Coal fire abnormal temperature intelligent identification method based on multi-source time sequence thermal infrared satellite image

Abstract

The invention discloses a coal fire abnormal temperature intelligent identification method based on a multi-source time sequence thermal infrared satellite image, and the method comprises the steps: carrying out the selection of recent multi-source thermal infrared and visible light data of a plurality of satellites, and reducing the influence of a cloud layer on a remote measurement result; a specific correction model is combined with digital elevation data to carry out topographic correction on the satellite thermal infrared data, and the influence of the topographic effect in the thermal infrared satellite image on a temperature value in coal field fire area identification is weakened; then training a YOLOv5 model to detect a coal field area and a non-coal field area; visible light and thermal infrared satellite image data are fused at the same time, the influence of human activities on coal field fire area thermal infrared satellite image recognition is weakened, finally, time sequence change characteristics of satellite image temperature data with the continuous time of at least one year are extracted, and a coal field fire area thermal infrared satellite image recognition method is obtained by fusing a U-Net-CBAM model of an attention mechanism. Finally, the identification efficiency and accuracy of thermal infrared satellite image thermal anomaly are effectively improved, and accurate identification and positioning of a coal field abnormal temperature region in a complex terrain global region are realized.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of coalfield fire area identification, in particular to a coal fire abnormal temperature intelligent identification method based on multi-source time-series thermal infrared satellite images. BACKGROUND

[0002] Coal fire combustion will produce and release CO, C2H2, CH4 and other toxic and harmful gases and emit a large amount of heat into the atmosphere, while forming a large area of burned-out area and causing ground subsidence, which seriously endangers the health of local residents and the ecological environment. Coal fires are mainly distributed in some complex terrain and rarely visited areas. Conventional underground coal fire detection technology is limited by detection equipment and personnel costs, and it is difficult to find burning coalfield fire areas in time and effectively manage them. Remote sensing detection method can use the high temperature, smoke and dust and other associated characteristics of coal fire to remotely monitor the entire range of coalfield in unpopulated areas, so as to achieve the purpose of early detection and early management of coal fire.

[0003] The existing remote sensing detection technology mainly uses thermal infrared remote sensing to detect coalfield fire areas, but this detection technology is easily affected by factors such as cloud cover, terrain, specific heat capacity of surface deposits, human activities and seasonal changes, which makes it prone to misjudgment of non-fire areas when extracting thermal anomalies. Especially in complex terrain areas, the misjudgment probability is higher due to the influence of terrain.

[0004] Therefore, how to provide a new coal fire abnormal temperature intelligent identification method can effectively reduce the influence of complex terrain on identification, so as to realize accurate identification and positioning of abnormal temperature areas of coalfield in complex terrain, which is the research direction of the present application. SUMMARY

[0005] In view of the problems existing in the prior art, the present application provides a coal fire abnormal temperature intelligent identification method based on multi-source time-series thermal infrared satellite images, which combines visible light images, thermal infrared satellite images, terrain correction, ground feature mask and time-series temperature change characteristics, can effectively reduce the influence of complex terrain on identification, so as to realize accurate identification and positioning of abnormal temperature areas of coalfield in complex terrain.

[0006] In order to achieve the above purpose, the technical scheme adopted by the present application is as follows: a coal fire abnormal temperature intelligent identification method based on multi-source time-series thermal infrared satellite images, comprising the following steps: Step 1: Determine the geographic location information of the coalfield area to be monitored; Step 2: According to the geographic location of the coalfield area, obtain the multi-source thermal infrared satellite image data and visible light image data of the area within a certain period of time through satellite; Step 3: Perform temperature inversion on the obtained multi-source thermal infrared satellite image data to obtain the surface inversion temperature of different positions in the coalfield area, and form a thermal infrared temperature image; Step four, according to the geographical location of the coalfield region, the corresponding digital elevation data is obtained, then the terrain correction of the thermal infrared temperature image is carried out combined with the digital elevation data, and the corresponding label is identified from the visible light image according to the preset target detection label; Step five, the corrected thermal infrared temperature image of the coalfield region and the coalfield region identified from the visible light image are fused to obtain the corrected thermal infrared temperature image; Step six, the temperature time series variation characteristics of the corrected thermal infrared temperature image are extracted; Step seven, the temperature time series variation characteristics obtained in step six are used to train the abnormal temperature identification model, and after completion, the model is used to identify the abnormal temperature region in the coalfield region.

[0007] Further, the geographical position information in step one includes the central position longitude and latitude coordinates of the coalfield region and the boundary range longitude and latitude coordinates of the coalfield region.

[0008] Further, the multi-source thermal infrared and visible light image data used in step two includes Landsat satellite thermal infrared and visible light image data, SDGSAT-1 satellite thermal infrared and visible light image data, and ZY1F satellite thermal infrared and visible light image data; and the data acquisition time period is at least one year of thermal infrared data from the current time.

[0009] Further, the thermal infrared temperature inversion function is used to perform temperature inversion on the multi-source thermal infrared satellite image data, specifically: The Landsat satellite thermal infrared temperature inversion function is: Where T L-LST is the Landsat satellite thermal infrared surface inversion temperature, T L-DN is the detection value of Landsat satellite thermal infrared data; The SDGSAT-1 satellite three-band (i.e. 8~10.5, 10.3~11.3, 11.5~12.5 μm) thermal infrared temperature inversion function is: Where d ra is the radiation brightness, d DN is the reflectivity value of SDGSAT-1 satellite thermal infrared data, a GAIN is the gain value of radiation calibration, bBIAS The offset value calibrated for radiation, c BG T is the cold space background correction coefficient; T BT The brightness temperature, k1 and k2 are the radiation calibration constants; h represents the Planck constant, c is the speed of light, k is the Boltzmann constant, and λ ef The effective wavelength of the sensor; T SDGSAT-1-LST The SDGSAT-1 satellite thermal infrared surface retrieval temperature, T1, T2, and T3 are the brightness temperatures of the three thermal infrared bands, and a1 and b1 are empirical coefficients. The ZY1F satellite thermal infrared temperature retrieval function is: Wherein E em The surface emissivity; T BT The brightness temperature; T ZY1F-LST The ZY1F satellite thermal infrared surface retrieval temperature; λ Wl The center wavelength of the thermal infrared band.

[0010] Further, the specific process of topographic correction in step four is: The thermal infrared temperature image is topographically corrected by using digital elevation data combined with C topographic correction model and SCS+C topographic correction model: The C topographic correction model: Wherein L cor The reflectivity of the topographically corrected thermal infrared temperature image; L obs The reflectivity of the original thermal infrared temperature image; θ i And θ z The solar zenith angle and the incidence angle determined according to the digital elevation data, respectively; C is an empirical coefficient; The SCS+C topographic correction model: Wherein θ v The solar reflection angle determined according to the digital elevation data.

[0011] Further, the target detection label in step four includes coalfield area and non-coalfield area such as town, water area, etc. The preset target detection label is input into the YOLOv5 target detection model for training, and after completion, the coalfield area and non-coalfield area are identified from the visible light image.

[0012] Further, the step five adopts a space matching method to fuse the coalfield area recognized by the visible light image and the thermal infrared temperature image to obtain a corrected thermal infrared temperature image. The specific process is as follows: the coalfield area recognized by the visible light image is corresponded to the corresponding area of the thermal infrared temperature image, and the area is extracted to be the corrected thermal infrared temperature image (thermal infrared temperature image of the coalfield area).

[0013] Further, the abnormal temperature recognition model in the step seven is a U-Net-CBAM model with a fusion attention mechanism, and the U-Net model architecture is as follows: ①, down-sampling; Convolution layer: wherein the input temperature time series change feature map , H is the height, W is the width, and C is the number of channels; the convolution kernel , k is the kernel size, D is the output channel number; the bias ; the convolution operation is represented by *; σ a is the ReLU activation function, ; the output feature size is ; the maximum pooling layer: ; ②, up-sampling; Transposed convolution: ; wherein the input temperature time series change feature map ; the transposed convolution kernel ; the output feature map size is expanded to 2h×2w; the bias value is b; Skip connection: ; The feature map of the first layer of the encoder is spliced with the feature map of the corresponding layer of the decoder; ③, joint loss function; Weighted cross-entropy loss: Dice loss: Joint loss function: ; wherein w is the weight of the thermal anomaly pixel, is the real label, L Dice is the predicted value, and α is the hyperparameter adjustment weight; ④, back propagation and optimization; ; wherein θ is the gradient descent optimization parameter, η is the learning rate, is the gradient of the loss function to the parameter.

[0014] Further, the CBAM module is embedded between the encoder, the decoder and the skip connection layer of the U-Net model architecture, and the CBAM module comprises a channel attention module and a spatial attention module; The channel attention module comprises: Wherein represents the input temperature time series variation feature map; and are global average pooling and global maximum pooling in the spatial dimension respectively, and the output dimension is ; the MLP is a multi-layer perceptron, comprising two fully connected layers with ReLU activation functions in between; and a is a Sigmoid activation function, which normalizes the output; The spatial attention module comprises: Wherein represents that the average pooling and maximum pooling results in the channel dimension are spliced into a two-channel feature map; is a convolution operation using a 7*7 convolution kernel; and a is a Sigmoid function, and the output is a spatial attention map .

[0015] Further, the method further comprises step eight: based on the abnormal temperature area identified in step seven, collecting temperature data of the abnormal temperature area on site and forming a temperature three-dimensional distribution map of the area to verify the abnormal temperature area on site.

[0016] Compared with the prior art, the present application adopts the combination of visible light images, thermal infrared satellite images, terrain correction, ground object mask and time series temperature variation characteristics, and has the following advantages: 1. In view of the problem that the coalfield fire area in the remote sensing image is affected by clouds and terrain, and it is difficult to realize accurate detection of temperature anomalies of the coalfield fire area; in the present application, a plurality of satellite multi-source thermal infrared and visible light image data in recent years are selected according to the geographical location information of the coalfield, thereby reducing the influence of clouds on the telemetry result. In addition, C correction and SCS+C correction models are used in combination with digital elevation data to perform terrain correction on the satellite thermal infrared data, thereby weakening the influence of the terrain effect in the thermal infrared satellite image on the temperature value in the identification of the coalfield fire area, and improving the cleanliness of the input thermal infrared data of the coalfield anomaly identification model.

[0017] 2. In view of the problem of misjudgment of non-mining area temperature anomaly area caused by human activities, a YOLOv5 model is trained based on visible light satellite image data to detect coalfield area and non-coalfield area; meanwhile, by fusing visible light and thermal infrared satellite image data, the influence of human activities on the identification of coalfield fire area in thermal infrared satellite image is weakened, and the cleanliness of the input thermal infrared data of the coalfield fire area thermal anomaly identification model is improved.

[0018] 3. In view of the problems of low efficiency and lagging identification of thermal anomaly of coalfield fire area in remote sensing image, the thermal infrared satellite image after terrain correction and target identification is used as source data, the time sequence change characteristics of satellite image temperature data for at least 1 year are extracted, the U-Net-CBAM model with attention mechanism is fused, and the above processing of infrared data cleanliness is combined, so that the identification efficiency and accuracy of thermal infrared satellite image thermal anomaly are effectively improved, and the precise identification and positioning of the abnormal temperature area of the coalfield in the complex terrain are realized. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 It is a coalfield temperature anomaly area identification and verification flowchart of the embodiment of the present application; Figure 2 It is a satellite thermal infrared satellite image temperature inversion result graph of a certain coalfield mining area in the embodiment of the present application; Figure 3 It is a C, SCS+C terrain correction model training flowchart in the embodiment of the present application; Figure 4 It is a terrain correction comparison graph based on Landsat satellite thermal infrared satellite image data in the embodiment of the present application; Among them, (a) is the original temperature inversion graph, (b) is the C terrain correction effect graph, and (c) is the SCS+C terrain correction effect graph; Figure 5 It is a thermal infrared satellite image ground feature mask flowchart based on the YOLOv5 model in the embodiment of the present application; Figure 6 It is a time sequence change characteristic graph of the average temperature of a certain mining area thermal infrared satellite image after processing in the embodiment of the present application; Figure 7 It is a U-Net model structure graph in the embodiment of the present application; Figure 8 It is a CBAM module structure graph in the embodiment of the present application; Figure 9 It is a comparison graph of known abnormal temperature area of coalfield and abnormal temperature identification result of the embodiment of the present application; Figure 10 It is an abnormal temperature distribution graph of part of the coalfield in Xinjiang identified by the embodiment of the present application; Figure 11An abnormal temperature region temperature distribution map obtained by using an unmanned aerial vehicle on-site verification method; Figure 12 An abnormal temperature region temperature distribution map obtained by using a handheld thermal infrared instrument on-site verification method. DETAILED DESCRIPTION

[0020] The present application will be further described below.

[0021] As shown in Figure 1 , the present embodiment comprises the following steps: Step one, determine the geographical position information of the required monitoring coalfield region, including the central position latitude and longitude coordinates of the coalfield region and the boundary range latitude and longitude coordinates of the coalfield region.

[0022] Step two, according to the geographical position of the coalfield region, obtain the multi-source thermal infrared satellite image data and visible light image data of the region within a previous period of time through satellite; the multi-source thermal infrared and visible light image data used includes Landsat satellite thermal infrared and visible light image data, SDGSAT-1 satellite thermal infrared and visible light image data and ZY1F satellite thermal infrared and visible light image data; and the data acquisition time period is at least one year of thermal infrared data from the current time.

[0023] Step three, use the thermal infrared temperature inversion function to perform temperature inversion on the multi-source thermal infrared satellite image data to form a thermal infrared temperature image as shown in Figure 2 , specifically: The Landsat satellite thermal infrared temperature inversion function is: Where T L-LST is the Landsat satellite thermal infrared surface inversion temperature, T L-DN is the detection value of the Landsat satellite thermal infrared data; The SDGSAT-1 satellite three-band (i.e. 8~10.5, 10.3~11.3, 11.5~12.5 μm) thermal infrared temperature inversion function is: Where d ra is the radiation brightness, d DN is the reflectivity value of the SDGSAT-1 satellite thermal infrared data, a GAIN is the gain value of radiation calibration, b BIASOffset value for radiation calibration, c BG is a cold space background correction coefficient; T BT is a brightness temperature, k1 and k2 are radiation calibration constants; h represents Planck's constant, c is the speed of light, k is the Boltzmann constant, and λ ef is a sensor effective wavelength; T SDGSAT-1-LST is a ZY1F satellite thermal infrared surface retrieval temperature, T1, T2, and T3 are brightness temperatures of three thermal infrared bands, and a1 and b1 are empirical coefficients; The ZY1F satellite thermal infrared temperature retrieval function is: wherein E em is a surface emissivity, [0.9, 0.99]; T BT is a brightness temperature; T ZY1F-LST is a ZY1F satellite thermal infrared surface retrieval temperature; λ Wl is a center wavelength of a thermal infrared band.

[0024] Step four, according to the geographical location of the coalfield region, corresponding digital elevation data is obtained, and the digital elevation data is resampled to match the satellite image resolution; then, the thermal infrared temperature image is terrain corrected in combination with the digital elevation data as Figure 3 shown, and the specific process is as follows: the thermal infrared temperature image is terrain corrected by using the digital elevation data in combination with a C terrain correction model and an SCS+C terrain correction model: The C terrain correction model is: wherein L cor is a reflectivity of the thermal infrared temperature image after terrain correction; L obs is a reflectivity of the original thermal infrared temperature image; θ i and θ z are a solar zenith angle and an incident angle respectively determined according to the digital elevation data; and C is an empirical coefficient; The SCS+C terrain correction model is: wherein θ v is a solar reflection angle determined according to the digital elevation data. As an example, Figure 4 is a terrain correction comparison chart based on Landsat satellite thermal infrared satellite image data.

[0025] Meanwhile, according to a preset target detection label, it includes a coalfield region and a non-coalfield region such as a town, a water area, etc., as Figure 5As shown, the preset target detection labels are input into the YOLOv5 target detection model for training. After completion, coalfield areas and non-coalfield areas are identified from visible light images.

[0026] Step 5: Use spatial matching to fuse the corrected thermal infrared temperature image of the coalfield area with the coalfield area identified from the visible light image. The specific process is as follows: map the coalfield area identified from the visible light image to the corresponding area of ​​the thermal infrared temperature image, and extract the area to obtain the corrected thermal infrared temperature image (thermal infrared temperature image of the coalfield area).

[0027] Step Six: Extract temperature time-series variation features from the corrected thermal infrared temperature image for at least one year, such as... Figure 6 As shown; Step 7: Train the abnormal temperature identification model using the temperature time-series variation features obtained in Step 6. After completion, use the model to identify abnormal temperature regions in the coalfield area. Specifically, the abnormal temperature identification model is a U-Net-CBAM model that incorporates an attention mechanism, such as... Figure 7 As shown, the U-Net model architecture is as follows: ① Downsampling; Convolutional layers: The input temperature time-series variation feature map Where H is the height, W is the width, and C is the number of channels; convolution kernel Where k is the core size, D is the number of output channels; bias ; * indicates convolution operation; σ a It is the ReLU activation function. The output feature size is Max pooling layer: The spatial dimension is compressed by downsampling, the pooling window is 2×2, and the stride is 2. ② Upsampling; Transposed convolution: ; The input temperature time-series variation feature map Transposed convolution kernel The output feature map size is increased to 2h × 2w; the bias value is b. Jump links: ; The feature map of the encoder's first layer is concatenated with the feature map of the corresponding layer of the decoder; ③ Joint loss function; Weighted cross-entropy loss: Dice loss: The combined loss function is: ; Where w is the weight of the thermal anomaly pixel, is the true label, L Dice is the predicted value, and alpha is the hyperparameter adjustment weight. ④, back propagation and optimization; ; Where theta is the gradient descent optimization parameter, and eta is the learning rate, is the gradient of the loss function with respect to the parameters.

[0028] As Figure 8 shown, the CBAM module is embedded between the encoder, decoder and skip connection layer of the U-Net model architecture, and the CBAM module includes a channel attention module and a spatial attention module. Wherein the channel attention module: Wherein represents the input temperature time series variation feature map; and are global average pooling and global maximum pooling in the spatial dimension, respectively, and the output dimension is ; the MLP is a multilayer perceptron, including two fully connected layers with ReLU activation functions in between; sigma a is a Sigmoid activation function, which normalizes the output ([0, 1]) processing; The spatial attention module: Wherein represents the result of concatenating the average pooling and maximum pooling in the channel dimension into a two-channel feature map; is a convolution operation using a 7x7 convolution kernel; sigma a is a Sigmoid function, and the output spatial attention map .

[0029] Step eight: according to the abnormal temperature area identified in step seven, compare it with the abnormal temperature area of the proven coalfield as Figure 9 shown, the method of the present application can accurately locate the proven abnormal temperature area; Then, the method of the present application is used to locate the abnormal temperature area of the whole coalfield in Xinjiang as Figure 10 shown; in order to ensure the accuracy of the positioning of the present application, the temperature data of any abnormal temperature area is collected on site for on-site verification, specifically: according to the exploration range and terrain conditions of the coalfield site, the on-site temperature image data collection method is divided into two kinds: unmanned aerial vehicle and handheld thermal infrared instrument; The unmanned aerial vehicle on-site verification mode is: in a coalfield fire area with gentle topography and large monitoring range, hot infrared satellite image data of the coalfield is obtained by the unmanned aerial vehicle, EXIF information and RGB information in the original image data are combined into the converted hot infrared temperature image data in the back-end processing process, the number of feature points of the unmanned aerial vehicle hot infrared temperature image is increased, and a temperature distribution diagram of an abnormal temperature area is spliced. Figure 11 The temperature distribution diagram is verified for the abnormal temperature area by the unmanned aerial vehicle.

[0030] The handheld hot infrared instrument on-site verification mode is: in a coalfield fire area with large topography, temperature distribution data of the abnormal temperature area is remotely obtained by the handheld hot infrared instrument at a detection slope position. Figure 12 The temperature distribution diagram is verified for the abnormal temperature area by the handheld hot infrared instrument.

[0031] As shown in Figures 9 to 12 After a plurality of verifications, it is known that the method can effectively reduce the influence of complex topography on identification, so that accurate identification and positioning of the abnormal temperature area of the coalfield in the whole complex topography can be realized.

[0032] The above only describes the preferred embodiments of the present application, and it should be noted that: for ordinary skilled persons in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should also be considered as the protection scope of the present application.

Claims

1. A coal fire abnormal temperature intelligent identification method based on multi-source time-series thermal infrared satellite images, characterized in that, The method comprises the following steps: Step 1: determining the geographical position information of the coalfield area to be monitored; Step 2: obtaining multi-source thermal infrared satellite image data and visible light image data of the area within a previous period of time according to the geographical position of the coalfield area through a satellite; Step 3: performing temperature inversion on the obtained multi-source thermal infrared satellite image data to obtain the surface inversion temperature of different positions in the coalfield area and form a thermal infrared temperature image; Step 4: obtaining corresponding digital elevation data according to the geographical position of the coalfield area, then combining the digital elevation data to perform terrain correction on the thermal infrared temperature image, and simultaneously identifying the corresponding label from the visible light image according to a preset target detection label; Step 5: fusing the corrected thermal infrared temperature image of the coalfield area and the coalfield area identified from the visible light image to obtain a corrected thermal infrared temperature image; Step 6: extracting the temperature time sequence variation characteristics of the corrected thermal infrared temperature image; Step 7: training an abnormal temperature identification model using the temperature time sequence variation characteristics obtained in Step 6, and then using the model to identify abnormal temperature areas in the coalfield area.

2. The coal fire abnormal temperature intelligent identification method based on multi-source timing thermal infrared satellite images according to claim 1, characterized in that, The geographical position information in Step 1 includes the central position longitude and latitude coordinates of the coalfield area and the boundary range longitude and latitude coordinates of the coalfield area.

3. The coal fire abnormal temperature intelligent identification method based on multi-source timing thermal infrared satellite images according to claim 1, characterized in that, The multi-source thermal infrared and visible light image data used in Step 2 includes Landsat satellite thermal infrared and visible light image data, SDGSAT-1 satellite thermal infrared and visible light image data, and ZY1F satellite thermal infrared and visible light image data; and the data acquisition time period is at least one year of thermal infrared data from the current time.

4. The coal fire abnormal temperature intelligent identification method based on multi-source timing thermal infrared satellite images according to claim 3, characterized in that, In Step 3, a thermal infrared temperature inversion function is used to perform temperature inversion on the multi-source thermal infrared satellite image data, specifically: The Landsat satellite thermal infrared temperature inversion function is: where T L-LST is the Landsat satellite thermal infrared land surface retrieval temperature, T L-DN is the detected value of the Landsat satellite thermal infrared data; The SDGSAT-1 satellite three-band thermal infrared temperature inversion function is: where d ra is the radiance, d DN is the reflectance value of the SDGSAT-1 satellite thermal infrared data, a GAIN is the gain value of the radiometric calibration, b BIAS is the offset value of the radiometric calibration, c BG is the cold space background correction coefficient; T BT is the brightness temperature, k1 and k2 are the radiometric calibration constants; h represents the Planck constant, c is the speed of light, k is the Boltzmann constant, λ ef is the effective wavelength of the sensor; T SDGSAT-1-LST is the SDGSAT-1 satellite thermal infrared surface retrieval temperature, T1, T2, T3 are the brightness temperatures of the three thermal infrared bands, a1, b1 are empirical coefficients; The ZY1F satellite thermal infrared temperature inversion function is: where E em is the surface emissivity; T BT is the brightness temperature; T ZY1F-LST is the ZY1F satellite thermal infrared surface retrieval temperature; λ Wl is the center wavelength of the thermal infrared band.

5. The coal fire abnormal temperature intelligent identification method based on multi-source timing thermal infrared satellite images according to claim 1, characterized in that, In Step 4, the specific process of terrain correction is: Terrain correction of the thermal infrared temperature image is performed using digital elevation data combined with a C terrain correction model and an SCS+C terrain correction model: The C terrain correction model is: where L cor is the reflectance of the terrain-corrected thermal infrared temperature image; L obs is the reflectance of the original thermal infrared temperature image; θ i and θ z are the solar zenith angle and the incidence angle, respectively, determined from digital elevation data; and C is an empirical coefficient. The SCS+C terrain correction model is: where θ v is the solar reflection angle determined from digital elevation data.

6. The coal fire abnormal temperature intelligent identification method based on multi-source timing thermal infrared satellite images according to claim 1, characterized in that, In Step 4, the target detection label includes a coalfield area and a non-coalfield area, the preset target detection label is input into a YOLOv5 target detection model for training, and after the training, the coalfield area and the non-coalfield area are identified from the visible light image.

7. The coal fire abnormal temperature intelligent identification method based on multi-source timing thermal infrared satellite images according to claim 1, characterized in that, In Step 5, a spatial matching method is used to fuse the coalfield area identified from the visible light image with the thermal infrared temperature image to obtain a corrected thermal infrared temperature image.

8. The coal fire abnormal temperature intelligent identification method based on multi-source timing thermal infrared satellite images according to claim 1, characterized in that, In Step 7, the abnormal temperature identification model is a U-Net-CBAM model with a fusion attention mechanism, and the U-Net model architecture is: ① Down-sampling; Convolution layer: wherein the input temperature time series feature map wherein H is height, W is width, and C is the number of channels; a convolution kernel wherein k is the kernel size, and D is the number of output channels; a bias * denotes a convolution operation; σ a is a ReLU activation function, ; The output feature size is ; max pooling layer: ; ② Up-sampling; Transposed convolution: ; wherein the input temperature time-series feature map ; transpose convolution kernel ; The output feature map size is expanded to 2h×2w; the bias value is b; Jump connection: ; The feature map of the first layer of the encoder is spliced with the feature map of the corresponding layer of the decoder; ③ Joint loss function; Weighted cross-entropy loss: Dice loss: Joint loss function: ; where w is a thermal anomaly pixel weight, is the true label, L Dice is the predicted value, and a is a hyperparameter adjustment weight. (4) backpropagation and optimization; ; where θ is a gradient descent optimization parameter, and η is a learning rate, is the gradient of the loss function with respect to the parameters.

9. The coal fire abnormal temperature intelligent identification method based on multi-source timing thermal infrared satellite images according to claim 8, characterized in that, The CBAM module is embedded between the encoder, the decoder and the skip connection layer of the U-Net model architecture, and the CBAM module comprises a channel attention module and a spatial attention module. The channel attention module comprises: wherein represents the input temperature time series variation feature map; and are global average pooling and global max pooling in spatial dimension respectively, and the output dimension is ; MLP is a multi-layer perceptron, which contains two fully connected layers with ReLU activation function in between; σ a is a Sigmoid activation function to normalize the output. The spatial attention module comprises: wherein represents concatenating the results of average pooling and max pooling in the channel dimension into a two-channel feature map; is a convolution operation using a 7x7 convolution kernel; σ a is a Sigmoid function, outputting a spatial attention map .

10. The coal fire abnormal temperature intelligent identification method based on multi-source timing thermal infrared satellite images according to claim 1, characterized in that, The method further comprises the following step eight: based on the abnormal temperature area identified in step seven, collecting temperature data of the abnormal temperature area on site and forming a temperature three-dimensional distribution map of the area to verify the abnormal temperature area on site.

Images