Open pit coal mine spontaneous combustion identification method and system based on unmanned aerial vehicle thermal imaging image

Abstract

The invention discloses an open pit coal mine spontaneous combustion identification method and system based on an unmanned aerial vehicle thermal imaging image, and belongs to the technical field of end-to-end open pit coal mine spontaneous combustion analysis. The unmanned aerial vehicle mounts a camera to take off to execute a task and inspects the coal mining area; thermal infrared image acquisition is carried out on the monitored area in a continuous frame mode, and video data streams are transmitted back to the work station; and starting a spontaneous combustion analysis algorithm flow, extracting image frames in the video stream, inputting the image frames into an image processing unit, and identifying coal seam spontaneous combustion through an algorithm. The advantages of high temperature sensitivity and large-area rapid scanning of the infrared thermal imaging technology are utilized, early accurate recognition of the spontaneous combustion phenomenon of the open pit coal mine is achieved based on image analysis, tiny temperature changes can be detected, problems can be found in time in the early stage of spontaneous combustion risks, and resource waste, environmental pollution and safety accident risks caused by spontaneous combustion are reduced.

Description

Technical Field

[0001] The present invention belongs to the technical field of end-to-end open-pit coal mine spontaneous combustion analysis, and in particular relates to an open-pit coal mine spontaneous combustion identification method and system based on unmanned aerial vehicle thermal imaging images. Background Art

[0002] In terms of manual inspection, the efficiency is low and the labor intensity is high, making it difficult to conduct comprehensive and timely monitoring of large-scale open-pit coal mines. At the same time, affected by the experience and subjective judgment of the inspectors, the accuracy is not high, and it is easy to miss inspections and misjudgments, and manual inspections have safety hazards.

[0003] Existing monitoring equipment, such as temperature sensors and smoke detectors, is highly dependent on human intervention and cannot independently complete automatic monitoring tasks. Humans are required to regularly check whether the equipment is operating normally, read and analyze monitoring data, which not only increases labor costs, but also makes it difficult to ensure the timeliness and accuracy of monitoring. In addition, both manual inspections and existing monitoring equipment have problems such as limited coverage and restrictions on terrain conditions.

[0004] The limitations of manual inspections have seriously affected the effectiveness and efficiency of spontaneous combustion monitoring in open-pit coal mines, increasing safety risks and the possibility of resource waste. The manual dependence of existing monitoring equipment makes monitoring costs high and the reliability insufficient, which cannot meet the needs of open-pit coal mines for large-scale, efficient and accurate spontaneous combustion monitoring. Summary of the invention

[0005] In view of the shortcomings of the prior art, the purpose of the present invention is to provide a method and system for identifying spontaneous combustion in open-pit coal mines based on unmanned aerial vehicle thermal imaging images.

[0006] The technical solution adopted by the invention is: a method for identifying spontaneous combustion of open-pit coal mines based on UAV thermal imaging images, which is implemented by using an open-pit coal mine spontaneous combustion identification system based on UAV thermal imaging images. The technical key points are as follows:

[0007] Step 1: Plan the inspection route according to the actual terrain of the mining area;

[0008] Step 2: The drone takes off with the camera mounted on it to perform a mission and inspect the coal mining area;

[0009] Step 3: Collect thermal infrared images of the monitored area in a continuous frame manner and transmit the video data stream back to the workstation;

[0010] Step 4: Start the spontaneous combustion analysis algorithm process, extract the image frames in the video stream and input them into the image processing unit, and identify the spontaneous combustion of the coal seam through the algorithm;

[0011] Step 5: Store the algorithm output results to the workstation for subsequent visualization applications, and determine whether to trigger the alarm system based on the results.

[0012] Furthermore, the spontaneous combustion analysis algorithm described in step 4 is as follows:

[0013] Step 4.1, image acquisition: The thermal imaging camera acquires images of the monitored area at a rate of multiple frames per second;

[0014] Step 4.2, image preprocessing: use median filtering to remove image noise and improve image quality;

[0015] Step 4.3, feature extraction: complete instance segmentation of spontaneous combustion area through clustering;

[0016] Step 4.4, temperature change modeling: transform the self-ignition area from image space to temperature distribution space.

[0017] Furthermore, the step of completing the instance segmentation of the spontaneous combustion area through clustering described in step 4.3 is: through the FireNet model, all spontaneous combustion areas are automatically detected and segmented from the infrared image through semantic segmentation, and semantic segmentation, embedded vectorization, and clustering algorithms are used to instantiate each area, that is, to clarify the boundaries and pixel sets of each spontaneous combustion area, providing input for subsequent temperature modeling.

[0018] Furthermore, the semantic segmentation described in step 4.3 refers to:

[0019] The input infrared image is semantically segmented and a binary classification result of whether each pixel belongs to the spontaneous combustion area is output; the U-Net architecture is used to capture the local and global features of the image by adding the contextual information of the pixel in the encoder and decoder process.

[0020] Furthermore, the embedding vectorization described in step 4.3 means: generating an embedding vector for each pixel for subsequent clustering operations, using the U-Net architecture, and outputting a high-dimensional embedding vector; in the case of a small number of spontaneous combustion samples and imbalance: using the softmax_cross_entropy loss function and introducing the bounded inverse class weight, the formula is as follows:

[0021]

[0022] Among them, w i =c·(1-p i ), p i (2)

[0023] In the formula, p iis the probability of whether category i is a spontaneous combustion area in the overall sample, c is a hyperparameter, usually set to 1.02; y i,c is the true label (one-hot encoding value) of pixel i belonging to category c

[0024] The embedding loss is composed of variance loss and distance loss, ensuring that the distance between pixel vectors in the same area is small, for example, less than 1.5, while the distance between pixel vectors in different temperature areas is large, for example, greater than 1.5: A distance less than this range indicates that the pixels belong to the same area, and the formula is:

[0025] Loss embed =variance loss+distance loss (3)

[0026] in,

[0027]

[0028] Where C is the number of spontaneous combustion areas, n i is the number of pixels belonging to the same spontaneous combustion area, m i is the mean vector of the spontaneous combustion region, e j is a pixel vector, m is a hyperparameter, m k represents the kth spontaneous combustion area in the mean vector (or called the centroid vector, cluster center).

[0029] Furthermore, the clustering algorithm described in step 4.3 refers to clustering the embedded vectors into different spontaneous combustion regions. In the clustering process, all pixels in the circle are selected as the same spontaneous combustion region with the cluster center as the center and r as the radius:

[0030] The specific steps are as follows:

[0031] Step 4.3.1 Initialize each pixel as a candidate cluster center;

[0032] Step 4.3.2 calculates the density gradient direction of each candidate cluster center;

[0033] Step 4.3.3: Move the center of each candidate cluster along the density gradient until it converges to a stable point;

[0034] All pixels that converge to the same stable point are classified as the same spontaneous combustion area.

[0035] Furthermore, the temperature change modeling process described in step 4.4 is: through the T-Net model, the spontaneous combustion area obtained by FireNet is further modeled, the temperature change matrix TTT of each spontaneous combustion area is predicted, and the temperature distribution model is fitted to complete the conversion from image space to temperature space.

[0036] Furthermore, step 4.4 also includes a time series data verification step: the detection results of FireNet and T-Net are verified by combining historical data through the TimesNet model.

[0037] Furthermore, the time series data verification process described in step 4.5 is:

[0038] Real-time monitoring: The spontaneous combustion area data detected by FireNet and T-Net are input into the TimeSNet model;

[0039] Historical data comparison: The model reconfirms whether the detected spontaneous combustion area is a real spontaneous combustion event based on the time series characteristics of historical data;

[0040] Result output: Output the final spontaneous combustion confirmation result, including the location of the spontaneous combustion area and temperature change trend information.

[0041] A spontaneous combustion identification system for open-pit coal mines based on UAV thermal imaging images, the technical key points of which are that it is mainly composed of a visible light camera, an infrared camera, a UAV, a ground control center and a workstation. The visible light camera and the infrared camera are mounted on the UAV, and the UAV is connected to the ground control center for communication.

[0042] The beneficial effects of the present invention are: the method and system for identifying spontaneous combustion in open-pit coal mines based on thermal imaging images of unmanned aerial vehicles, the process of which is: planning inspection routes according to the actual terrain of the mining area; the unmanned aerial vehicle with a camera mounted on it takes off to perform tasks and inspect the coal mining area; the thermal infrared image of the monitored area is collected in a continuous frame manner, and the video data stream is transmitted back to the workstation; the spontaneous combustion analysis algorithm process is started, the image frames in the video stream are extracted and input into the image processing unit, and the coal seam spontaneous combustion is identified through the algorithm. Its beneficial effects are:

[0043] 1. Overcoming the drawbacks of traditional inspection methods

[0044] In response to the problems of low efficiency, high labor intensity, inaccuracy and potential safety hazards of manual inspections, the present invention utilizes a drone equipped with an infrared camera to collect thermal imaging images, automatically identifies spontaneous combustion phenomena based on image analysis, and realizes rapid and comprehensive spontaneous combustion monitoring of open-pit coal mines, greatly improving monitoring efficiency and reducing labor intensity and safety risks.

[0045] It solves the problem of manual dependence of traditional monitoring equipment. It does not require manual regular inspection of equipment operation status, reading and analyzing monitoring data, improves the automation of monitoring, and ensures the timeliness and accuracy of monitoring. At the same time, automated monitoring reduces human errors and makes monitoring results more reliable.

[0046] 2. Expanding the scope of monitoring and improving safety

[0047] Through the flexible flight of drones, which are not restricted by terrain conditions, they can quickly reach areas that are difficult to reach by humans, such as steep slopes and complex terrain, expanding the monitoring range and solving the problem of limited coverage of traditional monitoring equipment.

[0048] Selecting a drone platform suitable for the open-pit coal mine environment, with stable flight performance and long endurance, ensures that the inspection mission is carried out efficiently and safely, and improves the safety of open-pit coal mine monitoring. The automated monitoring system can monitor the status of the drone in real time and adjust the flight parameters in time, further ensuring flight safety.

[0049] 3. Achieve early and accurate identification and improve management level

[0050] Taking advantage of the high temperature sensitivity and large-area rapid scanning of infrared thermal imaging technology, early and accurate identification of spontaneous combustion in open-pit coal mines can be achieved based on image analysis. It can detect tiny temperature changes, which is conducive to timely discovery of problems at the early stages of spontaneous combustion risks and reducing resource waste, environmental pollution and safety accident risks caused by spontaneous combustion.

[0051] Automated monitoring can continuously and regularly monitor open-pit coal mines, detect abnormal situations in a timely manner and issue alarms, buying valuable time for taking countermeasures.

[0052] In summary, the present invention provides a more efficient, accurate and safe solution for monitoring spontaneous combustion in open-pit coal mines, and provides a strong guarantee for safe production and environmental protection in open-pit coal mines. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0054] Figure 1 The open-pit coal mine spontaneous combustion identification system equipment based on UAV thermal imaging images of the present invention is composed of;

[0055] Figure 2 This is a workflow diagram for spontaneous combustion identification in open-pit coal mines based on UAV thermal imaging images according to the present invention;

[0056] Figure 3 It is a flow chart of the open-pit coal mine spontaneous combustion analysis algorithm of the present invention. DETAILED DESCRIPTION

[0057] To make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the following is a brief description of the present invention in conjunction with the attached Figure 1-Figure 3The present invention is further described in detail with reference to the accompanying drawings and specific embodiments.

[0058] In this embodiment, an open-pit coal mine spontaneous combustion identification system based on drone thermal imaging images is mainly composed of a visible light camera, an infrared camera, a drone, a ground control center and a workstation. The visible light camera and infrared camera are mounted on the drone, and the drone is communicated with the ground control center.

[0059] Visible light cameras collect visible light images of open-pit coal mines, including the topography of the mining area, the distribution of the mining area, assist in comprehensive judgment of spontaneous combustion phenomena, and provide a basis for subsequent visualization applications. At the same time, they provide an application basis for UAV flights and subsequent visualization applications.

[0060] Infrared cameras collect thermal images of open-pit coal mines, detect infrared radiation in coal mining areas and represent temperatures in different colors or grayscales, providing a data basis for spontaneous combustion identification;

[0061] The ground control center receives the image data transmitted by the drone for preliminary storage and management, and sends flight instructions to the drone to control its flight trajectory, altitude, speed and other parameters to ensure that the inspection mission is carried out efficiently and safely, and monitor the drone status in real time.

[0062] The workstation, as the core data processing and analysis center, receives the image data transmitted by the ground control center and performs in-depth processing and analysis on the visible light image and thermal image. It performs auxiliary analysis on the visible light image and detects the temperature abnormality area through the spontaneous combustion recognition algorithm for thermal image to judge the spontaneous combustion phenomenon. Finally, the analysis results are visualized.

[0063] In this embodiment, a method for identifying spontaneous combustion in open-pit coal mines based on UAV thermal imaging images includes the following steps:

[0064] The visible light camera and infrared camera are installed on the drone. When the drone flies in the open-pit coal mining area according to the preset route, the visible light camera collects visible light images of the open-pit coal mine, and the infrared camera collects thermal imaging images. The drone is connected to the ground control center through wireless communication. The drone transmits the collected image data to the ground control center in real time and receives the flight instructions issued by the ground control center. The ground control center is connected to the workstation through wired or wireless means, and the received image data is transmitted to the ground workstation for processing and analysis. The workflow is shown in Figure 3 .

[0065] Step 1: Plan the inspection route according to the actual terrain of the mining area;

[0066] Step 2: The drone takes off with the camera mounted on it to perform a mission and inspect the coal mining area;

[0067] Step 3: Collect thermal infrared images of the monitored area in a continuous frame manner and transmit the video data stream back to the workstation;

[0068] Step 4: Start the spontaneous combustion analysis algorithm process, extract the image frames in the video stream and input them into the image processing unit, and identify the spontaneous combustion of the coal seam through the algorithm;

[0069] Step 4.1, image acquisition: The thermal imaging camera acquires images of the monitored area at a rate of multiple frames per second;

[0070] Step 4.2, image preprocessing: Use median filtering to remove image noise and improve image quality.

[0071] Step 4.3, feature extraction: complete the instance segmentation of the spontaneous combustion area through clustering. Specifically: through the FireNet model, all spontaneous combustion areas are automatically detected and segmented from the infrared image through semantic segmentation, and each area is instantiated (that is, the boundaries and pixel sets of each spontaneous combustion area are clarified) by using semantic segmentation, embedding vectorization, and clustering operations to provide input for subsequent temperature modeling.

[0072] 4.3.1 Instance segmentation of spontaneous combustion area (FireNet):

[0073] 4.3.1 Semantic segmentation branch: responsible for semantic segmentation of the input infrared image and outputting a binary classification result of whether each pixel belongs to the spontaneous combustion area; this branch adopts the U-Net architecture and adds the contextual information of the pixel in the encoder and decoder process, which can effectively capture the local and global features of the image.

[0074] 4.3.2 Embedding branch: Generate an embedding vector for each pixel for subsequent clustering operations. This branch also uses the U-Net architecture, but outputs a high-dimensional embedding vector instead of a binary classification result.

[0075] There are few spontaneous combustion samples in this paper, which is unbalanced: use the softmax_cross_entropy loss function and introduce bounded inverse class weight.

[0076] The formula is as follows:

[0077]

[0078] Among them, w i =c·(1-p i ), p i (2)

[0079] Where p is the probability of category (i) appearing in the overall sample, and c is a hyperparameter (usually set to 1.02).

[0080] Embedding loss: It consists of variance loss and distance loss, which ensures that the distance between pixel vectors in uniform areas (such as areas with the same temperature) is small, and the distance between pixel vectors in areas with different temperatures is large.

[0081] Loss embed =variance loss+distance loss (3)

[0082] in,

[0083]

[0084] Where C is the number of spontaneous combustion areas, n i is the number of pixels belonging to the same spontaneous combustion area, m i is the mean vector of the spontaneous combustion region, e j is a pixel vector, j is a pixel index, m is a hyperparameter, and m k is the center vector (mean vector) of the kth region.

[0085] Step 4.3.3 uses the Mean-Shift clustering algorithm to cluster the embedded vectors into different spontaneous combustion regions. In the clustering process, the cluster center is used as the circle center and (r) is the radius. All pixels in the circle are selected and classified as the same spontaneous combustion region.

[0086] Mean-Shift clustering: Mean-Shift is an unsupervised clustering algorithm based on kernel density estimation, which can automatically discover patterns in data without prior knowledge. The specific steps are as follows:

[0087] Step 4.3.3.1 Initialize each pixel as a candidate cluster center.

[0088] Step 4.3.3.2 calculates the density gradient direction of each candidate cluster center.

[0089] Step 4.3.3.3 moves each candidate cluster center along the density gradient direction until it converges to a stable point.

[0090] All pixels that converge to the same stable point are classified as the same spontaneous combustion area.

[0091] Step 4.4, convert the spontaneous combustion area from the image space to the temperature distribution space. Specifically: through the T-Net model, further model the spontaneous combustion area obtained by FireNet, predict the temperature change matrix TTT of each spontaneous combustion area, and fit the temperature distribution model to complete the conversion from image space to temperature space.

[0092] Temperature change modeling (T-Net) is used to convert the spontaneous combustion area from image space to temperature distribution space, so as to model the spontaneous combustion area more accurately. The continuity of natural areas is mainly taken into account, and the natural areas are predicted and regressed through continuity.

[0093] Step 4.4.1 Network structure

[0094] T-Net consists of 6 layers of ordinary convolutional networks and one layer of fully connected network, outputting a 6-dimensional vector representing the temperature change matrix (T).

[0095] Convolutional layers: The first five layers are convolutional layers, each containing 64 3x3 convolution kernels, using the ReLU activation function, with a step size of 1 and padding of 1 to keep the input and output sizes consistent.

[0096] Fully connected layer: The last layer is a fully connected layer, which outputs a 6-dimensional vector representing the temperature change matrix (T).

[0097] Constraints: Constrain the transposed matrix by setting it to 0 to ensure that the temperature change remains consistent under transformation, that is, the horizontal line remains horizontal under transformation.

[0098] Step 4.4.2 Temperature modeling:

[0099] Use the temperature change matrix (T) predicted by T-Net to remove the brackets and use the brackets to transform the temperature of the pixels in the spontaneous combustion area.

[0100] Coordinate transformation: The pixel coordinates ((x, y)) of the spontaneous combustion area are transformed into the new coordinate system through the matrix (T).

[0101] T′=T·T (6)

[0102] Temperature fitting: The least squares method is used to fit the parameters of the third-order polynomial to predict the accurate temperature distribution of the spontaneous combustion area.

[0103] T=A·y 3 +B·y 2 +C·y+D (7)

[0104] Among them, T and y are the transformed temperature and coordinate respectively, and A, B, C, and D are fitting parameters.

[0105] Step 4.4.3 Loss Function

[0106] MSE loss: The mean squared error (MSE) loss function is used to optimize the prediction accuracy of T-Net.

[0107]

[0108] Where N is the number of samples, Tpred is the predicted temperature change matrix, T true is the true label.

[0109] Step 4.5, risk assessment: The extracted features are input into the anomaly detection module for analysis in combination with historical data to determine whether there is a risk of spontaneous combustion. Is it right to put it here? This is the algorithm result output module. Risk assessment can be placed on the application side. Here we can briefly mention the time series data verification: Through the TimesNet model, the detection results of FireNet and T-Net are verified in combination with historical data to further improve the accuracy and reliability of detection and avoid false positives or negatives.

[0110] 4.5.1 Timing Analysis Algorithm

[0111] 4.5.2 Data preprocessing:

[0112] Normalization: Normalize the temperature data to the range of [0,1] for model training.

[0113] Sliding window: The time series data is divided into sliding windows of fixed length, each of which contains data of multiple time steps.

[0114] 4.5.3 Feature Engineering:

[0115] Temperature Trend: Calculate the temperature trend within each window, such as the linear regression slope.

[0116] Temperature Peak: Extracts the maximum temperature value within each window.

[0117] Temperature gradient: Calculate the temperature gradient within each window, that is, the temperature difference between adjacent time points.

[0118] 4.5.4 Model Architecture:

[0119] Input layer: accepts multiple time-step data in a sliding window.

[0120] Encoder: Use LSTM (Long Short-Term Memory Network) to encode time series data and extract time-dependent features.

[0121] Attention mechanism: Introduce the attention mechanism to highlight the features of important time steps.

[0122] Decoder: Uses fully connected layers to map the features output by the encoder to predictions.

[0123] Output layer: Output the probability of spontaneous combustion in each window.

[0124] 4.5.5 Loss Function:

[0125] Binary cross entropy loss: used to optimize the classification performance of the model.

[0126] LossTimeSNet=-∑i[yilog(y i )+(1-yi)log(1-y i )] (12)

[0127] Among them, yi is the true label (0 or 1) of the i-th window, y i is the probability of spontaneous combustion predicted by the model.

[0128] 4.5.6 Spontaneous combustion confirmation.

[0129] Real-time monitoring: The spontaneous combustion area data detected by FireNet and T-Net are input into the TimeSNet model.

[0130] Historical data comparison: The model reconfirms whether the detected spontaneous combustion area is a real spontaneous combustion event based on the time series characteristics of historical data.

[0131] Result output: Output the final spontaneous combustion confirmation result, including the location of the spontaneous combustion area, temperature change trend and other information.

[0132] Step 4.6, Alarm Triggering: Once a risk is detected, the alarm system is immediately activated and a notification is sent.

[0133] Step 4.7, data storage and analysis: Store the images and analysis results in a database for subsequent analysis.

[0134] Step 5: Store the algorithm output results to the workstation for subsequent visualization applications, and determine whether to trigger the alarm system based on the results.

[0135] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

Claims

1. A method for identifying spontaneous combustion in open-pit coal mines based on UAV thermal imaging images, characterized in that: The following steps are involved: Step 1: Plan the inspection route according to the actual terrain of the mining area; Step 2: The drone takes off with the camera mounted on it to perform a mission and inspect the coal mining area; Step 3: Collect thermal infrared images of the monitored area in a continuous frame manner and transmit the video data stream back to the workstation; Step 4: Start the spontaneous combustion analysis algorithm process, extract the image frames in the video stream and input them into the image processing unit, and identify the spontaneous combustion of the coal seam through the algorithm; Step 5: Store the algorithm output results to the workstation for subsequent visualization applications, and determine whether to trigger the alarm system based on the results.

2. The method for identifying spontaneous combustion in open-pit coal mines based on UAV thermal imaging images as claimed in claim 1, characterized in that: The auto-ignition analysis algorithm described in step 4 is as follows: Step 4.1, image acquisition: The thermal imaging camera acquires images of the monitored area at a rate of multiple frames per second; Step 4.2, image preprocessing: use median filtering to remove image noise and improve image quality; Step 4.3, feature extraction: complete instance segmentation of spontaneous combustion area through clustering; Step 4.4, temperature change modeling: transform the self-ignition area from image space to temperature distribution space.

3. The method for identifying spontaneous combustion in open-pit coal mines based on UAV thermal imaging images as claimed in claim 2, characterized in that: The steps for completing the instance segmentation of the spontaneous combustion area through clustering described in step 4.3 are: through the FireNet model, all spontaneous combustion areas are automatically detected and segmented from the infrared image through semantic segmentation, and semantic segmentation, embedded vectorization, and clustering algorithms are used to instantiate each area, that is, to clarify the boundaries and pixel sets of each spontaneous combustion area, providing input for subsequent temperature modeling.

4. The method for identifying spontaneous combustion in open-pit coal mines based on UAV thermal imaging images as claimed in claim 3, characterized in that: The semantic segmentation described in step 4.3 refers to: The input infrared image is semantically segmented and a binary classification result of whether each pixel belongs to the spontaneous combustion area is output; the U-Net architecture is used to capture the local and global features of the image by adding the contextual information of the pixel in the encoder and decoder process.

5. The method for identifying spontaneous combustion in open-pit coal mines based on UAV thermal imaging images as claimed in claim 3, characterized in that: The embedding vectorization described in step 4.3 means: generating an embedding vector for each pixel for subsequent clustering operations, using the U-Net architecture, and outputting a high-dimensional embedding vector; in the case of a small number of spontaneous combustion samples and imbalance: using the softmax_cross_entropy loss function and introducing bounded inverse class weight, The formula is as follows: Among them, w i =c·(1-p i ), p i (2) In the formula, p i is the probability of whether category i is in the spontaneous combustion area in the overall sample, c is a hyperparameter, usually set to 1.02; y is the actual spontaneous combustion, spontaneous combustion represents 1, and no spontaneous combustion represents 0; The embedding loss is composed of variance loss and distance loss, which ensures that the distance between pixel vectors in the same area is small, while the distance between pixel vectors in different temperature areas is large. The formula is: Loss embed =variance loss+distance loss (3) in, Where C is the number of spontaneous combustion areas, n i is the number of pixels belonging to the same spontaneous combustion area, m i is the mean vector of the spontaneous combustion region, e j is a pixel vector, m is a hyperparameter, and mk represents the mean vector of the kth spontaneous combustion area.

6. The method for identifying spontaneous combustion in open-pit coal mines based on UAV thermal imaging images as claimed in claim 3, characterized in that: The clustering algorithm described in step 4.3 is to cluster the embedded vectors into different spontaneous combustion areas. In the clustering process, the cluster center is taken as the circle center and r is the radius. All pixels in the circle are selected to be classified as the same spontaneous combustion area: The specific steps are as follows: Step 4.3.1 Initialize each pixel as a candidate cluster center; Step 4.3.2 calculates the density gradient direction of each candidate cluster center; Step 4.3.3: Move the center of each candidate cluster along the density gradient until it converges to a stable point; All pixels that converge to the same stable point are classified as the same spontaneous combustion area.

7. The method for identifying spontaneous combustion in open-pit coal mines based on UAV thermal imaging images as claimed in claim 3, characterized in that: The temperature change modeling process described in step 4.4 is: through the T-Net model, the spontaneous combustion area obtained by FireNet is further modeled, the temperature change matrix TTT of each spontaneous combustion area is predicted, and the temperature distribution model is fitted to complete the conversion from image space to temperature space.

8. The method for identifying spontaneous combustion in open-pit coal mines based on UAV thermal imaging images as claimed in claim 3, characterized in that: Step 4.4 also includes the step of time series data verification: the detection results of FireNet and T-Net are verified by combining historical data through the TimesNet model.

9. The method for identifying spontaneous combustion in open-pit coal mines based on UAV thermal imaging images as claimed in claim 8, characterized in that: The time series data verification process described in step 4.5 is: Real-time monitoring: The spontaneous combustion area data detected by FireNet and T-Net are input into the TimeSNet model; Historical data comparison: The model reconfirms whether the detected spontaneous combustion area is a real spontaneous combustion event based on the time series characteristics of historical data; Result output: Output the final spontaneous combustion confirmation result, including the location of the spontaneous combustion area and temperature change trend information.

10. An open-pit coal mine spontaneous combustion identification system based on UAV thermal imaging images, characterized in that: It is mainly composed of a visible light camera, an infrared camera, an unmanned aerial vehicle, a ground control center and a workstation. The visible light camera and the infrared camera are mounted on the unmanned aerial vehicle, and the unmanned aerial vehicle is connected to the ground control center for communication.

Images