundary information, calculating contour area, aspect ratio, and shape complexity. The contour area is the total number of pixels within the equipment region multiplied by the actual area of ​​a single pixel. The aspect ratio is the ratio of the length to the width of the equipment's smallest bounding rectangle. The shape complexity is the ratio of the square of the contour perimeter to the contour area. Temperature feature parameters include the average temperature of the equipment region, temperature standard deviation, and temperature gradient. The temperature gradient is obtained by calculating the average temperature difference between adjacent pixels. A comprehensive feature vector combines geometric and temperature features into a nine-dimensional feature array, with normalization ensuring a consistent range for each dimension's feature values. Similarity is calculated using the intersection and union ratio. The intersection of two equipment feature vectors is the sum of the minimum values ​​of their corresponding dimensions, and the union is the sum of the maximum values. The similarity is equal to the ratio of the intersection to the union. Iterative clustering calculations group equipment based on the similarity matrix, setting a similarity threshold of 0.8. Equipment with similarity higher than the threshold is grouped into the same category.

[0037] Calculate the standardized temperature deviation of similar electrical equipment and determine fire hazards. Group temperature statistical characteristics are obtained by summing the temperatures of all equipment within each group and dividing by the number of equipment. The temperature standard deviation is obtained by summing the squares of the differences between each equipment temperature and the group average temperature, dividing by the number of equipment minus 1, and then taking the square root. The standardized temperature deviation of equipment is calculated by subtracting the group average temperature from the individual equipment temperature and then dividing by the group standard deviation to obtain a dimensionless standardized deviation value. An anomaly threshold is set at 2.5; when the standardized deviation of a piece of equipment exceeds this threshold, the equipment is marked as having an abnormal temperature state. Fire hazard risk level assessment combines equipment location coordinates, temperature deviation degree, and equipment type risk coefficient to calculate a comprehensive risk score, and classifies the risk into four levels: low risk, medium risk, high risk, and extremely high risk, based on the score range.

[0038] In one specific embodiment, step S1 includes:

[0039] Infrared thermal imagers continuously scan and collect data on the monitoring area of ​​electrical equipment to obtain raw infrared image sequences containing temperature information.

[0040] Calculate the neighborhood variance of each frame in the original infrared image sequence, determine the filter window size parameter based on the variance distribution, and obtain the adaptive filter parameter set.

[0041] An adaptive filtering parameter set is applied to the original infrared image sequence to remove noise, eliminating random noise and isolated noise points in the image, resulting in a denoised infrared image.

[0042] Based on real-time ambient temperature data collected by an ambient temperature sensor, temperature drift compensation processing is performed on the noise-reduced infrared image to obtain a preprocessed infrared image.

[0043] Specifically, when the infrared thermal imager continuously scans and collects data on the monitoring area of ​​electrical equipment, the long-wave infrared sensor captures electromagnetic radiation signals in the 8-14 micrometer band at a frequency of 30 frames per second. The infrared radiation intensity received by each pixel unit in the sensor array is converted into digital grayscale values ​​by an analog-to-digital converter, forming a sequence of original infrared images containing temperature information. Each frame contains 640×480 pixels, and each pixel corresponds to the infrared radiation intensity and preliminary temperature estimate at a specific location within the monitoring area.

[0044] Pixel neighborhood variance calculation is a core step in determining adaptive filtering parameters. For each frame in the original infrared image sequence, the algorithm traverses each pixel in the image, establishing a 3×3 or 5×5 neighborhood window centered on that pixel. The variance of all pixel grayscale values ​​within the window is calculated. Variance is obtained by first calculating the arithmetic mean of the pixel grayscale values ​​within the window, then calculating the square of the difference between each pixel's grayscale value and the mean, summing all squared values, and dividing by the total number of pixels minus one. The variance distribution reflects the noise level and detail complexity of local image regions. A large variance in a region indicates significant noise interference or rich edge details, requiring a larger filtering window for noise suppression. Conversely, a small variance indicates a relatively smooth region, necessitating a smaller filtering window to avoid over-smoothing and loss of detail. The filter window size parameter is mapped according to the distribution range of the variance value. A variance value in the range of 0-100 corresponds to a 3×3 filter window, a variance value in the range of 100-300 corresponds to a 5×5 filter window, and a variance value exceeding 300 corresponds to a 7×7 filter window. The adaptive filter parameter set contains the optimal filter window size information corresponding to each pixel position in the image.

[0045] An adaptive filtering parameter set is applied to noise removal of the original infrared image sequence using an improved version of the median filtering algorithm. Traditional median filtering uses a fixed-size filtering window, while adaptive median filtering dynamically adjusts the window size based on the variance characteristics of each pixel position. During filtering, the algorithm reads the filtering window size parameter corresponding to the current pixel position, establishes a neighborhood window of the corresponding size around the pixel, extracts the gray values ​​of all pixels within the window and sorts them by value. The value at the middle position after sorting is selected as the output value of that pixel. Median selection can effectively suppress the impact of salt-and-pepper noise and impulse noise on image quality. Random noise usually manifests as random fluctuations in pixel gray values, and isolated noise points manifest as individual pixel gray values ​​that deviate abnormally from the average level of surrounding pixels. Median filtering replaces these outliers with the median value of the local area through neighborhood statistical characteristics, eliminating noise interference while maintaining the integrity of image edges and details. The gray value of each pixel in the denoised infrared image is processed by median filtering, and the overall signal-to-noise ratio of the image is significantly improved.

[0046] Real-time ambient temperature data collected by an ambient temperature sensor is used to perform temperature drift compensation processing on the noise-reduced infrared image. Changes in ambient temperature affect the detector performance and optical system characteristics of the infrared thermal imager, leading to systematic deviations in temperature measurement results. The temperature sensor is placed near the infrared thermal imager and collects ambient temperature values ​​every 10 seconds. The temperature drift compensation algorithm establishes a linear relationship model between ambient temperature and measurement deviation. When the ambient temperature is higher than the standard reference temperature, the measurement results of the infrared thermal imager tend to be lower; when the ambient temperature is lower than the standard reference temperature, the measurement results tend to be higher. The compensation process calculates the difference between the ambient temperature and the 25-degree Celsius standard reference temperature, multiplies this difference by a temperature compensation coefficient of 0.02, and adds the compensation amount to the temperature value of each pixel in the noise-reduced infrared image. The temperature compensation coefficient, obtained through experimental calibration, reflects the sensitivity of the infrared thermal imager to changes in ambient temperature. By ensuring that the temperature value of each pixel in the preprocessed infrared image is compensated for ambient temperature, the interference of environmental factors on the accuracy of temperature measurement is eliminated.

[0047] In one specific embodiment, step S2 includes:

[0048] The preprocessed infrared image is input into the input layer of the convolutional neural network for data format conversion to obtain standardized network input data;

[0049] Based on standardized network input data, feature extraction is performed through five convolutional blocks. Each convolutional block includes convolution operations, batch normalization, and activation function processing to obtain deep feature vectors of electrical equipment.

[0050] The deep feature vector of electrical equipment is input into a fully connected classification layer for material category discrimination. The probability distribution of six material types—metal shell, plastic insulation material, ceramic component, rubber seal, glass insulator, and composite material shell—is output to obtain the material classification probability result.

[0051] The material type label for each pixel region is determined based on the maximum probability value in the material classification probability results. The material label is then mapped to the original image coordinates to obtain the material distribution map of the electrical equipment.

[0052] Specifically, when preprocessing infrared images into the input layer of the convolutional neural network, data format conversion first reshapes the 640×480 pixel infrared image data into a three-dimensional tensor structure. The first and second dimensions of the tensor correspond to the height and width of the image, respectively, and the third dimension corresponds to the number of channels in the image. Infrared images are usually single-channel grayscale images, but need to be copied into a three-channel format to adapt to the network architecture. Pixel value normalization converts the original pixel grayscale values ​​from the integer range of 0-255 to the floating-point range of 0-1. Normalization is done by dividing each pixel value by 255. Standardizing the network input data ensures the stability of gradient propagation and the convergence speed during network training. Data augmentation techniques are simultaneously applied to the input layer, including transformations such as random rotation, scaling, and brightness adjustment, to enhance the diversity of training data and the network's generalization ability.

[0053] Five convolutional blocks form the feature extraction backbone of the electrical equipment material recognition network. Each convolutional block processes the standardized network input data sequentially according to the feature abstraction hierarchy from shallow to deep. The first convolutional block uses 64 large 7×7 convolutional kernels with a stride of 2 to perform primary feature extraction on the input image. The large convolutional kernels can capture the overall contour and basic shape information of the electrical equipment. The convolution operation calculates the inner product pixel by pixel on the image by sliding the convolutional kernel to obtain the feature map. Batch normalization calculates the mean and variance of each feature map output by convolution. Then, the mean and the square root of the variance are subtracted from each pixel value in the feature map. Batch normalization eliminates the influence of differences in the distribution of data from different batches on network training. The ReLU activation function sets the negative values ​​after batch normalization to zero, while keeping the positive values ​​unchanged, enhancing the non-linear expressive ability of the network. The second to fifth convolutional layers progressively employ a greater number and smaller kernel size, using 128, 256, 512, and 512 3×3 kernels respectively. Smaller kernels extract finer local texture features and material surface characteristics. The output feature map size of each convolutional layer gradually decreases while the number of channels gradually increases, forming a hierarchical feature representation from coarse to fine. The global average pooling layer compresses the multiple feature maps output by the last convolutional layer into a fixed-length feature vector. The deep feature vector of the electrical equipment contains 512 dimensions, each representing a specific material feature pattern.

[0054] The fully connected classification layer receives deep feature vectors from electrical equipment and performs material category discrimination. This layer contains two fully connected neural network layers: the first maps the 512-dimensional feature vector to a 256-dimensional intermediate representation, and the second maps this intermediate representation to a 6-dimensional output vector. Each dimension of the output vector corresponds to a type of electrical equipment material. The weight matrix and bias vector are obtained through backpropagation training. Each element in the weight matrix represents the correlation strength between the input feature and the output category. The Softmax activation function transforms the 6-dimensional output vector into a probability distribution. The Softmax function calculates the exponential function for each output value, and then sums all the exponential values ​​as a normalization factor. Dividing the exponent of each output value by the normalization factor yields the probability of the corresponding category. The sum of the probability distributions for the six material types—metal shell, plastic insulation, ceramic components, rubber seals, glass insulators, and composite material shells—equals 1. The material classification probability results reflect the network's confidence in the material type of each pixel region.

[0055] The process of determining the maximum probability value in the material classification probability results adopts a pixel-by-pixel analysis method. The probability distribution output by the network corresponds to each pixel position in the preprocessed infrared image. The algorithm traverses all pixels in the image, extracts the 6-dimensional probability vector corresponding to each pixel position, compares the numerical values ​​in the probability vectors, and selects the material type with the highest probability value as the material type label for that pixel. A probability threshold filtering mechanism sets a minimum confidence requirement of 0.6. When the maximum probability value of a pixel position is lower than the threshold, the pixel is marked as an unidentified area to avoid low-confidence prediction results affecting the accuracy of subsequent processing. Material labels are mapped to the original image coordinate positions to establish pixel-level material distribution information. The mapping process considers the image size transformation and coordinate offset in the preprocessing stage to ensure accurate correspondence between material labels and actual electrical equipment locations. The electrical equipment material distribution map visualizes the spatial distribution of different material types using color coding: metal shell areas are displayed in blue, plastic insulation material areas in red, ceramic component areas in green, rubber sealing material areas in yellow, glass insulator areas in purple, and composite material shell areas in orange.

[0056] In one specific embodiment, the material type label of each pixel region is determined based on the maximum probability value in the material classification probability results, and the material label is mapped to the original image coordinate position to obtain the electrical equipment material distribution map, including:

[0057] The probability values ​​of the six material types in the material classification probability results are compared and sorted. The material type with the highest probability value is selected as the candidate material label for the corresponding pixel region, thus obtaining a set of pixel-level material labels.

[0058] Connectivity analysis is performed on pixel-level material tag sets to merge adjacent pixels with the same material type into a single device region, resulting in device region segmentation.

[0059] Based on the area threshold and shape characteristics of each equipment region in the equipment region segmentation results, an effectiveness screening process is performed to filter out noise regions with an area smaller than the preset minimum value, thereby obtaining effective electrical equipment regions.

[0060] The material label information in the effective electrical equipment area is spatially marked according to the original image coordinate system, and the correspondence between material type and image coordinate is established to obtain the material distribution map of electrical equipment.

[0061] Specifically, the comparison and sorting of probability values ​​for the six material types in the material classification probability results employs a pixel-by-pixel traversal strategy. The algorithm reads the six-dimensional probability vector corresponding to each pixel position output by the neural network. These six vectors represent the predicted probabilities for metal shells, plastic insulation materials, ceramic components, rubber seals, glass insulators, and composite material shells, respectively. The comparison and sorting are achieved by determining the numerical magnitude; the algorithm compares the six probability values ​​at the same pixel position and identifies the probability with the highest value and its corresponding material type index. The candidate material label selection process sets a minimum confidence threshold of 0.6. When the maximum probability value of a pixel is lower than this threshold, the pixel is marked as unidentified to avoid low-confidence predictions affecting the accuracy of subsequent processing. The pixel-level material label set records the material type information for each pixel position in the image. Each element in the set contains pixel coordinates and a corresponding material type identifier, with the identifier using integer encoding to represent different material types.

[0062] Connected component analysis is performed based on a set of pixel-level material labels to detect spatial adjacency. A connected component refers to a region in an image that consists of pixels with the same attributes and are spatially adjacent to each other. The algorithm uses an eight-connectivity approach to define pixel adjacency, meaning that each pixel is considered adjacent to pixels in all eight directions around it, including the four orthogonal directions (up, down, left, right) and the four diagonal directions. Connected component analysis starts scanning line by line from the top left corner of the image. When an unlabeled pixel is encountered, the algorithm initiates a depth-first search or breadth-first search strategy, expanding along the paths of adjacent pixels with the same material type, and grouping all connected pixels of the same material into a single device region. Region labeling uses unique integer identifiers to distinguish different connected components, and all pixels within each connected component share the same region identifier and material type label. The device region segmentation result transforms the original pixel-level data into a region-level data structure. Each region contains the boundary coordinates, total number of pixels, material type, and shape feature parameters.

[0063] In the equipment region segmentation results, area thresholds and shape feature validity screening processes are used to eliminate noise and false target regions generated during image segmentation. The area threshold is set based on the minimum size requirements of actual electrical equipment; regions smaller than a preset minimum value are considered noise regions or segmentation errors. The preset minimum area threshold is calculated based on image resolution and monitoring distance, with a typical value of one hundred pixels. Area calculation is achieved by counting the total number of pixels in each connected component; the total number of pixels multiplied by the actual physical area corresponding to a single pixel yields the actual area size of the region. Shape features include geometric parameters such as the region's aspect ratio, roundness, and convex hull area ratio. The aspect ratio is calculated as the ratio of the length to the width of the region's smallest bounding rectangle; roundness is calculated as the ratio of the square of the region's perimeter to its area multiplied by four times pi; and the convex hull area ratio is calculated as the ratio of the region's area to its convex hull area. The screening process filters out abnormal regions with overly long or complex shapes based on the shape feature parameters, retaining valid regions that conform to the typical shape characteristics of electrical equipment. The set of valid electrical equipment regions includes all regions that have passed both area and shape screening; each region has a clear material type identifier and accurate spatial boundary information.

[0064] Material label information within the effective electrical equipment area is spatially marked according to the original image coordinate system, establishing a precise mapping relationship between material type and image coordinates. Coordinate system transformation considers size changes and positional offsets during image preprocessing to ensure accurate positioning of material labels within the original image coordinate system. Spatial location marking is achieved through region centroid calculation and boundary extraction. The region centroid is calculated as the arithmetic mean of all pixel coordinates within the region, representing the center position of the material region. Boundary extraction identifies all pixel coordinates on the region's contour, forming a spatial description of the material region's extent. The correspondence between material type and image coordinates is stored in a data table format. Each row in the table records information for a material region, including region identifier, material type, centroid coordinates, boundary coordinate set, and region area. The electrical equipment material distribution map uses a color-coded visualization method to present the spatial distribution of different material types: metal shell areas are displayed in blue, plastic insulation material areas in red, ceramic component areas in green, and other material types correspond to different color codes, forming an intuitive visualization of material distribution.

[0065] In one specific embodiment, step S3 includes:

[0066] Based on the material type label in the electrical equipment material distribution map, query the preset material emissivity parameter database to obtain the reference emissivity value and temperature coefficient parameter corresponding to each material type;

[0067] An emissivity correction function model is constructed based on the baseline emissivity value and temperature coefficient parameters. The initial temperature reading and material type label of each pixel are substituted into the correction function for calculation to obtain the pixel-level dynamic emissivity value.

[0068] The original infrared radiance value is calculated and processed by emissivity compensation based on the pixel-level dynamic emissivity value. The true temperature value is then solved by inversely using the Stefan-Boltzmann law to obtain the corrected pixel temperature matrix.

[0069] The corrected pixel temperature matrix is ​​integrated and processed according to the boundaries of the electrical equipment area. The average temperature, maximum temperature and temperature distribution characteristic parameters of each equipment area are calculated to obtain emissivity-corrected temperature data.

[0070] Specifically, the dynamic emissivity correction algorithm in the infrared thermal imaging-based electrical fire hazard detection method solves the core technical problem of temperature measurement errors caused by differences in the emissivity of different electrical equipment materials in traditional infrared thermal imaging technology through precise material parameter lookup and temperature compensation calculation. A preset material emissivity parameter database stores emissivity characteristic data for six electrical equipment materials. The database adopts a relational table structure, with one record corresponding to each material. Each record contains four fields: material type identifier, reference emissivity value, primary temperature coefficient, and secondary temperature coefficient. The reference emissivity value for metal casing materials is set to 0.2, the primary temperature coefficient to 0.0008 degrees Celsius, and the secondary temperature coefficient to 1.2 × 10⁻⁶ degrees Celsius squared. These parameters were obtained through laboratory calibration and literature review. The reference emissivity value for plastic insulation materials is 0.92, the primary temperature coefficient to -0.0012 degrees Celsius, and the secondary temperature coefficient to 2.1 × 10⁻⁶ degrees Celsius squared. Ceramic components, rubber seals, glass insulators, and composite material shells each correspond to different parameter combinations. The query process uses the material type label in the electrical equipment material distribution map as the index key to extract the reference emissivity value and temperature coefficient parameter of the corresponding material from the database.

[0071] The emissivity correction function model is constructed using a quadratic polynomial to describe the variation of emissivity with temperature. The correction function takes material type, reference emissivity value, temperature coefficient parameter, and pixel temperature reading as input variables. The mathematical expression of the function model is: emissivity equals the reference emissivity value plus the first temperature coefficient multiplied by the temperature difference plus the second temperature coefficient multiplied by the square of the temperature difference, where the temperature difference is the current pixel temperature minus the reference temperature of 25 degrees Celsius. The pixel-level dynamic emissivity calculation process iterates through each pixel position in the corrected pixel temperature matrix, reads the initial temperature reading and material type label for that position, queries the corresponding reference emissivity value and temperature coefficient parameter based on the material type label, and substitutes the initial temperature reading into the correction function to calculate the dynamic emissivity value for that pixel position. The calculation process considers iterative optimization of temperature measurement. The initial calculation uses the initial temperature reading to obtain a preliminary dynamic emissivity value, then recalculates the temperature using this emissivity value, updating the emissivity value again. This iterative process continues until the emissivity value converges or the preset number of iterations (3) is reached.

[0072] The inverse calculation of the Stefan-Boltzmann law applies pixel-level dynamic emissivity values ​​to the temperature conversion of the original infrared radiance values. The Stefan-Boltzmann law describes the relationship between the surface temperature of an object and its infrared radiation intensity. The emissivity compensation calculation first obtains the original infrared radiance values ​​acquired by the infrared thermal imager, reflecting the infrared radiation power density emitted by the object's surface. Then, the original radiance value is divided by the dynamic emissivity value to obtain the equivalent radiance value under ideal blackbody conditions. The true temperature value is calculated through the inverse calculation of the Stefan-Boltzmann law. The calculation process divides the equivalent radiance value by the Stefan-Boltzmann constant, then by the atmospheric transmittance coefficient, and finally takes the fourth root to obtain the absolute temperature value in Kelvin. Converting to Celsius requires subtracting 273.15. The corrected pixel temperature matrix contains the true temperature value of each pixel in the image after emissivity compensation. The matrix's row and column dimensions are consistent with the original infrared image, and the data type uses floating-point format to maintain the accuracy of the temperature calculation.

[0073] Temperature data integration processing involves grouping and statistically calculating the corrected pixel temperature matrix according to the boundaries of electrical equipment regions. The region boundary information comes from the valid equipment regions identified in the electrical equipment material distribution map. Average temperature calculation involves summing the temperature values ​​of all pixels within each equipment region and dividing by the total number of pixels to obtain the average temperature of that region. The average temperature reflects the overall thermal state of the equipment region. Maximum temperature calculation searches for the maximum value of all pixel temperature values ​​within each equipment region, identifying hotspot locations and abnormal heating levels within the region. Temperature distribution characteristic parameters include statistical quantities such as temperature standard deviation, temperature coefficient of variation, and temperature gradient. The temperature standard deviation is calculated as the square root of the sum of the squares of the differences between the temperature of each pixel within the region and the average temperature, reflecting the dispersion of the temperature distribution. The temperature coefficient of variation is the ratio of the temperature standard deviation to the average temperature. The temperature gradient calculates the average rate of temperature change between adjacent pixels. Emissivity-corrected temperature data is stored in tabular form, containing fields such as equipment region identifier, material type, average temperature, maximum temperature, minimum temperature, temperature standard deviation, temperature coefficient of variation, and temperature gradient.

[0074] In one specific embodiment, step S4 includes:

[0075] Based on the boundary information of each electrical equipment region in the emissivity-corrected temperature data, the equipment outline area, aspect ratio and shape complexity are extracted as geometric feature parameters to obtain the set of geometric features of electrical equipment.

[0076] Based on the emissivity-corrected temperature data, the average temperature, temperature standard deviation, and temperature gradient of each electrical equipment area are calculated as temperature feature parameters. These parameters are then combined with the set of geometric features of the electrical equipment to construct a comprehensive feature vector, thereby obtaining multidimensional feature data of the electrical equipment.

[0077] The distance between feature vectors of the multidimensional feature data of electrical equipment is calculated, and the similarity between equipment is quantified by the intersection and union ratio method to obtain the equipment similarity matrix.

[0078] Clustering iterative calculation is performed based on the equipment similarity matrix. Electrical equipment with similar features is grouped into the same category group according to the similarity threshold, and the clustering grouping results of electrical equipment of the same category are obtained.

[0079] Specifically, the clustering and identification process of similar equipment in the infrared thermal imaging-based electrical fire hazard perception method solves the technical challenge of traditional infrared thermal imaging technology's inability to intelligently identify individual abnormal devices among similar electrical equipment through the fusion analysis of geometric and temperature features. Geometric feature parameter extraction of electrical equipment is based on contour analysis and shape calculation using boundary information of each equipment region from emissivity-corrected temperature data. The boundary information originates from the valid electrical equipment regions identified in the equipment region segmentation results. Equipment contour area calculation is achieved by counting the total number of pixels within each equipment region. The total number of pixels multiplied by the actual physical area corresponding to a single pixel yields the actual surface area of ​​the equipment. The physical area conversion factor is determined based on the field of view angle and monitoring distance of the infrared thermal imager. Aspect ratio calculation requires first determining the minimum bounding rectangle of the equipment region. The minimum bounding rectangle is the rectangle that completely contains the equipment contour and has the smallest area. The aspect ratio is equal to the ratio of the longer side to the shorter side of the rectangle, reflecting the slenderness and overall shape characteristics of the equipment. Shape complexity is quantified using the concept of circumference ratio, calculated as the ratio of the square of the equipment contour perimeter to the contour area multiplied by 4 times pi. The shape complexity of circular equipment is close to 1, and the more irregular the shape, the greater the complexity value. The set of geometric features of electrical equipment combines the outline area, aspect ratio and shape complexity of each device into a 3D geometric feature vector, where each dimension of the vector represents a geometric attribute.

[0080] Temperature feature parameter calculations are based on the temperature distribution information of each device area in the emissivity-corrected temperature data. The average temperature calculation calculates the arithmetic mean of the corrected temperature values ​​of all pixels within the device area, reflecting the overall thermal state level of the device. The temperature standard deviation is calculated by summing the squares of the differences between the temperature of each pixel and the average temperature of the area, dividing by the total number of pixels minus 1, and then taking the square root to quantify the uniformity and dispersion of the internal temperature distribution of the device. The temperature gradient calculation uses the spatial difference method to calculate the average rate of temperature change between adjacent pixels within the device area. A large temperature gradient indicates the presence of significant temperature variation areas within the device, while a small temperature gradient indicates a relatively uniform temperature distribution. The comprehensive feature vector construction merges the 3D geometric feature vector and the 3D temperature feature vector into a 6-dimensional comprehensive feature vector. The first three dimensions of the vector correspond to geometric features, and the last three dimensions correspond to temperature features. The multi-dimensional feature data of electrical equipment undergoes data standardization to eliminate differences in the numerical range between different feature dimensions. The standardization process subtracts the mean of each feature dimension from its value and divides by the standard deviation, ensuring that all feature dimensions are within a similar numerical range for subsequent calculations.

[0081] The distance calculation between feature vectors uses the Tanimoto similarity measure instead of the traditional Euclidean distance calculation. The Tanimoto similarity measure is specifically designed to handle the similarity quantification of multi-dimensional feature vectors. The intersection and union ratio calculation process first calculates the minimum value of the corresponding dimension values ​​of the two feature vectors as the intersection contribution, then calculates the maximum value of the corresponding dimension values ​​as the union contribution. The sum of the intersection contributions of all dimensions yields the total intersection value, and the sum of the union contributions of all dimensions yields the total union value. The similarity is equal to the total intersection value divided by the total union value. The device similarity matrix is ​​a symmetric matrix, with rows and columns corresponding to different electrical devices. Each element in the matrix represents the similarity value between two corresponding devices, ranging from 0 to 1. The closer the value is to 1, the more similar the devices are; the closer the value is to 0, the greater the difference between the devices. A value of 1 on the diagonal of the matrix indicates that the device is completely similar to itself.

[0082] The iterative clustering process groups devices based on a similarity matrix, employing a hierarchical clustering method with a similarity threshold. The similarity threshold is set to 0.8; devices with a similarity exceeding this threshold are grouped into the same cluster. The clustering process searches the similarity matrix for all similarity values ​​greater than the threshold, establishing connections between devices. Then, connectivity analysis merges devices with transitive connections into the same cluster. The iterative calculation process includes four stages: initialization, similarity comparison, connection establishment, and cluster update. In the initialization stage, each device is treated as an independent cluster. The similarity comparison stage scans all elements in the similarity matrix. The connection establishment stage establishes connections between devices based on the threshold condition. The cluster update stage merges groups of devices with connections. The clustering results for similar electrical equipment are stored in a list format. Each element in the list represents a device group, containing information such as the identifier, mean geometric features, mean temperature features, and number of devices within that group.

[0083] In one specific embodiment, step S5 includes:

[0084] Based on the equipment temperature data of each group in the clustering results of similar electrical equipment, the average temperature value and temperature standard deviation within the group are calculated as statistical benchmark parameters to obtain the statistical characteristic data of group temperature.

[0085] Based on the statistical characteristics of temperature in each group, the standardized deviation of the individual equipment temperature in each group is calculated. The difference between the individual equipment temperature and the average temperature in the group is divided by the standard deviation of the temperature in the group to obtain the standardized deviation of the equipment temperature.

[0086] The standardized deviation of the equipment temperature is compared and analyzed with a preset anomaly judgment threshold. When the deviation value is greater than the threshold, the corresponding equipment is marked as having an abnormal temperature state, and the abnormal state identification result of the equipment is obtained.

[0087] Based on the abnormal equipment information in the equipment abnormality status identification results, combined with the equipment location coordinates and temperature deviation degree, fire hazard risk level assessment data is generated to obtain electrical fire hazard perception results.

[0088] Specifically, the statistical characteristic data calculation for group temperature is based on the statistical analysis of equipment temperature data within each group of similar electrical equipment clusters. This statistical analysis includes two core parameters: the intra-group average temperature and the temperature standard deviation. The intra-group average temperature is calculated by taking the arithmetic mean of the average temperature data of all equipment within the same cluster. This is done by summing the average temperature values ​​of each device within the group and then dividing by the total number of devices in that group, yielding a benchmark value representing the normal operating temperature level of that type of equipment. The intra-group temperature standard deviation is calculated by summing the squares of the differences between the average temperature of each device and the intra-group average temperature, dividing by the total number of devices minus 1, and then taking the square root. This quantifies the dispersion and range of variation in temperature distribution among similar equipment. These statistical benchmark parameters establish a mathematical model of the temperature distribution of similar equipment, providing an objective comparison standard for subsequent anomaly detection, replacing the fixed temperature thresholds set based on experience in traditional methods.

[0089] Standardized deviation calculation converts the temperature data of individual devices into dimensionless standardized values, eliminating the impact of temperature benchmark differences between different types of equipment on anomaly detection. The standardized deviation calculation process first obtains the average temperature data of individual devices within each group, then calculates the difference between the device's temperature and the group's average temperature. This difference reflects the degree of temperature deviation of the individual device relative to similar devices. Standardization divides the temperature difference by the group's temperature standard deviation to obtain a standard normally distributed deviation value. A positive standardized deviation value indicates that the device temperature is higher than the average level of similar devices, while a negative value indicates that the device temperature is lower than the average level of similar devices. The absolute magnitude of the value reflects the severity of the deviation. Standardized deviation calculation solves the problem of difficult alarm settings caused by large differences in the normal operating temperatures of different devices, as described in the manual, by automatically adapting to the temperature characteristics of various types of equipment through relative comparison.

[0090] The anomaly detection threshold comparative analysis process employs the outlier detection principle from statistics, with a preset threshold of 2.5 standard deviations. This threshold is determined based on the 3-sigma criterion of normal distribution; data points exceeding 2.5 standard deviations in a normal distribution are considered statistically outliers. The comparative analysis process examines the standardized deviation value of each device individually. When the deviation value of a device exceeds the anomaly detection threshold, the device is marked as having an abnormal temperature state. The anomaly status marker includes the device identifier, anomaly type, deviation value, and detection time. The device anomaly status identification results record information for all devices judged to be abnormal, forming an abnormal device list and status database. The anomaly detection process is entirely based on the relative temperature relationships of similar devices, avoiding the technical shortcomings of traditional methods where fixed thresholds cannot adapt to complex field environments.

[0091] The fire hazard risk level assessment data is generated based on a comprehensive risk analysis of abnormal equipment information from the equipment anomaly status identification results. The risk assessment considers multiple dimensions, including equipment location coordinates, temperature deviation, equipment type, and environmental factors. Equipment location coordinate information is derived from spatial positioning data in infrared images. This location information is used to determine whether abnormal equipment is located in densely populated areas, near important equipment, or in areas where flammable materials are stored; different locations have different risk weight coefficients. Temperature deviation is quantified by standardized deviation values; a larger deviation value indicates a more severe temperature anomaly and a higher fire hazard risk level. A four-level system is used for risk level classification: low risk, medium risk, high risk, and extremely high risk. The risk level is determined by a weighted comprehensive score, and the scoring formula comprehensively considers factors such as temperature deviation weight, location risk weight, equipment type weight, and duration weight. The electrical fire hazard perception results are output in report form, including a list of abnormal equipment, risk level distribution, handling recommendations, and monitoring trends. The handling recommendations provide corresponding emergency measures and maintenance requirements based on the risk level.

[0092] The above describes the electrical fire hazard detection method based on infrared thermal imaging in the embodiments of this application. The following describes the electrical fire hazard detection system based on infrared thermal imaging in the embodiments of this application. Please refer to [link / reference]. Figure 2 One embodiment of the electrical fire hazard detection system based on infrared thermal imaging in this application includes:

[0093] The filtering module is used to collect infrared image data of the monitoring area of ​​electrical equipment by the infrared thermal imager, and to eliminate image noise through adaptive median filtering to obtain a pre-processed infrared image.

[0094] The input module is used to input the preprocessed infrared image into the electrical equipment material recognition network, identify the material type labels of the electrical equipment in the image, and obtain the material distribution map of the electrical equipment.

[0095] The calculation module is used to calculate the corrected emissivity parameters of each pixel based on the material type label in the material distribution map of the electrical equipment, and to obtain emissivity correction temperature data.

[0096] The matching module is used to extract the geometric and temperature features of electrical equipment based on the emissivity-corrected temperature data, perform equipment feature matching using similarity measures, and obtain clustering results of similar electrical equipment.

[0097] The judgment module is used to calculate the temperature standardization deviation value of each device in the clustering results of the same type of electrical equipment. When the deviation value exceeds the preset abnormal threshold, it is judged as a fire hazard, and the electrical fire hazard perception result is obtained.

[0098] above Figure 2 The electrical fire hazard sensing system based on infrared thermal imaging in this embodiment of the invention will be described in detail from the perspective of modular functional entities. The electrical fire hazard sensing device based on infrared thermal imaging in this embodiment of the invention will be described in detail from the perspective of hardware processing.

[0099] Reference Figure 3 This invention also provides an electrical fire hazard detection device based on infrared thermal imaging. This device can be a server, and its internal structure can be as follows: Figure 3 As shown, the infrared thermal imaging-based electrical fire hazard detection device includes a processor, memory, display screen, input device, network interface, and database connected via a system bus. The processor, designed as a computer, provides computing and control capabilities. The memory of the infrared thermal imaging-based electrical fire hazard detection device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database of the infrared thermal imaging-based electrical fire hazard detection device stores the data corresponding to this embodiment. The network interface of the infrared thermal imaging-based electrical fire hazard detection device is used for communication with external terminals via a network connection. When the computer program is executed by the processor, it implements the above-described method.

[0100] Those skilled in the art will understand that Figure 3 The structure shown is merely a block diagram of a portion of the structure related to the present invention and does not constitute a limitation on the infrared thermal imaging-based electrical fire hazard sensing device to which the present invention is applied.

[0101] The present invention also provides a computer-readable storage medium, which can be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium, wherein the computer-readable storage medium stores instructions that, when the instructions are executed on a computer, cause the computer to perform the steps of the method for sensing electrical fire hazards based on infrared thermal imaging.

[0102] Those skilled in the art will clearly understand that, for the sake of convenience and brevity, the specific working processes of the systems and units described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0103] If the integrated unit is implemented as 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, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause an infrared thermal imaging-based electrical fire hazard detection device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0104] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting electrical fire hazards based on infrared thermal imaging, characterized in that, The method includes: Step S1: The infrared thermal imager acquires infrared image data of the monitoring area of ​​the electrical equipment, and the image noise is eliminated by adaptive median filtering to obtain a preprocessed infrared image; Step S2: Input the preprocessed infrared image into the electrical equipment material recognition network, identify the material type labels of the electrical equipment in the image, and obtain the electrical equipment material distribution map; Step S3: Based on the material type labels in the electrical equipment material distribution map, apply a dynamic emissivity correction algorithm to calculate the corrected emissivity parameters of each pixel to obtain emissivity correction temperature data. This includes: querying a preset material emissivity parameter database based on the material type labels in the electrical equipment material distribution map to obtain the reference emissivity value and temperature coefficient parameter corresponding to each material type; constructing an emissivity correction function model based on the reference emissivity value and temperature coefficient parameter, substituting the initial temperature reading and material type label of each pixel into the correction function for calculation to obtain pixel-level dynamic emissivity values; performing emissivity compensation calculation on the original infrared radiation brightness value based on the pixel-level dynamic emissivity values, and inversely solving for the true temperature value using the Stefan-Boltzmann law to obtain the corrected pixel temperature matrix; integrating the temperature data of the corrected pixel temperature matrix according to the electrical equipment area boundaries, calculating the average temperature, maximum temperature, and temperature distribution characteristic parameters of each equipment area to obtain the emissivity correction temperature data. Step S4: Extract the geometric and temperature features of the electrical equipment based on the emissivity-corrected temperature data, and use a similarity metric to perform equipment feature matching to obtain clustering results for similar electrical equipment; Step S5: Calculate the temperature standardization deviation of each device in the clustering results of the same type of electrical equipment. When the deviation value exceeds the preset abnormal threshold, it is determined as a fire hazard, and the electrical fire hazard perception result is obtained.

2. The method for detecting electrical fire hazards based on infrared thermal imaging according to claim 1, characterized in that, Step S1 includes: Infrared thermal imagers continuously scan and collect data on the monitoring area of ​​electrical equipment to obtain raw infrared image sequences containing temperature information. Calculate the pixel neighborhood variance value for each frame of the original infrared image sequence, determine the filter window size parameter based on the variance value distribution, and obtain the adaptive filter parameter set. The adaptive filtering parameter set is applied to the original infrared image sequence to perform noise filtering, eliminating random noise and isolated noise points in the image, and obtaining a denoised infrared image. Based on real-time ambient temperature data collected by an ambient temperature sensor, the noise-reduced infrared image is subjected to temperature drift compensation processing to obtain the preprocessed infrared image.

3. The method for detecting electrical fire hazards based on infrared thermal imaging according to claim 1, characterized in that, Step S2 includes: The preprocessed infrared image is input into the input layer of the convolutional neural network for data format conversion to obtain standardized network input data; Based on the standardized network input data, feature extraction processing is performed through five convolutional blocks. Each convolutional block includes convolution operation, batch normalization processing and activation function processing to obtain the deep feature vector of electrical equipment. The deep feature vector of the electrical equipment is input into a fully connected classification layer for material category discrimination. The probability distribution of six material types—metal shell, plastic insulation material, ceramic component, rubber seal, glass insulator, and composite material shell—is output to obtain the material classification probability result. The material type label for each pixel region is determined based on the maximum probability value in the material classification probability results. The material label is then mapped to the original image coordinate position to obtain the material distribution map of the electrical equipment.

4. The method for detecting electrical fire hazards based on infrared thermal imaging according to claim 3, characterized in that, The step of determining the material type label for each pixel region based on the maximum probability value in the material classification probability results, mapping the material label to the original image coordinate position, and obtaining the material distribution map of the electrical equipment includes: The probability values ​​of the six material types in the material classification probability results are compared and sorted. The material type with the highest probability value is selected as the candidate material label for the corresponding pixel region, thus obtaining a set of pixel-level material labels. Based on the pixel-level material tag set, connected component analysis is performed to merge adjacent pixels with the same material type into a single device region, thus obtaining the device region segmentation result. Based on the area threshold and shape characteristics of each equipment region in the equipment region segmentation result, an effectiveness screening process is performed to filter out noise regions with an area smaller than a preset minimum value, thereby obtaining effective electrical equipment regions. The material label information in the effective electrical equipment area is spatially marked according to the original image coordinate system to establish the correspondence between material type and image coordinates, thereby obtaining the material distribution map of the electrical equipment.

5. The method for detecting electrical fire hazards based on infrared thermal imaging according to claim 1, characterized in that, Step S4 includes: Based on the boundary information of each electrical equipment region in the emissivity-corrected temperature data, the equipment outline area, aspect ratio and shape complexity are extracted as geometric feature parameters to obtain a set of geometric features of electrical equipment. Based on the emissivity-corrected temperature data, the average temperature, temperature standard deviation, and temperature gradient of each electrical equipment area are calculated as temperature feature parameters. These parameters are then combined with the set of geometric features of the electrical equipment to construct a comprehensive feature vector, thereby obtaining multidimensional feature data of the electrical equipment. The distance between feature vectors of the multidimensional feature data of the electrical equipment is calculated, and the similarity between the equipment is quantified by the intersection and union ratio method to obtain the equipment similarity matrix. Clustering iterative calculation is performed based on the device similarity matrix. Electrical devices with similar features are grouped into the same category group according to the similarity threshold, and the clustering grouping results of the same type of electrical devices are obtained.

6. The method for detecting electrical fire hazards based on infrared thermal imaging according to claim 1, characterized in that, Step S5 includes: Based on the equipment temperature data of each group in the clustering results of the same type of electrical equipment, the average temperature value and temperature standard deviation within the group are calculated as statistical benchmark parameters to obtain the statistical characteristic data of group temperature. Based on the temperature statistical characteristic data of the group, the standardized deviation of the individual equipment temperature in each group is calculated. The difference between the individual equipment temperature and the average temperature in the group is divided by the standard deviation of the temperature in the group to obtain the standardized deviation value of the equipment temperature. The standardized deviation of the equipment temperature is compared and analyzed with a preset anomaly judgment threshold. When the deviation value is greater than the threshold, the corresponding equipment is marked as having an abnormal temperature state, and the abnormal state identification result of the equipment is obtained. Based on the abnormal equipment information in the abnormal equipment status identification results, fire hazard risk level assessment data is generated by combining the equipment location coordinates and temperature deviation, and the electrical fire hazard perception results are obtained.

7. An electrical fire hazard detection system based on infrared thermal imaging, characterized in that, For implementing the method for detecting electrical fire hazards based on infrared thermal imaging as described in any one of claims 1-6, the system for detecting electrical fire hazards based on infrared thermal imaging comprises: The filtering module is used by the infrared thermal imager to acquire infrared image data of the monitoring area of ​​electrical equipment, and to eliminate image noise through adaptive median filtering to obtain a pre-processed infrared image. The input module is used to input the preprocessed infrared image into the electrical equipment material recognition network, identify the material type labels of the electrical equipment in the image, and obtain the material distribution map of the electrical equipment. The calculation module is used to calculate the corrected emissivity parameters of each pixel based on the material type labels in the electrical equipment material distribution map, and to obtain emissivity-corrected temperature data. This includes: querying a preset material emissivity parameter database based on the material type labels in the electrical equipment material distribution map to obtain the reference emissivity value and temperature coefficient parameter corresponding to each material type; constructing an emissivity correction function model based on the reference emissivity value and temperature coefficient parameter; substituting the initial temperature reading and material type label of each pixel into the correction function for calculation to obtain pixel-level dynamic emissivity values; performing emissivity compensation calculation on the original infrared radiation brightness value based on the pixel-level dynamic emissivity values; inversely solving for the true temperature value using the Stefan-Boltzmann law to obtain the corrected pixel temperature matrix; and integrating the corrected pixel temperature matrix according to the electrical equipment area boundaries to calculate the average temperature, maximum temperature, and temperature distribution characteristic parameters of each equipment area, thereby obtaining the emissivity-corrected temperature data. The matching module is used to extract the geometric and temperature features of electrical equipment based on the emissivity-corrected temperature data, perform equipment feature matching using similarity measures, and obtain clustering results of similar electrical equipment. The judgment module is used to calculate the temperature standardization deviation value of each device in the clustering results of the same type of electrical equipment. When the deviation value exceeds the preset abnormal threshold, it is judged as a fire hazard, and the electrical fire hazard perception result is obtained.

8. An electrical fire hazard detection device based on infrared thermal imaging, characterized in that, The device includes a memory and a processor, wherein the memory stores a computer program that can run on the processor, and the processor executes the computer program to implement the electrical fire hazard perception method based on infrared thermal imaging as described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is run by the processor, it causes the processor to execute the electrical fire hazard perception method based on infrared thermal imaging as described in any one of claims 1 to 6.

Images