Temperature detection method and device, electronic equipment and storage medium

Abstract

The invention provides a temperature detection method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a region distribution image and a heat distribution matrix of a target material, carrying out the region segmentation of the region distribution image through a preset material distribution region segmentation model, and obtaining a target material region, the material distribution region segmentation model is obtained by training an image sample with a material distribution region label, mapping a region distribution image to a heat distribution matrix to obtain a homography transformation matrix, and constructing a deep learning registration model based on the homography transformation matrix to obtain a material distribution region segmentation model; performing registration alignment on the area distribution image and the heat distribution matrix according to a deep learning registration model to obtain a registration result, and performing statistics according to the registration result of the target material area to obtain temperature statistical data so as to complete temperature detection of the target material; through the method, a more practical and more accurate temperature detection method is provided.

Description

Technical Field

[0001] This application relates to the field of industrial temperature detection technology, and in particular to a temperature detection method, device, electronic device and storage medium. Background Technology

[0002] In the advancement of industrial automation and intelligent manufacturing, real-time monitoring of material status during transportation, processing, and storage has become crucial for ensuring the safe, stable, and efficient operation of production lines. Temperature, as an important parameter characterizing the physical and chemical state of materials, directly reflects whether abnormal conditions such as overheating, localized blockage, or pre-combustion signs are occurring. Therefore, continuous and accurate monitoring of material temperature is of significant importance for improving product quality, ensuring production safety, and optimizing energy efficiency. Currently, various temperature detection methods have been developed in this field, mainly including contact temperature sensor detection, non-contact infrared thermal imaging detection, and registration detection based on visible light and infrared images.

[0003] However, the relevant technologies still have several obvious limitations in practical applications: First, detection schemes based on a single sensing mode (such as relying solely on infrared thermal imagers or a single type of sensor) have narrow information sources, making it difficult to comprehensively capture the diverse characteristics of the material's state, resulting in limited data analysis dimensions and low recognition accuracy. Second, when attempting to fuse multi-source images such as visible light and infrared, inaccurate positioning often occurs due to insufficient image registration accuracy and deviations in regional correspondence, making it impossible to accurately correlate temperature information with the actual area of ​​the material. In addition, most existing systems still rely on preset rules and fixed threshold judgment mechanisms, making it difficult to adapt to dynamic temperature changes under complex working conditions, unable to achieve early prediction of abnormal trends, and exhibiting problems of response lag and untimely warnings. Clearly, a new temperature detection method is urgently needed to solve at least one of the above problems.

[0004] It should be noted that the above content only provides background information related to this application and does not necessarily constitute prior art. Summary of the Invention

[0005] This application provides a temperature detection method, device, electronic device, and storage medium to solve the technical problems of low accuracy and low real-time performance in temperature detection of material areas in high-temperature manufacturing scenarios.

[0006] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0007] This application provides a temperature detection method, comprising: acquiring a regional distribution image and a heat distribution matrix of a target material; segmenting the regional distribution image using a preset material distribution region segmentation model to obtain a target material region, wherein the material distribution region segmentation model is trained based on image samples labeled with material distribution regions; mapping the regional distribution image to the heat distribution matrix to obtain a homography transformation matrix; constructing a deep learning registration model based on the homography transformation matrix; registering and aligning the regional distribution image and the heat distribution matrix according to the deep learning registration model to obtain a registration result; and statistically analyzing the registration result of the target material region to obtain temperature statistics data, thereby completing the temperature detection of the target material.

[0008] In one embodiment of this application, based on the aforementioned scheme, the material distribution area segmentation model is trained using image samples labeled with material distribution areas, including: acquiring material distribution area labels and sample image data; labeling the sample image data according to the material distribution area labels to generate image samples with material distribution area labels; and training a preset material distribution area segmentation model based on the image samples with material distribution area labels to obtain the material distribution area segmentation model, wherein the preset material distribution area segmentation model is obtained by adding a scaling factor based on a self-attention mechanism.

[0009] In one embodiment of this application, based on the aforementioned scheme, the heat distribution matrix is ​​determined by the following method: obtaining the original temperature of the object and the wavelength of the electromagnetic wave; calculating the spectral radiant exitance of each pixel based on Planck's radiation law, the original temperature of the object, and the wavelength of the electromagnetic wave; calculating the radiation intensity received by the thermal imager detector at each pixel based on the spectral radiant exitance, the comprehensive response coefficient, and the operating band of the infrared thermal imager; and calculating the surface temperature of the object at each pixel based on the radiation intensity, the calibration constant, the emissivity of the object surface, and the environmental reflectivity, thereby obtaining the heat distribution matrix.

[0010] In one embodiment of this application, based on the aforementioned scheme, the regional distribution image is mapped to a heat distribution matrix to obtain a homography transformation matrix. A deep learning registration model is constructed based on the homography transformation matrix. The regional distribution image and the heat distribution matrix are registered and aligned according to the deep learning registration model to obtain a registration result. This includes: extracting multiple feature points and feature descriptors corresponding to each feature point from the regional distribution image and the heat distribution matrix, respectively; calculating feature matching pairs between the regional distribution image and the heat distribution matrix based on the feature descriptors, and filtering the feature matching pairs using the nearest neighbor distance ratio and a bidirectional matching strategy to obtain a preliminary matching point set; and using a random sampling consensus algorithm combined with least binary... The process involves multiplication, determining a homography transformation matrix based on the initial matching point set, which maps the coordinate system of the regional distribution image to the heat distribution matrix; using the homography transformation matrix as initialization parameters, constructing a deep learning registration model, and optimizing the deep learning registration model by minimizing a loss function to obtain an optimized deep learning registration model; registering and aligning the regional distribution image and the heat distribution matrix based on the optimized deep learning registration model to obtain a registration result; calculating a registration quality evaluation index for the registration result, and if the registration quality evaluation index is less than or equal to a preset threshold, re-extracting feature points until the registration quality evaluation index is greater than the preset threshold to update the registration result.

[0011] In one embodiment of this application, based on the aforementioned scheme, temperature statistics are obtained by performing statistical analysis on the registration results of the target material region, including: determining the maximum temperature of the target material region based on the registration results of the target material region; calculating the average temperature, standard deviation of temperature, and temperature gradient of the target material region based on the registration results of the target material region; and calculating the hot spot ratio of the target material region based on the average temperature and the standard deviation of temperature, wherein the temperature statistics include the maximum temperature, average temperature, standard deviation of temperature, temperature gradient, and hot spot ratio.

[0012] In one embodiment of this application, based on the aforementioned scheme, after obtaining temperature statistics data by statistically analyzing the registration results of the target material region, the method further includes: determining a temperature feature vector based on the temperature statistics data; obtaining an initial time series based on the temperature feature vectors of multiple target times; normalizing the initial time series to obtain a target time series; inputting the target time series into a preset anomaly probability calculation model to obtain an anomaly probability value; determining a target warning level based on the comparison result of the anomaly probability value and a preset probability threshold; and issuing a warning based on the target warning level, wherein the preset probability threshold is determined based on the actual warning frequency and the target warning frequency within a preset time period.

[0013] In one embodiment of this application, based on the aforementioned scheme, after issuing an early warning according to the target early warning level, the method further includes: enhancing the features of the target time series to obtain an enhanced feature sequence; inputting the enhanced feature sequence into a preset sequence prediction model to obtain predicted values ​​of temperature feature vectors for multiple future times; and calculating the temperature prediction mean and temperature prediction variance for each future time based on the predicted values ​​of the temperature feature vectors for multiple future times, wherein the temperature prediction mean is used as the prediction result for the future time, and the temperature prediction variance is used as an uncertainty confidence index characterizing the prediction result.

[0014] This application also provides a temperature detection device, comprising: an acquisition module for acquiring a regional distribution image and a heat distribution matrix of a target material; a region segmentation module for segmenting the regional distribution image using a preset material distribution region segmentation model to obtain a target material region, wherein the material distribution region segmentation model is trained based on image samples labeled with material distribution regions; a registration module for mapping the regional distribution image to the heat distribution matrix to obtain a homography transformation matrix, constructing a deep learning registration model based on the homography transformation matrix, and registering and aligning the regional distribution image and the heat distribution matrix according to the deep learning registration model to obtain a registration result; and a statistics module for performing statistics based on the registration result of the target material region to obtain temperature statistics data, thereby completing the temperature detection of the target material.

[0015] This application also provides an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the temperature detection method as described in any of the above embodiments.

[0016] This application also provides a computer-readable storage medium storing a computer program, which is executed by a processor to implement the temperature detection method as described in any of the above embodiments.

[0017] The beneficial effects of this application are as follows: By acquiring the regional distribution image and heat distribution matrix of the target material, the regional distribution image is segmented using a preset material distribution region segmentation model to obtain the target material region. The material distribution region segmentation model is trained based on image samples labeled with material distribution regions. The regional distribution image is mapped to the heat distribution matrix to obtain a homography transformation matrix. A deep learning registration model is constructed based on the homography transformation matrix. The regional distribution image and heat distribution matrix are registered and aligned using the deep learning registration model to obtain the registration result. Statistical analysis is performed on the registration result of the target material region to obtain temperature statistics, thereby completing the temperature detection of the target material. By fusing multi-source image information, achieving precise positioning and temperature sensing of the material region, and introducing a deep learning model for intelligent analysis, the accuracy, real-time performance, and robustness of temperature detection can be improved, meeting the higher requirements for safety and efficiency in modern high-temperature industrial manufacturing.

[0018] In addition, this application predicts temperature trends and detects anomalies through time series analysis, and achieves multi-level early warning by combining dynamic threshold adjustment, which can avoid safety problems caused by untimely or no early warning.

[0019] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0020] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0021] In the attached diagram: Figure 1 This is a schematic diagram illustrating an exemplary system architecture as shown in an exemplary embodiment of this application; Figure 2 This is a schematic flowchart illustrating a temperature detection method according to an exemplary embodiment of this application; Figure 3 This is a schematic diagram of the regional distribution image acquired by a visible light camera, illustrating an exemplary embodiment of the temperature detection method of this application; Figure 4 This is a schematic diagram of an infrared thermal image acquired by a thermal imaging camera illustrating a temperature detection method according to an exemplary embodiment of this application; Figure 5 This is a schematic flowchart illustrating a temperature detection method according to another exemplary embodiment of this application; Figure 6This is a schematic flowchart illustrating a temperature detection method in yet another exemplary embodiment of this application; Figure 7 This is a block diagram illustrating a temperature detection device in an exemplary embodiment of this application; Figure 8 A schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present application is shown. Detailed Implementation

[0022] The following specific examples illustrate the implementation of this application. Those skilled in the art can easily understand other advantages and effects of this application from the content disclosed in this specification. This application can also be implemented or applied through other different specific embodiments. Various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of this application. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0023] It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of this application. The drawings only show the components related to this application and are not drawn according to the actual number, shape and size of the components in the actual implementation. In the actual implementation, the shape, quantity and proportion of each component can be arbitrarily changed, and the layout of the components may also be more complex.

[0024] In the following description, numerous details are explored to provide a more thorough explanation of embodiments of the present application. However, it will be apparent to those skilled in the art that embodiments of the present application may be practiced without these specific details. In other embodiments, well-known structures and devices are shown in block diagram form rather than in detail to avoid obscuring embodiments of the present application.

[0025] First, it's important to clarify that RANSAC (Random Sample Consensus) is an efficient method for estimating mathematical model parameters from a sample set containing outliers. This algorithm assumes the data contains inliers that can be described by the model and outliers that deviate from the model. It initializes the model by randomly sampling a minimum sample set and iteratively filters the inlier set to optimize parameters based on a set threshold. The core process involves randomly sampling a subset of samples to construct the initial model, calculating the remainder error to filter for a consensus set, and selecting the largest consensus set through multiple samplings to determine the final model. Its optimization strategy can involve re-estimating parameters by improving the sampling method or expanding the inlier set. As a fundamental algorithm in computer vision, RANSAC is commonly used in stereo vision for camera matching point problems and fundamental matrix calculations. The iterative mechanism proposed in the original paper provides the basic framework for subsequent model fitting.

[0026] Image registration is the process of matching and overlaying two or more images acquired at different times, by different sensors (imaging devices), or under different conditions (weather, illumination, camera position and angle, etc.). It has been widely used in remote sensing data analysis, computer vision, image processing and other fields.

[0027] A homography matrix is ​​a 3×3 matrix in computer vision that describes the perspective transformation between two planes. It achieves point-to-point mapping using homogeneous coordinates and is widely used in image correction, stitching, and viewpoint transformation.

[0028] Figure 1 This is a schematic diagram illustrating an exemplary system architecture as shown in an exemplary embodiment of this application.

[0029] Reference Figure 1 As shown, the system architecture may include an image acquisition device 101 and a computer device 102. The computer device 102 may be at least one of a desktop graphics processing unit (GPU) computer, a GPU computing cluster, or a neural network computer. The image acquisition device 101 is used to acquire regional distribution images and heat distribution matrices of the target material. In this embodiment, after acquiring the above data, the image acquisition device 101 provides it to the computer device 102 for processing. Those skilled in the art can use the computer device 102 to segment the regional distribution image using a preset material distribution region segmentation model to obtain the target material region. The material distribution region segmentation model is trained based on image samples with material distribution region labels. The regional distribution image is mapped to the heat distribution matrix to obtain a homography transformation matrix. A deep learning registration model is constructed based on the homography transformation matrix. The regional distribution image and the heat distribution matrix are registered and aligned according to the deep learning registration model to obtain the registration result. Statistics are then performed on the registration result of the target material region to obtain temperature statistics, thereby completing the temperature detection of the target material. It should be noted that the image acquisition device 101 and computer device 102 provided in this embodiment are merely examples and should not impose any limitations on the functions and scope of use of the embodiments of this application.

[0030] It should be noted that the temperature detection method provided in this application embodiment is generally executed by computer device 102, and correspondingly, the temperature detection device is generally installed in computer device 102.

[0031] Figure 2This is a schematic flowchart illustrating a temperature detection method in an exemplary embodiment of this application. The temperature detection method can be executed by a computing processing device, which may be... Figure 1 The computer device 102 shown is illustrated. (Refer to...) Figure 2 As shown, the temperature detection method includes at least steps S210 to S240, which are described in detail below: In step S210, the regional distribution image and heat distribution matrix of the target material are obtained.

[0032] In one embodiment of this application, a regional distribution image of the target material is acquired using one or more of a visible light camera, a multispectral imaging device, etc., while a heat distribution matrix and an infrared thermal image (infrared thermal map) are acquired using one or more of an infrared thermal imaging camera, an infrared thermal imager, a thermal imaging camera, a thermal imager, etc. (See reference) Figure 3 and Figure 4 , Figure 3 This is a schematic diagram of the regional distribution image acquired by a visible light camera, illustrating an exemplary embodiment of the temperature detection method of this application. Figure 4 This is a schematic diagram of an infrared thermal image acquired by a thermal imaging camera, illustrating an exemplary embodiment of the temperature detection method of this application. It is understood that the infrared thermal image is an image corresponding to the heat distribution matrix, and the pixel coordinates in the infrared thermal image are the same as the corresponding pixel coordinates in the heat distribution matrix.

[0033] In one embodiment of this application, the heat distribution matrix is ​​determined by: obtaining the original temperature of the object and the wavelength of the electromagnetic wave; calculating the spectral radiant exitance of each pixel based on Planck's radiation law, the original temperature of the object, and the wavelength of the electromagnetic wave; calculating the radiation intensity received by the thermal imager detector at each pixel based on the spectral radiant exitance, the comprehensive response coefficient, and the operating band of the infrared thermal imager; and calculating the surface temperature of the object at each pixel based on the radiation intensity, the calibration constant, the emissivity of the object surface, and the environmental reflectivity, thereby obtaining the heat distribution matrix.

[0034] In this embodiment, taking an infrared thermal imager as an example, before connecting the infrared thermal imager, preset key parameters such as default thermal imager parameters, image registration related variables, and temperature analysis cache are set; during connection, the original temperature data of the object is acquired, and the matrix data is of floating-point type. The spectral radiative exitance of each pixel is calculated using Planck's radiation law. The temperature of each pixel is calculated by the thermal imager detector. The entire heat distribution matrix has been calculated, and the relevant formulas are as follows: Equation (1) Equation (2) Equation (3) in, For the pixels of an object at wavelength and temperature Spectral radiative exitance at the following levels The wavelength of electromagnetic waves, The original temperature of the object (in K). is Planck's constant. At the speed of light, Boltzmann's constant; This represents the radiation intensity (digital value) received by the thermal imager detector at pixel (i,j). To achieve a comprehensive response coefficient, including detector sensitivity and gain, the radiated power is converted into an electrical signal. These represent the lower and upper limits of the operating wavelength of the infrared thermal imager, respectively, defining the spectral range that the thermal imager's detector is sensitive to. Indicates temperature as The object at wavelength Spectral radiative exitance at that location The spectral transmittance function represents the spectral transmittance of an optical system or medium, including the transmittance of lenses, filters, windows, etc., as well as the spectral response of detectors, with a value range of 0 to 1. This represents the surface temperature of the object obtained by inversion at pixel (i,j). For calibration constant, The emissivity of the object's surface (0.01~1.0). It is an environmental reflectance factor (related to the ambient temperature). These are calibration coefficients.

[0035] In step S220, the region distribution image is segmented using a preset material distribution region segmentation model to obtain the target material region.

[0036] The material distribution area segmentation model is trained based on image samples labeled with material distribution areas.

[0037] In one embodiment of this application, material distribution area labels and sample image data are obtained; the sample image data are labeled according to the material distribution area labels to generate image samples with material distribution area labels; a preset material distribution area segmentation model is trained based on the image samples with material distribution area labels to obtain a material distribution area segmentation model, wherein the preset material distribution area segmentation model is obtained by adding a scaling factor based on a self-attention mechanism.

[0038] In this embodiment, when annotating the sample image data used to train the visible light camera detection, a polygonal frame mask is used to annotate the material distribution area in the image, and the annotated image is stored in the computer memory.

[0039] When training a material distribution region segmentation model for material distribution region segmentation, the models used include, but are not limited to, Segformer-B0 and the YOLO series. To reduce the overall computational complexity of the model, a scaling factor R is added to the self-attention mechanism to reduce the computational complexity of each self-attention mechanism module. The principle of the self-attention mechanism can be expressed as follows: Equation (4) The computational complexity of self-attention is Based on this, a scaling factor R is added, specifically as follows: Equation (5) Equation (6) Self-attention mechanism Q , K , V All N*C The feature map, where N It is the number of all patches (image tiles). C This is the dimension corresponding to each patch. (Through...) Reshape Operation N*C The feature map is transformed into The feature map is then processed through a fully connected layer, and then... Transform into At this point, the computational complexity is . It is the attention head dimension, which is the subspace dimension of Q, K, and V that each attention head processes in a multi-head attention mechanism. Used for scaling dot products and stabilizing training.

[0040] In step S230, the regional distribution image is mapped to the heat distribution matrix to obtain the homography transformation matrix. A deep learning registration model is constructed based on the homography transformation matrix. The regional distribution image and the heat distribution matrix are registered and aligned according to the deep learning registration model to obtain the registration result.

[0041] In one embodiment of this application, multiple feature points and corresponding feature descriptors are extracted from the regional distribution image and the heat distribution matrix, respectively. Based on the feature descriptors, feature matching pairs between the regional distribution image and the heat distribution matrix are calculated, and the feature matching pairs are filtered using the nearest neighbor distance ratio and a bidirectional matching strategy to obtain a preliminary matching point set. A homography transformation matrix is ​​determined based on the preliminary matching point set using a random sampling consensus algorithm combined with the least squares method. The homography transformation matrix is ​​used to map the coordinate system of the regional distribution image to the heat distribution matrix. A deep learning registration model is constructed using the homography transformation matrix as initialization parameters. The deep learning registration model is optimized by minimizing the loss function to obtain an optimized deep learning registration model. The regional distribution image and the heat distribution matrix are registered and aligned based on the optimized deep learning registration model to obtain the registration result. The registration quality evaluation index of the registration result is calculated. If the registration quality evaluation index is less than or equal to a preset threshold, feature points are re-extracted until the registration quality evaluation index is greater than the preset threshold to update the registration result.

[0042] In this embodiment, the obtained regional distribution image and heat distribution matrix are registered. A feature point extraction and matching algorithm is then performed using a homography matrix projection geometric model, with the relevant formulas as follows: Equation (7) in, It is a visible light coordinate system matrix (i.e., the pixel coordinates of points in the regional distribution image). The coordinate system matrix is ​​the thermal imaging coordinate system matrix (i.e., the point coordinates in the heat distribution matrix are pixel coordinates). The H matrix is ​​the homography transformation matrix, used to describe the projection transformation between two planes (including rotation, translation, scaling, affine transformation, etc.). ~ These are the nine parameters of the homography transformation matrix.

[0043] Feature point detection scale space extrema, key point localization: Equation (8) Equation (9) Equation (10)

[0044] in, It is the scale-space representation obtained by convolving image I with Gaussian kernel G. This is a Gaussian kernel function used to construct the image scale space, the width of which is determined by the scale parameter. control, This represents the grayscale value of the original input image at coordinates (x, y). The Taylor expansion used for precise keypoint localization is defined as follows: x is the offset relative to the candidate point, which can also be understood as a second-order Taylor expansion approximation of the Gaussian difference response value at offset x; D is the Gaussian difference response value at the candidate keypoint (initial detection position), used to determine whether it is a stable feature point; and x is an offset vector representing the coordinate offset from the candidate keypoint position to the true extremum point. It is typically two-dimensional (x,y) or three-dimensional (x,y,σ) (including scale offset). In FAST corner detection, the grayscale value of the center pixel p is... Let t be the grayscale value of a pixel x on the Bresenham circle (i.e., the circle obtained by the midpoint circle drawing algorithm) surrounding the center pixel p, and t be the grayscale difference threshold of FAST corner detection, used to determine whether it is a corner.

[0045] When calculating feature matching, the nearest neighbor distance ratio and bidirectional matching must meet the following requirements: Equation (11) Equation (12) Equation (13) NNDR, or Nearest Neighbor Distance Ratio, is used to filter matches. The smaller the ratio, the more unique and reliable the match. The distance (minimum distance) for the nearest neighbor match. and The distance for matching the next nearest neighbor (the second and third smallest distances). The nearest neighbor distance ratio threshold, and This is a descriptor vector for the i-th feature point in the visible light image (i.e., the regional distribution image) and the j-th feature point in the infrared thermal image. and It is the result of bidirectional matching, ensuring that the optimal match from the visible light image to the infrared thermal image is consistent with the reverse match.

[0046] The above process extracts multiple feature points and their corresponding feature descriptors from the regional distribution image and the heat distribution matrix. Feature matching pairs are calculated using the feature descriptors and filtered using Nearest Neighbor Distance Ratio (NNDR) and bidirectional matching to obtain a preliminary set of matching points. Stable key points are detected using Gaussian kernels, and corner points are quickly detected using pixel grayscale comparison. Reliable matches are filtered using Nearest Neighbor Distance Ratio (NNDR) to eliminate ambiguous corresponding points. Bidirectional matching ensures the consistency of the matching.

[0047] After establishing a mathematical coordinate model and completing feature extraction and matching, the transformation matrix is ​​estimated and deep learning registration is introduced to improve robustness and enhance adaptability to complex scenes; the registration quality is also evaluated to ensure system stability and facilitate automatic recalibration. The relevant least squares method and RANSAC algorithm are as follows: Equation (14) Equation (15) An overdetermined system of equations is constructed by matching point pairs (i.e. feature matching pairs), and the eight parameters of the homography transformation matrix H are solved (h33 is usually normalized to 1).

[0048] in, This is the distance threshold for determining interior points in the RANSAC algorithm. If the distance between the transformed point and the target point is less than... If , then it is considered an interior point. Let be the homogeneous coordinates of the i-th feature point in the regional distribution image, usually expressed as ,in These are the pixel coordinates of the point in the image; Let be the homogeneous coordinates of the i-th feature point in the infrared thermal image, usually expressed as H is the homography transformation matrix.

[0049] Deep learning registration loss function: Equation (16) in, It is the spatial deformation field learned by the deep learning registration network, which defines the displacement required for each pixel; These are infrared thermal images corresponding to the regional distribution image and the heat distribution matrix, respectively. Indicates the deformation field Effect on infrared thermal images Align it with the regional distribution image; This is a similarity metric function that measures the degree of alignment between two images. This is the regularization weight coefficient, used to balance the importance of the similarity term and the regularization term; For regularization terms, the deformation field Apply smoothing constraints to prevent it from producing unphysical distortions.

[0050] The RANSAC algorithm randomly selects four pairs of matching points and calculates the homography transformation matrix. Calculate the number of interior points until the largest set of interior points is found, then re-estimate using all interior points. Finally, reprojection error, mutual information, and structural similarity are calculated to assess the registration quality. Equation (17) Equation (18) Equation (19) in, The reprojection error is the average positional error of all matching points after projection through the homography transformation matrix H. The smaller the value, the higher the geometric alignment accuracy. N is the number of matching point pairs used to calculate the error. Mutual information is used to evaluate the statistical dependence between two registered images A and B. The larger the value, the higher the information correlation and the better the registration. Joint probability distribution of gray values ​​in images A and B; Represents the edge probability distribution of grayscale values ​​in images A and B; The structural similarity index is represented by brightness (μ), contrast (σ), and structure (σ). Image similarity is evaluated from three aspects, with a range of [-1, 1], and the closer to 1 the better; c1 and c2 are small constants in the SSIM calculation, used to prevent the denominator from being zero and to stabilize the calculation.

[0051] The above process uses the RANSAC algorithm combined with the least squares method to estimate the homography transformation matrix H from the matching points (i.e., the initial matching point set). Starting with the homography transformation matrix H, the deformation field is further optimized using a deep learning loss function, improving the registration robustness in complex scenarios. Reprojection error, MI, SSIM, and other metrics are calculated; if they fail to meet the requirements, automatic recalibration is triggered to ensure the long-term stability of the system.

[0052] In step S240, the temperature statistics are obtained by statistically analyzing the registration results of the target material area to complete the temperature detection of the target material.

[0053] In one embodiment of this application, the maximum temperature of the target material region is determined based on the registration result of the target material region; the average temperature, standard deviation of temperature, and temperature gradient of the target material region are calculated based on the registration result of the target material region; and the hot spot ratio of the target material region is calculated based on the average temperature and standard deviation of temperature. The temperature statistics include the maximum temperature, average temperature, standard deviation of temperature, temperature gradient, and hot spot ratio.

[0054] In this embodiment, a time-series database storage module is designed to store temperature statistics for each material area. Considering the high-frequency data acquisition requirements of industrial scenarios, a storage bucket is created using a time-series database, with the data retention period adjusted according to needs. It stores preset key statistics such as material area ID and material type. Temperature data and other key statistics are displayed as floating-point numbers within the domain; alternatively, the entire histogram can be stored as a string. Additional fields, such as the number of sample points, can also be stored, with key values ​​stored as separate fields. Integration with the temperature monitoring system is then completed. The relevant data and calculation formulas are as follows: Equation (20) Equation (21) Equation (22) Hotspot ratio: Equation (23) Temperature gradient: Equation (24) in, The average temperature is N, and the number of sample points is N. The temperature value calculated for the i-th pixel within the target material area; For temperature standard deviation, This represents the maximum temperature. Hotspot ratio is the core indicator for anomaly warning. It represents the proportion of pixels whose temperature exceeds "mean + 2 standard deviation". This threshold is based on statistical principles and can automatically adapt to normal fluctuations under different working conditions, and sensitively capture statistically abnormal high temperature areas. The temperature gradient characterizes the maximum rate of temperature change in a two-dimensional plane. In an image, it is obtained by calculating the partial derivatives of each pixel in its x and y directions. Areas with large gradients are the boundaries of drastic temperature changes.

[0055] The above process establishes a long-term, traceable temperature record for each material area (such as a piece of material on a conveyor belt).

[0056] In one embodiment of this application, after obtaining temperature statistics based on the registration results of the target material area, the method further includes: determining a temperature feature vector based on the temperature statistics; obtaining an initial time series based on the temperature feature vectors of multiple target times; normalizing the initial time series to obtain a target time series; inputting the target time series into a preset anomaly probability calculation model to obtain an anomaly probability value; determining a target warning level based on the comparison result between the anomaly probability value and a preset probability threshold; and issuing a warning based on the target warning level, wherein the preset probability threshold is determined based on the actual warning frequency and the target warning frequency within a preset time period.

[0057] In one embodiment of this application, after issuing an early warning based on the target warning level, the method further includes: enhancing the features of the target time series to obtain an enhanced feature sequence; inputting the enhanced feature sequence into a preset sequence prediction model to obtain predicted values ​​of temperature feature vectors for multiple future times; and calculating the temperature prediction mean and temperature prediction variance for each future time based on the predicted values ​​of the temperature feature vectors for multiple future times, wherein the temperature prediction mean is used as the prediction result for the future time, and the temperature prediction variance is used as an uncertainty confidence index characterizing the prediction result.

[0058] In this embodiment, the similarity between the current sequence (i.e., the target time series) and the historical template sequences in the historical abnormal sequence template library is calculated using a dynamic time warping algorithm, which serves as an additional feature reflecting the correlation between the current pattern and the historical abnormal pattern; the dynamic change coefficients implicit in the temperature feature vector sequence are extracted based on a time-varying autoregressive model, which serve as an additional feature reflecting the time-varying law of the sequence itself.

[0059] In this embodiment, a one-dimensional convolutional neural network (1D-CNN) combined with a long short-term memory network (LSTM) is used to construct a time-series temperature early warning model. Historical temperature data is obtained from a time-series database; the data is preprocessed (normalization, construction of time window datasets); after inputting the time-series data, the real-time anomaly probability and the predicted future temperature using a seq2seq structure are output. The convolution formula for the 1D convolutional layer and the calculation function for the LSTM unit are as follows: 1D Convolution: Equation (24) Equation (25) Equation (26) Equation (27) Equation (28) Equation (29) Equation (30) Where l is the network layer index, indicating that the current layer is l. Indicates the lth The output feature value of layer 1 at time point t+k, for the first layer of the CNN, its input x (0) The target time series after preprocessing K is the kernel size, which defines the length of the local time window through which the CNN observes the input sequence, and k is the index inside the kernel, ranging from 0 to K. 1. Used to iterate through each time point within the convolution window. These are the weights of the l-th convolutional kernel at the k-th position. These weights are learned by the model through training and are used to capture meaningful local patterns (such as sudden temperature rises, sustained high values, etc.). It is the bias term of the l-th layer. This is the output feature of the l-th layer at time t. It is a weighted sum of all input features within the convolutional window, representing the local temporal pattern extracted near time t. 1D-CNN acts like a small sliding window, automatically scanning time series and efficiently capturing short-term, localized temperature change patterns and dependencies, providing more informative features for subsequent LSTM.

[0060] For the input at the current time t, this is typically the output feature of a 1D-CNN at time t for the first layer of an LSTM. For subsequent LSTM layers, it is the hidden state of the previous LSTM layer at time t. ; It is the previous time, i.e., t. The hidden state at time 1 encodes a summary of all historical sequence information up to the previous time step, which is the key to the memory ability of LSTM. It's a forget gate, which uses the sigmoid function to output a vector between 0 and 1, determining the previous cell state C. t How much information needs to be forgotten (0 means completely forgotten, 1 means completely retained)? The input gate also outputs a vector between 0 and 1, which determines the current candidate value. How much new information needs to be stored in the cellular state? It is the candidate cell state, determined by the current input. and the previous hidden state It is generated using the tanh activation function and contains new information that may be updated at the current moment. This represents the current state of the cell. It is the output gate, which determines the current cell state. How much information needs to be output to the hidden state? middle; The hidden state at the current moment; , , and The weight matrices, corresponding to the forget gate, input gate, candidate state, and output gate respectively, are the core parameters that the model needs to train. , , and These correspond to the bias vectors of each of the above gates; It is the Sigmoid activation function, which compresses the input to the (0,1) interval and generates a gated signal; It is the hyperbolic tangent activation function, which compresses the input to the (-1, 1) interval to regulate the information flow. This is understandable. The Hadamard product is the element-wise multiplication of a matrix or vector. LSTM uses these gating parameters to selectively memorize long-term historical patterns (such as slow warming trends or periodic production rhythms) and forget irrelevant information, thereby understanding the complex dynamic relationships in temperature sequences that span long time intervals.

[0061] This model learns patterns from historical temperature data and predicts future temperature trends while detecting outliers. Real-time temperature data acquisition and preprocessing are performed: the system acquires temperature feature vectors from multiple sensor nodes at a fixed frequency, combines the current time-to-time feature with the features from the previous N-1 time-to-time points to form an initial time series, and normalizes the series using a pre-trained normalizer to obtain the target time series. The relevant formulas are as follows: Equation (31) Equation (32) Equation (33) in, For a single moment, the temperature feature vector, the average temperature, maximum temperature, temperature standard deviation, temperature gradient, and hotspot ratio distribution correspond to the average temperature, maximum temperature, temperature standard deviation, temperature gradient, and hotspot ratio in the temperature statistics data. The initial time series consists of the temperature feature vector at the current time and the temperature feature vectors from the previous N-1 time periods. It is a matrix of dimension (N,5) representing a historical sequence of N past moments, with 5 features at each moment; For the target time series, and These are the mean and standard deviation of each of the five features calculated on the training set, in order to eliminate the differences in the units and ranges of different features, accelerate model convergence, and improve stability.

[0062] The standardized target time series is input into a pre-trained 1D-CNN-LSTM model (i.e., a pre-defined anomaly probability calculation model), and the anomaly probability value at the current time step is output: Equation (34) in, This is the probability value for anomalies, ranging from (0,1). The closer it is to 1, the more anomalous the current state. Let e ​​be the target time series, and e be the natural constant. W is the final hidden state of the LSTM at time t, which encodes the summary information of the entire sequence. W and b are the weights and biases of the output layer.

[0063] The preset probability threshold is automatically adjusted based on the recent warning frequency. Equation (35) in, For the preset probability threshold, The basic early warning threshold is determined by historical data. To adjust the coefficient and control the adjustment range, This represents the actual warning frequency in the near future (within a preset timeframe). The target warning frequency is the frequency that the system expects.

[0064] The target warning level is determined based on the comparison between the anomaly probability value and the preset probability threshold, and a warning is issued according to the target warning level: Equation (36) in, This represents the probability value of anomalies. The first preset probability threshold can also be understood as the warning level threshold. The second preset probability threshold can also be understood as a severity level threshold. The preset probability threshold includes a first preset probability threshold and a second preset probability threshold, with the first preset probability threshold being lower than the second preset probability threshold. This allows for differentiated alarm responses, making it easier for operators to distinguish the severity of problems. (Reference) Figure 5 , Figure 5 This is a schematic flowchart illustrating a temperature detection method as shown in another exemplary embodiment of this application.

[0065] After making an early warning decision on whether to issue a real-time temperature alarm (understandably, when...), (In case of no real-time temperature alarm), the predicted values ​​of temperature feature vectors at multiple future time points are predicted using a sequence-to-sequence (Seq2Seq) structure (i.e., a sequence-to-sequence prediction model). Dynamic time warping similarity is used. and time-varying autoregressive coefficients Enhance time series features; modify the fully connected layer to output n nodes, calculate the binary cross-entropy for each node, and keep the warning decision tree for each prediction node unchanged. The relevant calculation formula is as follows: Equation (37) Equation (38) in, DTW similarity is the similarity between the current sequence Q (i.e., the target time series) and the historical typical anomalous sequence library C. High DTW similarity can serve as a feature to corroborate anomalous events. For the i-th target time series, Let j be the j-th historical template sequence. These are time-varying autoregressive coefficients, used to capture the dynamic changes of the temperature sequence itself over time, serving as supplementary features. and It is a coefficient that changes over time. It's noise.

[0066] Quantifying forecast uncertainty using Monte Carlo Dropout: Equation (39) Equation (40) in, The final predicted value is the average of N predictions, i.e., the average temperature prediction, where N is the number of random forward propagations. This represents the model's predicted output when the random Dropout mask is used for the i-th time. To predict uncertainty (temperature prediction variance), the larger the temperature prediction variance, the more uncertain the model is about this prediction.

[0067] The Seq2Seq model is used to predict future sequences, while Monte Carlo Dropout provides Bayesian uncertainty estimates. This not only gives the predicted value of the future temperature, but also the reliability of the predicted value, which is crucial for high-risk decision-making.

[0068] Figure 6 This is a schematic flowchart illustrating a temperature detection method in another exemplary embodiment of this application. In one exemplary embodiment, a temperature detection method includes the following steps: In step S1, a dual-channel camera is used to acquire the regional distribution image and heat distribution matrix of the material, respectively. The dual-channel camera includes a visible light camera and an infrared thermal imaging camera. In step S2, the image registration module performs image matching to obtain the mapping relationship between the regional distribution image and the heat distribution matrix; In step S3, the target material region in the regional distribution image is mapped to the heat distribution matrix, and temperature statistics are calculated based on the heat distribution matrix. In step S4, historical temperature data are collected and stored in a time series database; In step S5, a one-dimensional neural network early warning model is used to analyze the temperature data and make an early warning decision.

[0069] It should be noted that the temperature detection method provided in this embodiment belongs to the same concept as the temperature detection method provided in the above embodiments, and the specific methods of each step have been described in detail in the method embodiments, and will not be repeated here. In practical applications, the temperature detection method provided in the above embodiments can be assigned to different steps as needed, and this is not a limitation here.

[0070] This application simultaneously acquires regional distribution images and heat distribution matrices of materials using a visible light camera and an infrared thermal imaging camera. An image registration module establishes a mapping relationship between the visible light image and the infrared thermal image, achieving precise alignment of the material region to the temperature matrix. Based on Planck's radiation law, temperature data is calculated and statistical characteristics (including average temperature, standard deviation of temperature, hotspot ratio, and temperature gradient) are computed. Historical temperature data is stored in a time-series database. A one-dimensional CNN-LSTM neural network is used to construct an early warning model. Time-series analysis is used to predict temperature trends and detect anomalies. Combined with dynamic threshold adjustment, multi-level early warnings are achieved. This application features multi-modal accurate temperature measurement, high processing efficiency, and strong robustness, while also achieving dynamic monitoring, intelligent alarm, and timely early warning based on predicted future temperature trends.

[0071] Figure 7 This is a block diagram illustrating a temperature detection device according to an exemplary embodiment of this application. The device can be applied to… Figure 1 The implementation environment shown is specifically configured in computer device 102. This device can also be applied to other exemplary implementation environments and specifically configured in other devices. This embodiment does not limit the implementation environment to which this device is applicable.

[0072] like Figure 7 As shown, the exemplary temperature detection device includes: a data acquisition module 710, a region segmentation module 720, a registration module 730, and a statistics module 740.

[0073] The system includes: an acquisition module 710 for acquiring the regional distribution image and heat distribution matrix of the target material; a region segmentation module 720 for segmenting the regional distribution image using a preset material distribution region segmentation model to obtain the target material region, wherein the material distribution region segmentation model is trained based on image samples with material distribution region labels; a registration module 730 for mapping the regional distribution image to the heat distribution matrix to obtain a homography transformation matrix, constructing a deep learning registration model based on the homography transformation matrix, and registering and aligning the regional distribution image and heat distribution matrix according to the deep learning registration model to obtain the registration result; and a statistics module 740 for performing statistics based on the registration result of the target material region to obtain temperature statistics data, thereby completing the temperature detection of the target material.

[0074] It should be noted that the temperature detection device and the temperature detection method provided in the above embodiments belong to the same concept. The specific operation methods of each module and unit have been described in detail in the method embodiments and will not be repeated here. In practical applications, the temperature detection device provided in the above embodiments can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. This is not a limitation here.

[0075] Embodiments of this application also provide an electronic device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the temperature detection method as described in any of the above embodiments.

[0076] Figure 8 A schematic diagram of a computer system suitable for implementing the embodiments of this application is shown. It should be noted that... Figure 8 The computer system 800 of the electronic device shown is merely an example and should not impose any limitation on the functionality and scope of the embodiments of this application.

[0077] like Figure 8As shown, the computer system 800 includes a Central Processing Unit (CPU) 801, which can perform various appropriate actions and processes, such as executing the methods provided in the various embodiments above, based on a program stored in Read-Only Memory (ROM) 802 or a program loaded from storage portion 808 into Random Access Memory (RAM) 803. The RAM 803 also stores various programs and data required for system operation. The CPU 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0078] The following components are connected to I / O interface 805: an input section 806 including a keyboard, mouse, etc.; an output section 807 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN (Local Area Network) card, modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to I / O interface 805 as needed. A removable medium 811, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 810 as needed so that computer programs read from it can be installed into storage section 808 as needed.

[0079] Specifically, according to embodiments of this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program including a computer program for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 809, and / or installed from removable medium 811. When the computer program is executed by central processing unit (CPU) 801, it performs various functions defined in the system of this application.

[0080] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disc read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying a computer-readable computer program. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The computer program contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0081] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. Each block in a flowchart or block diagram may represent a module, segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0082] The units described in the embodiments of this application can be implemented in software or hardware, and the described units can also be located in a processor. The names of these units do not necessarily limit the specific unit itself.

[0083] Another aspect of this application provides a computer-readable storage medium storing a computer program that is executed by a processor to implement the temperature detection method as described in any of the above embodiments. This computer-readable storage medium may be included in the electronic device described in the above embodiments, or it may exist independently and not assembled into the electronic device.

[0084] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0085] Another aspect of this application provides a computer program product or computer program including computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the temperature detection methods provided in the various embodiments described above.

[0086] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the method according to the embodiments of this application.

[0087] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0088] The above embodiments are merely illustrative of the principles and effects of this application and are not intended to limit this application. Any person skilled in the art can modify or alter the above embodiments without departing from the spirit and scope of this application. Therefore, all equivalent modifications or alterations made by those skilled in the art without departing from the spirit and technical concept disclosed in this application should still be covered by the claims of this application.

Claims

1. A temperature detection method, characterized in that, include: Obtain the regional distribution image and heat distribution matrix of the target material; The region distribution image is segmented using a preset material distribution region segmentation model to obtain the target material region. The material distribution region segmentation model is trained based on image samples with material distribution region labels. The regional distribution image is mapped to the heat distribution matrix to obtain a homography transformation matrix. A deep learning registration model is constructed based on the homography transformation matrix. The regional distribution image and the heat distribution matrix are registered and aligned according to the deep learning registration model to obtain the registration result. Based on the registration results of the target material area, statistical data on temperature is obtained to complete the temperature detection of the target material.

2. The temperature detection method according to claim 1, characterized in that, The material distribution area segmentation model is trained based on image samples labeled with material distribution areas, including: Acquire material distribution area labels and sample image data; The sample image data is labeled according to the material distribution area label to generate image samples with material distribution area labels; The preset material distribution area segmentation model is trained based on the image samples with material distribution area labels to obtain the material distribution area segmentation model, wherein the preset material distribution area segmentation model is obtained by adding a scaling factor based on a self-attention mechanism.

3. The temperature detection method according to claim 1, characterized in that, The heat distribution matrix is ​​determined using the following method: To obtain the object's original temperature and electromagnetic wave wavelength; Based on Planck's radiation law, the original temperature of the object, and the wavelength of the electromagnetic wave, the spectral radiative exitance of each pixel is calculated. Based on the spectral radiative exitance, the comprehensive response coefficient, and the operating band of the infrared thermal imager, the radiation intensity received by the thermal imager detector at each pixel is calculated. The surface temperature of the object at each pixel is calculated based on the radiation intensity, calibration constant, emissivity of the object surface, and environmental reflectivity, thereby obtaining the heat distribution matrix.

4. The temperature detection method according to claim 1, characterized in that, The regional distribution image is mapped to a heat distribution matrix to obtain a homography transformation matrix. A deep learning registration model is constructed based on the homography transformation matrix. The regional distribution image and the heat distribution matrix are registered and aligned according to the deep learning registration model to obtain the registration result, including: Multiple feature points and feature descriptors corresponding to each feature point are extracted from the regional distribution image and the heat distribution matrix, respectively. Based on the feature descriptor, feature matching pairs between the regional distribution image and the heat distribution matrix are calculated, and the feature matching pairs are filtered using the nearest neighbor distance ratio and a bidirectional matching strategy to obtain a preliminary set of matching points. A homography transformation matrix is ​​determined based on the preliminary matching point set using a random sampling consensus algorithm combined with the least squares method. The homography transformation matrix is ​​used to map the coordinate system of the regional distribution image to the heat distribution matrix. Using the homography transformation matrix as initialization parameters, a deep learning registration model is constructed. The deep learning registration model is then optimized by minimizing the loss function to obtain the optimized deep learning registration model. Based on the optimized deep learning registration model, the regional distribution image and the heat distribution matrix are registered and aligned to obtain the registration result. Calculate the registration quality evaluation index of the registration result. If the registration quality evaluation index is less than or equal to a preset threshold, re-extract feature points until the registration quality evaluation index is greater than the preset threshold, so as to update the registration result.

5. The temperature detection method according to any one of claims 1 to 4, characterized in that, Based on the registration results of the target material region, statistical data on temperature is obtained, including: Based on the registration results of the target material region, determine the maximum temperature of the target material region; Based on the registration results of the target material region, calculate the average temperature, standard deviation of temperature, and temperature gradient of the target material region; The hot spot ratio of the target material area is calculated based on the average temperature and the standard deviation of the temperature, wherein the temperature statistics include the maximum temperature, the average temperature, the standard deviation of the temperature, the temperature gradient, and the hot spot ratio.

6. The temperature detection method according to any one of claims 1 to 4, characterized in that, After obtaining temperature statistics based on the registration results of the target material region, the method further includes: Determine the temperature feature vector based on the temperature statistics; The initial time series is obtained based on the temperature feature vectors of multiple target times; The initial time series is normalized to obtain the target time series; The target time series is input into a preset anomaly probability calculation model to obtain anomaly probability values; The target warning level is determined based on the comparison result between the abnormal probability value and the preset probability threshold, and a warning is issued according to the target warning level. The preset probability threshold is determined based on the actual warning frequency and the target warning frequency within a preset time period.

7. The temperature detection method according to claim 6, characterized in that, After issuing an early warning based on the target warning level, the method further includes: The target time series is subjected to feature enhancement to obtain an enhanced feature sequence; The enhanced feature sequence is input into a preset sequence prediction model to obtain predicted values ​​of temperature feature vectors for multiple future times. Based on the predicted values ​​of the temperature feature vectors at multiple future times, the mean temperature prediction and the variance temperature prediction for each future time are calculated respectively. The mean temperature prediction is used as the prediction result for the future time, and the variance temperature prediction is used as an uncertainty confidence index characterizing the prediction result.

8. A temperature detection device, characterized in that, include: The acquisition module is used to obtain regional distribution images and heat distribution matrices of the target material; The region segmentation module is used to segment the region distribution image using a preset material distribution region segmentation model to obtain the target material region. The material distribution region segmentation model is trained based on image samples with material distribution region labels. The registration module is used to map the regional distribution image to the heat distribution matrix to obtain a homography transformation matrix, construct a deep learning registration model based on the homography transformation matrix, and perform registration and alignment of the regional distribution image and the heat distribution matrix according to the deep learning registration model to obtain the registration result. The statistics module is used to perform statistics based on the registration results of the target material area to obtain temperature statistics data, so as to complete the temperature detection of the target material.

9. An electronic device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the temperature detection method as described in any one of claims 1 to 7.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it implements the temperature detection method as described in any one of claims 1 to 7.

Images