Mountain fire detection method and device based on satellite processing and electronic equipment

Abstract

The embodiment of the invention discloses a mountain fire detection method and device based on satellite processing and electronic equipment, and can solve the problems of how to improve the mountain fire recognition precision and meet the real-time requirement of power grid mountain fire monitoring. The method comprises the following steps: acquiring satellite remote sensing image data; detecting the satellite remote sensing image data through a preset target detection model in the satellite platform to obtain initial forest fire data included in the satellite remote sensing image data, the initial forest fire data being used for indicating a suspected forest fire-containing area determined through the target detection model; multi-dimensional feature extraction is carried out on the initial mountain fire data to obtain target features, and the target features comprise at least one of color features, texture features and shape features; the target features are detected through a preset decision tree classification model in the satellite platform, target forest fire data are determined, and the target forest fire data are used for indicating the forest fire-containing area determined through the decision tree classification model.

Description

Technical Field

[0001] The embodiments of the present application relate to the technical field of wildfire detection, and in particular, to a wildfire detection method, device and electronic device based on on-board processing. Background Art

[0002] Transmission lines usually pass through mountainous areas, forests and other areas prone to fire. Wildfires pose a serious threat to the safe operation of transmission lines. Through on-board processing, real-time monitoring and rapid response to fires can be achieved, data transmission delay can be reduced, and data security can be improved. However, at present, the on-board wildfire recognition technology has problems such as high consumption of computing resources, insufficient real-time performance in ground applications, and low efficiency, high cost and poor timeliness of traditional satellite remote sensing wildfire monitoring methods. Therefore, how to improve the wildfire recognition accuracy and meet the real-time requirements of power grid wildfire monitoring has become an urgent problem to be solved. Summary of the Invention

[0003] Based on this, it is necessary to provide a wildfire detection method, device and electronic device based on on-board processing for the above technical problems.

[0004] In a first aspect, the embodiments of the present application provide a wildfire detection method based on on-board processing, which is applied to a satellite platform. The wildfire detection method based on on-board processing includes: obtaining satellite remote sensing image data;

[0005] Detecting the satellite remote sensing image data through a preset target detection model in the satellite platform to obtain initial wildfire data included in the satellite remote sensing image data, where the initial wildfire data is used to indicate a region suspected of containing a wildfire determined by the target detection model;

[0006] Performing multi-dimensional feature extraction on the initial wildfire data to obtain target features, where the target features include at least one of the following: color feature, texture feature and shape feature;

[0007] Detecting the target features through a preset decision tree classification model in the satellite platform to determine target wildfire data, where the target wildfire data is used to indicate a region containing a wildfire determined by the decision tree classification model.

[0008] As an optional implementation manner, in the first aspect of the embodiments of the present application, the method further includes:

[0009] Obtaining a plurality of multi-spectral remote sensing image sample data;

[0010] Labeling the plurality of multi-spectral remote sensing image sample data respectively to obtain label data corresponding to each multi-spectral remote sensing image sample data, where the label data at least includes: wildfire label, background label, fire point label;

[0011] Train the initial detection model in the satellite platform using the multiple multispectral remote sensing image sample data and the label data corresponding to each multispectral remote sensing image sample data to obtain the target detection model.

[0012] As an alternative implementation, in the first aspect of the embodiments of the present application, after annotating the multiple multispectral remote sensing image sample data respectively to obtain the label data corresponding to each multispectral remote sensing image sample data, the method further includes:

[0013] Select wildfire sample data from the multiple multispectral remote sensing image sample data, where the wildfire sample data is the data indicating the existence of wildfire indicated by the wildfire label;

[0014] Extract features from the wildfire sample data to obtain sample features;

[0015] Train the initial decision tree model in the satellite platform using the sample features and the label data corresponding to the wildfire sample data to obtain the decision tree classification model.

[0016] As an alternative implementation, in the first aspect of the embodiments of the present application, the method further includes:

[0017] Divide the multiple multispectral remote sensing image sample data into a training data set, a validation data set, and a test data set;

[0018] After obtaining the decision tree classification model, the method further includes:

[0019] Validate the decision tree classification model using the validation data set and optimize the decision tree classification model based on the validation result to obtain an optimized decision tree classification model;

[0020] Test the optimized decision tree classification model using the test data set to obtain a test result, and evaluate the optimized decision tree classification model based on the test result.

[0021] As an alternative implementation, in the first aspect of the embodiments of the present application, the obtaining of the satellite remote sensing image data includes:

[0022] Obtain initial satellite data;

[0023] Preprocess the initial satellite data to obtain the satellite remote sensing image data;

[0024] Wherein, the preprocessing at least includes: radiometric correction, geometric correction, and cloud removal.

[0025] As an alternative implementation, in the first aspect of the embodiments of the present application, detecting the satellite remote sensing image data through a preset target detection model in the satellite platform to obtain initial wildfire data included in the satellite remote sensing image data includes:

[0026] Detecting the satellite remote sensing image data through the target detection model to obtain the fire point confidence corresponding to the satellite remote sensing image data;

[0027] Filtering the satellite remote sensing image data according to a preset confidence threshold to obtain the initial wildfire data, and the fire point confidence corresponding to the initial wildfire data is greater than the preset confidence threshold.

[0028] As an alternative implementation, in the first aspect of the embodiments of the present application, detecting the target feature through a preset decision tree classification model in the satellite platform to determine target wildfire data includes:

[0029] Classifying the target feature through the decision tree classification model to obtain the classification probability corresponding to the target feature;

[0030] Filtering the target feature according to a preset classification threshold to obtain the target wildfire data, and the classification probability corresponding to the target wildfire data is greater than the preset classification threshold.

[0031] In the second aspect, the embodiments of the present application provide a wildfire detection device in the satellite platform, which is applied to the satellite platform. The device includes: an acquisition module for acquiring satellite remote sensing image data;

[0032] A processing module for detecting the satellite remote sensing image data through a preset target detection model in the satellite platform to obtain initial wildfire data included in the satellite remote sensing image data, and the initial wildfire data is used to indicate an area suspected of containing wildfires determined by the target detection model;

[0033] The processing module is further configured to perform multi-dimensional feature extraction on the initial wildfire data to obtain target features, and the target features include at least one of the following: color feature, texture feature, and shape feature;

[0034] The processing module is further configured to detect the target feature through a preset decision tree classification model in the satellite platform to determine target wildfire data, and the target wildfire data is used to indicate an area containing wildfires determined by the decision tree classification model.

[0035] As an alternative implementation, in the second aspect of the embodiments of the present application, the acquisition module is further configured to acquire a plurality of multi-spectral remote sensing image sample data;

[0036] The processing module is further configured to separately label multiple multispectral remote sensing image sample data to obtain label data corresponding to each multispectral remote sensing image sample data, where the label data at least includes: wildfire label, background label, and fire point label;

[0037] The processing module is further configured to train an initial detection model in the satellite platform through the multiple multispectral remote sensing image sample data and the label data corresponding to each multispectral remote sensing image sample data to obtain a target detection model.

[0038] As an optional implementation manner, in the second aspect of the embodiments of the present application, the processing module is further configured to select wildfire sample data from the multiple multispectral remote sensing image sample data, where the wildfire sample data is data indicating the existence of a wildfire indicated by the wildfire label;

[0039] The processing module is further configured to extract features from the wildfire sample data to obtain sample features;

[0040] The processing module is further configured to train an initial decision tree model in the satellite platform through the sample features and the label data corresponding to the wildfire sample data to obtain a decision tree classification model.

[0041] As an optional implementation manner, in the second aspect of the embodiments of the present application, the processing module is further configured to divide the multiple multispectral remote sensing image sample data into a training data set, a validation data set, and a test data set;

[0042] The processing module is further configured to verify the decision tree classification model through the validation data set and optimize the decision tree classification model according to the verification result to obtain an optimized decision tree classification model;

[0043] The processing module is further configured to test the optimized decision tree classification model through the test data set to obtain a test result, and evaluate the optimized decision tree classification model according to the test result.

[0044] As an optional implementation manner, in the second aspect of the embodiments of the present application, the acquisition module is specifically configured to acquire initial satellite data;

[0045] The processing module is specifically configured to preprocess the initial satellite data to obtain satellite remote sensing image data;

[0046] Wherein, the preprocessing at least includes: radiometric correction, geometric correction, and cloud removal.

[0047] As an alternative implementation, in the second aspect of the embodiments of the present application, the processing module is specifically configured to detect the satellite remote sensing image data through a target detection model to obtain the fire point confidence corresponding to the satellite remote sensing image data;

[0048] The processing module is specifically configured to screen the satellite remote sensing image data according to a preset confidence threshold to obtain initial wildfire data, and the fire point confidence corresponding to the initial wildfire data is greater than the preset confidence threshold.

[0049] As an alternative implementation, in the second aspect of the embodiments of the present application, the processing module is specifically configured to classify the target features through a decision tree classification model to obtain the classification probability corresponding to the target features;

[0050] The processing module is specifically configured to screen the target features according to a preset classification threshold to obtain target wildfire data, and the classification probability corresponding to the target wildfire data is greater than the preset classification threshold.

[0051] In a third aspect, an embodiment of the present application provides an electronic device, which includes:

[0052] A memory storing executable program code;

[0053] A processor coupled to the memory;

[0054] The processor calls the executable program code stored in the memory and executes the wildfire detection method based on on-board processing in the first aspect of the embodiments of the present application.

[0055] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program, and the computer program enables a computer to execute the wildfire detection method based on on-board processing in the first aspect of the embodiments of the present application. The computer-readable storage medium includes ROM / RAM, a magnetic disk, an optical disc, etc.

[0056] In a fifth aspect, an embodiment of the present application provides a computer program product, which, when running on a computer, enables the computer to execute some or all of the steps of any one of the methods in the first aspect.

[0057] In a sixth aspect, an embodiment of the present application provides an application publishing platform, which is used to publish a computer program product, wherein, when the computer program product runs on a computer, it enables the computer to execute some or all of the steps of any one of the methods in the first aspect.

[0058] Compared with the prior art, the embodiments of the present application have the following beneficial effects:

[0059] The embodiments of the present application provide a wildfire detection method, device and electronic device based on on-board processing, which acquire satellite remote sensing image data; detect the satellite remote sensing image data through a preset target detection model in the satellite platform to obtain initial wildfire data included in the satellite remote sensing image data, and the initial wildfire data is used to indicate the area suspected of containing wildfires determined by the target detection model; extract multi-dimensional features from the initial wildfire data to obtain target features, and the target features include at least one of the following: color feature, texture feature and shape feature; detect the target features through a preset decision tree classification model in the satellite platform to determine target wildfire data, and the target wildfire data is used to indicate the area containing wildfires determined by the decision tree classification model. In this solution, the target detection algorithm is used to quickly identify the suspected wildfire areas in the remote sensing images, reduce the computational complexity and improve the processing speed; then, combined with multi-dimensional feature extraction and the decision tree classification model, the suspected wildfire areas are accurately identified, and the above operations are all integrated into the satellite platform to be able to identify wildfires in real time and accurately on the satellite, reduce the consumption of computing resources, meet the high-precision, high-frequency and refined requirements of the power grid for wildfire identification, effectively reduce the false alarm rate, improve the identification accuracy, reduce the data transmission delay, achieve a rapid response to wildfires, meet the timeliness requirements of the power grid for wildfire monitoring, effectively improve the efficiency and accuracy of the power grid wildfire monitoring, and provide an important technical support for ensuring the safe operation of power lines. Description of the Drawings

[0060] The drawings here are incorporated into the specification and form a part of this specification, showing the embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0061] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required to be used in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0062] Figure 1 is a flowchart of a wildfire detection method based on on-board processing provided by an embodiment of the present application Figure 1 ;

[0063] Figure 2 is a flowchart of a wildfire detection method based on on-board processing provided by an embodiment of the present application Figure 2 ;

[0064] Figure 3 is a system architecture diagram of a wildfire detection method based on on-board processing provided by an embodiment of the present application;

[0065] Figure 4It is a schematic structural diagram of a wildfire detection device based on on-orbit processing provided by an embodiment of the present application;

[0066] Figure 5 It is a schematic structural diagram of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0067] In order to more clearly understand the above objects, features and advantages of the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of the present application. It should be noted that, without conflict, the embodiments of the present application and the features in the embodiments may be combined with each other. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present application without creative efforts shall fall within the protection scope of the present application.

[0068] The terms "first" and "second" etc. in the specification and claims of the present application are used to distinguish different objects, rather than to describe a specific order of the objects.

[0069] The terms "including" and "having" and any variations thereof in the embodiments of the present application are intended to cover non-exclusive inclusion. For example, a process, method, system, product or device including a series of steps or units does not necessarily have to be limited to those steps or units clearly listed, but may include other steps or units not clearly listed or inherent to these processes, methods, products or devices.

[0070] It should be noted that in the embodiments of the present application, words such as "exemplary" or "for example" are used to represent examples, illustrations or explanations. Any embodiment or design solution described as "exemplary" or "for example" in the embodiments of the present application should not be construed as being more preferred or having more advantages than other embodiments or design solutions. Rather, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific manner.

[0071] Transmission lines usually pass through mountainous areas, forests and other areas prone to fires, and mountain fires pose a serious threat to the safe operation of transmission lines. To ensure the safe and stable operation of the power system, the power grid has put forward higher requirements for mountain fire recognition: (1) High precision: It is required to accurately identify mountain fires and precisely locate the fire points to avoid false alarms and missed alarms. (2) High frequency: It is required to monitor the fire situation frequently and detect the fire in time so as to take corresponding measures in time. (3) Refinement: It is required to identify information such as the type of mountain fire, small-scale near the power grid, and spreading trend, etc., to provide more refined data support for fire fighting and post-disaster assessment. To meet the high-precision, high-frequency, and refined requirements of the power grid for mountain fire recognition, a deep learning model is deployed on the satellite, and real-time monitoring and rapid response to fires can be achieved through on-board processing, reducing data transmission delay and improving data security.

[0072] However, on-board processing also faces some challenges: (1) Limited computing resources: The computing power and storage space of the satellite platform are limited, far less than that of ground servers. (2) Power consumption limitation: The energy supply of the satellite platform is limited, and the power consumption of the algorithm needs to be controlled. (3) Real-time requirement: Mountain fire monitoring needs to detect the fire in time, and there are high requirements for the processing speed of the algorithm. (4) Data transmission limitation: The communication bandwidth between the satellite and the ground is limited, and all original data cannot be transmitted to the ground for processing.

[0073] To solve some or all of the above technical problems, the embodiments of the present application provide a mountain fire detection method, device and electronic device based on on-board processing, which obtain satellite remote sensing image data; detect the satellite remote sensing image data through a preset target detection model in the satellite platform to obtain initial mountain fire data included in the satellite remote sensing image data, and the initial mountain fire data is used to indicate the area suspected of containing a mountain fire determined by the target detection model; perform multi-dimensional feature extraction on the initial mountain fire data to obtain target features, and the target features include at least one of the following: color feature, texture feature and shape feature; detect the target features through a preset decision tree classification model in the satellite platform to determine target mountain fire data, and the target mountain fire data is used to indicate the area containing a mountain fire determined by the decision tree classification model. In this solution, the target detection algorithm is used to quickly identify the suspected mountain fire area in the remote sensing image, reduce the computational complexity, and improve the processing speed; then, combined with multi-dimensional feature extraction and the decision tree classification model, the suspected mountain fire area is accurately identified. Integrating the above operations into the satellite platform can identify mountain fires on the satellite in real time and accurately, reduce the consumption of computing resources, meet the high-precision, high-frequency, and refined requirements of the power grid for mountain fire recognition, effectively reduce the false alarm rate, improve the recognition accuracy, reduce the data transmission delay, achieve a rapid response to mountain fires, meet the timeliness requirements of the power grid for mountain fire monitoring, effectively improve the efficiency and accuracy of the power grid mountain fire monitoring, and provide important technical support for ensuring the safe operation of power lines.

[0074] like Figure 1 As shown, Figure 1 A flowchart of a mountain fire detection method based on on-board processing provided in an embodiment of the present application, the method may include the following steps:

[0075] 101. Obtain satellite remote sensing image data.

[0076] In an embodiment of the present application, the satellite remote sensing image data may be data sent by a received satellite sensor. Specifically, a satellite infrared payload may be connected to receive multispectral image data.

[0077] 102. The satellite remote sensing image data is detected by using a target detection model preset in the satellite platform to obtain initial wildfire data included in the satellite remote sensing image data.

[0078] In an embodiment of the present application, the initial wildfire data can be used to indicate areas suspected of containing wildfires as determined by a target detection model. Since satellite remote sensing image data may include data from a larger area, an initial judgment can be made first to initially identify areas suspected of containing wildfires, and then classification detection can be performed again on these areas suspected of containing wildfires.

[0079] It should be noted that the preset target detection model in the satellite platform can be a YOLOv5 model, which can be deployed on a graphics processing unit (GPU) to achieve accelerated processing. The YOLOv5 model has been trained with ground wildfire sample data and can identify areas in the image that are suspected of containing wildfires.

[0080] In some embodiments, the model structure of YOLOv5 can be divided into four main parts: input, Backbone, Neck and Prediction. Among them, the input uses Mosaic data enhancement technology, which enriches the detection data set and improves the robustness of the network by randomly using 4 pictures for scaling, cropping and splicing, and realizes adaptive anchor frame calculation and adaptive picture scaling to adapt to picture inputs of different sizes and proportions; Backbone includes Focus structure, CBL structure, CSP structure and SPP structure. The Focus structure reduces the amount of calculation through slicing operations while retaining the original picture information as much as possible. The CSP (Cross Stage Partial Network) structure is used to reduce the amount of calculation and improve the speed and accuracy of reasoning. Two CSP structures are designed in YOLOv5, which are respectively applied to the Backbone backbone network and Neck. The SPP (Spatial PyramidPooling) module increases the reception range of the backbone features through the maximum pooling operations of different scales, and significantly separates the most important context features.

[0081] Neck adopts the structures of FPN (Feature Pyramid Network) and PAN (Pyramid Attention Network). FPN conveys strong semantic features from top to bottom, and PAN conveys strong localization features from bottom to top. The combination of the two maximally preserves the location information and category information of the target. Prediction adopts CIOU_Loss as the loss function for the BoundingBox to improve the accuracy of object detection.

[0082] The YOLOv5 model has high precision in object detection tasks, can achieve real-time processing on the CPU, and can obtain higher performance when deployed on the GPU. It can also process multiple tasks simultaneously, including object detection, face detection, pose estimation, etc., and has good versatility and flexibility.

[0083] In the embodiment of this application, satellite remote sensing image data can be input into the YOLOv5 model, and the satellite remote sensing image data can be detected by the YOLOv5 model to obtain the initial wildfire data included in the satellite remote sensing image data.

[0084] 103. Extract multi-dimensional features from the initial wildfire data to obtain target features.

[0085] In the embodiment of this application, after obtaining the initial wildfire data, it is necessary to further detect whether there is a wildfire. Therefore, multi-dimensional features can be extracted for subsequent classification and recognition. The target features can specifically include at least one of the following: color features, texture features, and shape features.

[0086] In some embodiments, the color features can specifically include: infrared band reflectivity, near-infrared band reflectivity, and the radiation brightness difference between the fire point pixels and surrounding pixels, etc.; the texture features can specifically include: gray-level co-occurrence matrix (GLCM) features and local binary pattern (LBP) features; the shape features can specifically include: the area, perimeter, and shape factor of the suspected wildfire area, solar flare features, high-temperature surface features, industrial heat source features, etc.

[0087] 104. Detect the target features through a preset decision tree classification model in the satellite platform to determine the target wildfire data.

[0088] In the embodiment of this application, after extracting the target features, classification and recognition can be performed through a preset decision tree classification model in the satellite platform to determine the target wildfire data from the initial wildfire data. The target wildfire data can be used to indicate the area containing wildfires determined by the decision tree classification model. It can be understood that it further screens out the areas where wildfires are determined to exist from the areas suspected of having wildfires.

[0089] It should be noted that the decision tree classification model in the satellite platform is a commonly used supervised learning algorithm, which makes classification decisions by constructing a tree-like structure. The decision tree classification model consists of nodes, branches, and decision paths. Among them, the nodes include: Root Node: The topmost node of the tree, which contains the entire dataset; Internal Node: The node used to test a certain attribute (feature), and the data is allocated to the child nodes according to the test results; Leaf Node: The node that does not contain any child nodes, indicating the final classification result; The branch is the path extending from the node, representing the result of a certain attribute test; The decision path is a path from the root node to the leaf node, indicating a complete decision-making process. The decision tree classification model can usually be used to perform classification tasks (such as email classification, disease diagnosis, etc.), feature selection (select features by evaluating the importance of features in the decision tree), rule extraction (extract classification rules from the decision tree), etc.

[0090] The embodiment of the present application provides a wildfire detection method based on on-board processing, which acquires satellite remote sensing image data; detects the satellite remote sensing image data through a preset target detection model in the satellite platform to obtain initial wildfire data included in the satellite remote sensing image data, and the initial wildfire data is used to indicate the area suspected of containing wildfires determined by the target detection model; performs multi-dimensional feature extraction on the initial wildfire data to obtain target features, and the target features include at least one of the following: color feature, texture feature, and shape feature; detects the target features through a preset decision tree classification model in the satellite platform to determine target wildfire data, and the target wildfire data is used to indicate the area containing wildfires determined by the decision tree classification model. In this solution, the target detection algorithm is used to quickly identify the suspected wildfire area in the remote sensing image, reduce the computational complexity, and improve the processing speed; then, combined with multi-dimensional feature extraction and the decision tree classification model, the suspected wildfire area is accurately identified, and the above operations are all integrated into the satellite platform to be able to identify wildfires in real time and accurately on the satellite, reduce the consumption of computing resources, meet the high-precision, high-frequency, and refined requirements of the power grid for wildfire identification, effectively reduce the false alarm rate, improve the identification accuracy, reduce the data transmission delay, achieve a rapid response to wildfires, meet the timeliness requirements of the power grid for wildfire monitoring, effectively improve the efficiency and accuracy of power grid wildfire monitoring, and provide an important technical support for ensuring the safe operation of power lines.

[0091] As Figure 2 shown, Figure 2 is a flowchart of a wildfire detection method based on on-board processing provided by the embodiment of the present application. This method may further include the following steps:

[0092] 201. Obtain multiple multispectral remote sensing image sample data.

[0093] In the embodiments of the present application, both the target detection model and the decision tree classification model need to be pre-trained with a large amount of sample data. Therefore, it is necessary to obtain multiple multispectral remote sensing image sample data, which are also data received from satellite sensors.

[0094] It should be noted that to ensure the accuracy of the target detection model and the decision tree classification model, when training the models, data in various scenarios, environments, seasons, and times need to be used for training. Therefore, the obtained multispectral remote sensing image sample data can cover wildfire and non-wildfire areas in different seasons, different times, and different geographical locations.

[0095] 202. Label each of the multiple multispectral remote sensing image sample data to obtain label data corresponding to each multispectral remote sensing image sample data.

[0096] In the embodiments of the present application, when training the model, it is necessary to learn based on the multispectral remote sensing image sample data and the corresponding wildfire detection results. That is to say, what the model learns is whether there is a wildfire in each multispectral remote sensing image sample data. Therefore, it is necessary to label the multispectral remote sensing image sample data, that is, it is necessary to label the label data corresponding to each multispectral remote sensing image sample data for the model to learn.

[0097] It should be noted that the label data can at least include: wildfire label, background label, and fire point label. The wildfire label is: presence of wildfire and absence of wildfire. The background label can be used to indicate the background description information of the current data, such as: forest, plain, lake, mountain, etc. The fire point label marks the specific location points where wildfires exist.

[0098] It should be noted that the labeling can be done manually. That is, manually label the wildfire areas, backgrounds, and fire points in each multispectral remote sensing image sample data. For example, use a labeling tool to frame the wildfire areas and mark the category as "wildfire". At the same time, label the samples of the background areas, such as forests, grasslands, water bodies, etc. and suspected fire points, such as photovoltaic panels, factory heat sources, etc., and mark the corresponding categories.

[0099] 203. Use the multiple multispectral remote sensing image sample data and the label data corresponding to each multispectral remote sensing image sample data to train the initial detection model in the satellite platform to obtain the target detection model.

[0100] In the embodiments of the present application, since the target detection model only performs preliminary detection to detect areas where wildfires are suspected to exist, during model training, the model can be directly trained using multiple multispectral remote sensing image sample data and the label data corresponding to each multispectral remote sensing image sample data. Specifically, multiple multispectral remote sensing image sample data are used as input items, and the label data corresponding to each multispectral remote sensing image sample data are used as output items to train the initial detection model to obtain the target detection model.

[0101] In some embodiments, during the process of model training, the model also needs to be verified and tested. Therefore, multiple multispectral remote sensing image sample data can be divided into a training data set, a validation data set, and a test data set; after obtaining the target detection model, the target detection model is verified using the validation data set, and the target detection model is optimized based on the verification results to obtain an optimized target detection model; the optimized target detection model is tested using the test data set to obtain test results, and the optimized target detection model is evaluated based on the test results.

[0102] It should be noted that when optimizing the target detection model based on the verification results, specifically, the target detection model is optimized using a loss function, and the optimized model is verified again. The loop iterates until the model converges to obtain the optimized target detection model. During testing, the test results can be compared with the standard results to evaluate the generalization ability and recognition accuracy of the model.

[0103] Exemplarily, during the training process of the target detection model, the training data set: contains 10,000 labeled multispectral remote sensing images, including wildfire and non-wildfire area samples. Model parameters: The input image size is 1024x1280 pixels, the pre-trained YOLOv5s model is used as the base model, the training iteration times are 100 times, and the learning rate is 0.001. Loss function: CIoU Loss is used as the loss function. Optimizer: Adam optimizer is used.

[0104] 204. Select wildfire sample data from multiple multispectral remote sensing image sample data.

[0105] In the embodiments of the present application, since the decision tree classification model selects areas where wildfires are definitely present, when selecting sample data, it is necessary to determine the sample data where wildfires are present, that is, wildfire sample data, and this wildfire sample data is the data indicated by the wildfire label as having a wildfire.

[0106] 205. Extract features from the wildfire sample data to obtain sample features.

[0107] In the embodiments of the present application, the sample features and the target features in the above embodiments are the same concept, which may specifically include color features, texture features, and shape features, and will not be elaborated here.

[0108] 206. Train the initial decision tree model in the satellite platform through the sample features and the label data corresponding to the wildfire sample data to obtain a decision tree classification model.

[0109] In the embodiments of the present application, the decision tree classification model can be understood as analyzing whether there is a wildfire through the multi-dimensional features of the data. Therefore, during training, it is necessary to learn the corresponding relationship between the sample features and the existence of wildfires. That is, train the initial decision tree model through the sample features and the label data corresponding to the wildfire sample data. Specifically, use the sample features as the input item and the label data corresponding to the wildfire sample data as the output item to train the initial decision tree model to obtain a decision tree classification model.

[0110] In some embodiments, during the model training process, it is also necessary to verify and test the model. Therefore, multiple multi-spectral remote sensing image sample data can be divided into a training data set, a validation data set, and a test data set; after obtaining the decision tree classification model, verify the decision tree classification model through the validation data set, and optimize the decision tree classification model based on the verification results to obtain an optimized decision tree classification model; test the optimized decision tree classification model through the test data set to obtain test results, and evaluate the optimized decision tree classification model based on the test results.

[0111] It should be noted that when optimizing the decision tree classification model based on the verification results, specifically optimize the decision tree classification model through a loss function, and verify the optimized model again, and iterate until the model converges to obtain an optimized decision tree classification model. During testing, the test results can be compared with the standard results to evaluate the generalization ability and recognition accuracy of the model.

[0112] Exemplarily, during the training process of the decision tree classification model, the training data set: the multi-dimensional features of the suspected wildfire areas extracted from the YOLOv5 model training data set, and the corresponding class labels. Model parameters: Use the CART algorithm to construct a decision tree, the maximum depth of the tree is 5, and the node splitting criterion is the Gini coefficient.

[0113] In some embodiments, an appropriate decision tree algorithm can be selected, such as CART, ID3, C4.5, etc., to optimize the model parameters, such as the depth of the tree, the node splitting criterion, etc. Use the validation data set and the test data set to evaluate the performance of the decision tree classification model and make adjustments to obtain the best classification effect.

[0114] In some embodiments, after obtaining the trained object detection model and decision tree classification model through the above steps, the object detection model and decision tree classification model can be converted into formats supported by the on-board processing platform, such as the.onnx format or.tflite format. Then, techniques such as model quantization and pruning can be used to compress and optimize the model, reducing the model volume and computational amount, so that it can operate efficiently on a satellite platform with limited resources. The converted model is loaded into the memory of the on-board processing platform, and the model running environment is configured, such as setting the model input and output formats, allocating computing resources, etc.

[0115] 207. Obtain initial satellite data.

[0116] 208. Preprocess the initial satellite data to obtain satellite remote sensing image data.

[0117] In the embodiments of the present application, the preprocessing may at least include: radiometric correction, geometric correction, and cloud removal.

[0118] Among them, radiometric correction may refer to converting the radiation value received by the sensor into surface reflectance or radiance, and eliminating sensor differences and atmospheric effects. Specifically, it can be achieved by using a calibration formula, and the calibration formula is:

[0119] L1 = DN * A + B

[0120] L2 = A * (DN - SW - b1) - B

[0121] Among them, L1 is the mid-wave and long-wave radiance, DN is the image output value, A and B are calibration coefficients, L2 is the short-wave reflectance, DN is the image output value, and SW - b1 is the low-frequency noise.

[0122] Among them, geometric correction may refer to eliminating the geometric distortion of the image, such as the distortion caused by terrain undulation, and making it match the geographic coordinate system. Specifically, it can be achieved by using a geometric polynomial, and the geometric polynomial is:

[0123] x = a0 + a1X + a2Y + a3X2 + a4XY + a5Y2

[0124] y = b0 + b1X + b2Y + b3X2 + b4XY + b5Y2

[0125] Among them, (x, y) refers to the corrected image coordinates, (X, Y) refers to the original image coordinates, and a0, a1, a2, a3, a4, a5, b0, b1, b2, b3, b4, b5 are all quadratic polynomial coefficients.

[0126] Among them, cloud removal processing is an important step in remote sensing image analysis because the occlusion of clouds often leads to incomplete data, thus affecting the accuracy of the analysis results. Specific algorithms are used to identify and segment the cloud parts in remote sensing images. The cloud removal methods can specifically include: regarding the cloud area as low-frequency noise, separating high-frequency and low-frequency information in the frequency domain, enhancing the high-frequency information and suppressing the low-frequency information to remove the clouds; or, statistical interpolation methods, inferring the data of unknown points through the data of known points, which can be used to fill the areas occluded by clouds; or, dividing the image into components of different frequencies and scales, which can more effectively remove the clouds while protecting the background information; or, according to the output of the cloud detection algorithm, setting a cloud removal threshold (such as: 80%, 85%), and performing masking processing on the areas where the cloud coverage rate exceeds this threshold.

[0127] In some embodiments, for subsequent calculations, the initial satellite data can also be subjected to data format conversion to convert the original data into a format that can be processed by the system.

[0128] 209. Detect the satellite remote sensing image data through the target detection model to obtain the fire point confidence corresponding to the satellite remote sensing image data.

[0129] In the embodiments of the present application, during the process of detecting the satellite remote sensing image data through the target detection model, it is specifically determined whether there is a suspected wildfire according to the fire point confidence corresponding to the satellite remote sensing image data. Therefore, the fire point confidence corresponding to the satellite remote sensing image data can be determined first, and this fire point confidence can be learned through annotation during the model training process in advance.

[0130] 210. Screen the satellite remote sensing image data according to the preset confidence threshold to obtain the initial wildfire data.

[0131] In the embodiments of the present application, the preset confidence threshold can be a preset value. The areas where the value is less than or equal to the preset confidence threshold are considered to definitely have no wildfire and are filtered out; the areas where the value is higher than the preset confidence threshold are considered to be suspected of having a wildfire and are determined as the initial wildfire data for subsequent detection. That is to say, the fire point confidence corresponding to this initial wildfire data is greater than the preset confidence threshold.

[0132] 211. Extract multi-dimensional features from the initial wildfire data to obtain the target features.

[0133] In the embodiments of the present application, for the description of step 211, please refer to the detailed description of step 103 in the above embodiments, and the embodiments of the present application will not be repeated.

[0134] 212. Classify the target features through the decision tree classification model to obtain the classification probability corresponding to the target features.

[0135] In the embodiment of the present application, during the process of detecting target features through a decision tree classification model, specifically, it is determined whether there is a wildfire based on the classification probability of the target features. This classification probability can refer to the probability of the existence of a wildfire and the probability of the non-existence of a wildfire. Therefore, the classification probability corresponding to the target features can be determined through the decision tree classification model.

[0136] 213. Screen the target features according to a preset classification threshold to obtain target wildfire data.

[0137] In the embodiment of the present application, the preset classification threshold can be a pre-set value. Features less than or equal to the preset classification threshold are considered to have no wildfire and are filtered out; features higher than the preset classification threshold are considered to have a wildfire and are determined as target wildfire data. That is to say, the classification probability corresponding to the target wildfire data is greater than the preset classification threshold.

[0138] 214. Output the target wildfire data.

[0139] In the embodiment of the present application, after obtaining the target wildfire data, it is also necessary to output the target wildfire data.

[0140] In some embodiments, the following content can be specifically output: wildfire location, area, confidence level, longitude and latitude of the fire point, row and column numbers, DN value, confidence level, corresponding area array, etc. The above data can be packaged into a data packet, stored in the on-board memory, and sent to the ground receiving station through the communication unit (satellite data transmission link / telemetry transmission). After combining the fire point data with the ground power transmission line data, an alarm level is formed and sent to each city company.

[0141] The embodiment of the present application provides a wildfire detection method based on on-board processing, which uses a target detection algorithm to quickly identify suspected wildfire areas in remote sensing images, reduces the computational complexity, and improves the processing speed. Then, combined with multi-dimensional feature extraction and a decision tree classification model, the suspected wildfire areas are accurately identified. Integrating the above operations on the satellite platform can identify wildfires in real time and accurately on the satellite, reduce the consumption of computing resources, meet the high-precision, high-frequency, and refined requirements of the power grid for wildfire identification, effectively reduce the false alarm rate, improve the identification accuracy, reduce data transmission delay, achieve a rapid response to wildfires, meet the timeliness requirements of the power grid for wildfire monitoring, effectively improve the efficiency and accuracy of power grid wildfire monitoring, and provide an important technical support for ensuring the safe operation of power lines.

[0142] The embodiment of the present application realizes the rapid and accurate identification of wildfires in the areas surrounding the power grid through on-board real-time processing, reduces the data transmission delay and dependence on ground computing resources, improves the real-time performance and efficiency of wildfire identification, and has important application value.

[0143] In some embodiments, the wildfire detection method based on on-orbit processing provided by the embodiments of the present application can be applied to an on-orbit real-time power grid wildfire intelligent recognition system that combines YOLOv5 and a decision tree, such as Figure 3 shown. The system includes: a data acquisition module, a YOLOv5 object detection module, a feature extraction module, a decision tree classification module, a result output module, and an on-orbit processing platform.

[0144] Among them, the data acquisition module is responsible for receiving image data from the payload and performing preprocessing. The preprocessing steps include radiometric calibration, which converts the radiation value received by the sensor into surface reflectance or radiance; geometric calibration, which uses geometric polynomials to eliminate the geometric distortion of the image and make it match the geographic coordinate system; and data format conversion, which converts the original data into a format that can be processed by the system.

[0145] Among them, the YOLOv5 object detection module is responsible for quickly detecting objects in the preprocessed multi-spectral image. The YOLOv5 model is trained with ground wildfire sample data and can identify areas in the image that are suspected to contain wildfires.

[0146] Among them, the feature extraction module is responsible for extracting multi-dimensional features for subsequent classification and recognition for the areas suspected of wildfires identified by the YOLOv5 module. The multi-dimensional features extracted by this module can include at least one of the following: color features (including: brightness temperature, thermal deviation, etc.), texture features (including: local binary pattern (LBP) features, etc.), and shape features (including: the area, perimeter, shape factor, etc. of the fire point area).

[0147] Among them, the decision tree classification module is responsible for using a pre-trained decision tree classification model to classify the multi-dimensional features output by the feature extraction module and determine whether the areas suspected of wildfires are real wildfires. The decision tree model makes node divisions based on the feature values and finally classifies the suspected areas into wildfire or non-wildfire categories to identify real fire points.

[0148] Among them, the result output module is responsible for packing the longitude and latitude, row and column numbers, DN values, confidence levels, and corresponding area arrays of the real fire points into a txt file / excel and downloading them.

[0149] Among them, the on-board processing platform adopts a superheterogeneous design of FPGA-SoC + GPU-SoC. The FPGA-SoC includes an FPGA and an ARM CPU module, which are responsible for high-speed data interaction with the infrared payload, receiving instructions from the satellite overall, and performing data preprocessing, etc. The GPU-SoC includes a GPU, a CPU, a DLA, a memory, and a storage module, which are responsible for processing data and internally caching the processed data. Among them, the CPU has a data acquisition module, a feature extraction module, and a result output module; the GPU has a YOLOv5 object detection module and a decision tree classification module.

[0150] In some embodiments, a GPU-based acceleration platform is adopted as the on-board processing system, and deep learning models such as YOLO are integrated to improve the recognition accuracy and robustness. The hardware components include: an FPGA, a GPU, a processor, a memory, an image acquisition unit, a communication unit, and a power module.

[0151] Among them, the FPGA of the Xilinx ZYNQ7000 series, this module internally contains an equivalent Kintex-7 FPGA (PL) and a dual-core Cortex-A9 ARM (PS); in order to ensure the stability and reliability during the high-speed data transmission process, 1GB of DDR is added to both the PL and PS ends in the FPGA-SoC, which can be used as a high-speed data cache.

[0152] Among them, the NVIDIA Jetson Xavier NX embedded GPU platform has powerful parallel computing capabilities.

[0153] Among them, inside the GPU-SoC unit, there is a 6-core CPU with a maximum main frequency of up to 1.4 GHz. In addition, in addition to having a general GPU processor, there are also 2 deep learning accelerator (DLA) modules, which are specifically used for efficient forward inference of convolutional neural networks.

[0154] Among them, the memory includes 8GB of DDR4 memory, and each GPU-SoC is also equipped with a large-capacity 1TB SSD storage, which can achieve a data read and write rate at the GB / s level, and can support the storage and management of on-board raw image data and processed data.

[0155] Among them, the image acquisition unit can be connected to the satellite infrared payload to receive multi-spectral image data; the communication unit can be used for data communication with the satellite platform; the power module can provide a stable power supply for the system.

[0156] Such as Figure 4As shown in the figure, an embodiment of the present application provides a wildfire detection device based on on-board processing, which is applied to a satellite platform. The wildfire detection device based on on-board processing may include: an acquisition module 401, configured to acquire satellite remote sensing image data;

[0157] A processing module 402, configured to detect the satellite remote sensing image data through a preset target detection model in the satellite platform to obtain initial wildfire data included in the satellite remote sensing image data, where the initial wildfire data is used to indicate a region suspected of containing a wildfire determined by the target detection model;

[0158] The processing module 402 is further configured to perform multi-dimensional feature extraction on the initial wildfire data to obtain target features, where the target features include at least one of the following: color feature, texture feature, and shape feature;

[0159] The processing module 402 is further configured to detect the target features through a preset decision tree classification model in the satellite platform to determine target wildfire data, where the target wildfire data is used to indicate a region containing a wildfire determined by the decision tree classification model.

[0160] In some embodiments, the acquisition module 401 is further configured to acquire multiple multi-spectral remote sensing image sample data;

[0161] The processing module 402 is further configured to label each of the multiple multi-spectral remote sensing image sample data to obtain label data corresponding to each multi-spectral remote sensing image sample data, where the label data includes at least: wildfire label, background label, and fire point label;

[0162] The processing module 402 is further configured to train an initial detection model in the satellite platform through the multiple multi-spectral remote sensing image sample data and the label data corresponding to each multi-spectral remote sensing image sample data to obtain a target detection model.

[0163] In some embodiments, the processing module 402 is further configured to select wildfire sample data from the multiple multi-spectral remote sensing image sample data, where the wildfire sample data is data indicating the existence of a wildfire indicated by the wildfire label;

[0164] The processing module 402 is further configured to perform feature extraction on the wildfire sample data to obtain sample features;

[0165] The processing module 402 is further configured to train an initial decision tree model in the satellite platform through the sample features and the label data corresponding to the wildfire sample data to obtain a decision tree classification model.

[0166] In some embodiments, the processing module 402 is further configured to divide the multiple multi-spectral remote sensing image sample data into a training data set, a validation data set, and a test data set;

[0167] The processing module 402 is further configured to verify the decision tree classification model through a verification data set, and optimize the decision tree classification model according to the verification result to obtain an optimized decision tree classification model;

[0168] The processing module 402 is further configured to test the optimized decision tree classification model through a test data set to obtain a test result, and evaluate the optimized decision tree classification model according to the test result.

[0169] In some embodiments, the obtaining module 401 is specifically configured to obtain initial satellite data;

[0170] The processing module 402 is specifically configured to preprocess the initial satellite data to obtain satellite remote sensing image data;

[0171] Wherein, the preprocessing at least includes: radiometric correction, geometric correction, and cloud removal.

[0172] In some embodiments, the processing module 402 is specifically configured to detect the satellite remote sensing image data through a target detection model to obtain the fire point confidence corresponding to the satellite remote sensing image data;

[0173] The processing module 402 is specifically configured to screen the satellite remote sensing image data according to a preset confidence threshold to obtain initial wildfire data, and the fire point confidence corresponding to the initial wildfire data is greater than the preset confidence threshold.

[0174] In some embodiments, the processing module 402 is specifically configured to classify the target features through a decision tree classification model to obtain the classification probability corresponding to the target features;

[0175] The processing module 402 is specifically configured to screen the target features according to a preset classification threshold to obtain target wildfire data, and the classification probability corresponding to the target wildfire data is greater than the preset classification threshold.

[0176] In the embodiments of the present application, each module can implement the wildfire detection method provided in the above method embodiments, and can achieve the same technical effects. To avoid repetition, it will not be elaborated here.

[0177] As Figure 5 shown, the embodiments of the present application further provide an electronic device, which may include:

[0178] A memory 501 storing executable program code;

[0179] A processor 502 coupled to the memory 501;

[0180] Wherein, the processor 502 calls the executable program code stored in the memory 501 and executes the wildfire detection method based on on-board processing performed by the electronic device in the above method embodiments.

[0181] An embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements each process of the on-board processing-based wildfire detection method in the above method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be elaborated here.

[0182] The embodiment of the present application further provides a computer program product, which stores a computer program. When the computer program is executed by a processor, it implements each process of the on-board processing-based wildfire detection method in the above method embodiment, and can achieve the same technical effect. To avoid repetition, it will not be elaborated here.

[0183] Those skilled in the art should understand that the embodiments of the present application can be provided as a method, a system, or a computer program product. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media containing computer-usable program code.

[0184] In several embodiments provided by the present application, it should be understood that the disclosed device and method can also be implemented in other ways. The device embodiments described above are merely illustrative. For example, the flowcharts and block diagrams in the accompanying drawings show the possible architectures, functions, and operations of the device, method, and computer program product according to multiple embodiments of the present application. In this regard, each block in the flowchart or block diagram may represent a module, a program segment, or a part of code, and a module, a program segment, or a part of code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the blocks may occur in a different order from that marked in the accompanying drawings. For example, two consecutive blocks can actually be executed substantially in parallel, and they can sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and the combination of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system for performing the specified functions or actions, or can be implemented by a combination of dedicated hardware and computer instructions.

[0185] In this application, the processor may be a Central Processing Unit (CPU), or may also be other general-purpose processors, Digital Signal Processors (DSPs), Application Specific Integrated Circuits (ASICs), Field-Programmable Gate Arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor, or the processor may also be any conventional processor, etc.

[0186] In this application, the memory may include non-permanent memory in a computer-readable medium, in the form of random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0187] In this application, those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing relevant hardware through a program, and this program can be stored in a computer-readable storage medium. The computer-readable medium includes permanent and non-permanent, removable and non-removable storage media. The storage medium can implement information storage by any method or technology, and the information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer storage media include, but are not limited to, phase change memory (Parallel Random Access Memory, PRAM), static random access memory (Static Random Access Memory, SRAM), dynamic random access memory (Dynamic Random Access Memory, DRAM), programmable read-only memory (Programmable Read-only Memory, PROM), erasable programmable read-only memory (Erasable Programmable Read Only Memory, EPROM), other types of random access memory (Random Access Memory, RAM), read-only memory (Read-Only Memory, ROM), one-time programmable read-only memory (One-time Programmable Read-Only Memory, OTPROM), electrically-erasable programmable read-only memory (Electrically-Erasable Programmable Read-Only Memory, EEPROM), flash memory or other memory technologies, compact disc read-only memory (Compact Disc Read-Only Memory, CD-ROM), digital versatile disc (DVD) or other optical storage, magnetic cassette tapes, disk storage or other magnetic storage devices, or any other non-transmission medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media do not include transitory computer-readable media, such as modulated data signals and carrier waves.

[0188] It should be noted that, in this text, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising" or any other variant thereof is intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements not only includes those elements but also includes other elements not expressly listed, or elements inherent to such process, method, article, or device. Without further limitation, an element defined by the statement "comprising a..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0189] It should be understood that the "one embodiment" or "an embodiment" mentioned throughout the specification means that a specific feature, structure, or characteristic related to the embodiment is included in at least one embodiment of the present application. Therefore, the "in one embodiment" or "in an embodiment" that appears throughout the specification does not necessarily refer to the same embodiment. In addition, these specific features, structures, or characteristics can be combined in one or more embodiments in any suitable manner. Those skilled in the art should also be aware that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to the present application. The above-mentioned multiple embodiments are not necessarily multiple independent embodiments. Dividing them into multiple embodiments is only used to highlight different technical features in different embodiments. Those skilled in the art should be aware that the above-mentioned multiple embodiments can also be combined arbitrarily.

[0190] In various embodiments of the present application, it should be understood that the magnitude of the sequence numbers of the above processes does not necessarily mean the inevitable sequence of execution. The execution sequence of each process should be determined by its function and internal logic, and should not constitute any limitation to the implementation process of the embodiments of the present application.

[0191] The units described above as separate components may or may not be physically separated. The components shown as units may or may not be physical units, that is, they can be located in one place, or can be distributed to multiple network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0192] In addition, in each embodiment of the present application, each functional unit can be integrated in a processing unit, or each unit can exist physically alone, or two or more units can be integrated in one unit. The above-mentioned integrated units can be implemented in the form of hardware or in the form of software functional units.

[0193] When the above integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-accessible memory. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, or all or part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a memory and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc., specifically, the processor in the computer device) to execute some or all of the steps of the above method in various embodiments of the present application.

[0194] The above are only specific implementation manners of the present application, enabling those skilled in the art to understand or implement the present application. Various modifications to these embodiments will be obvious to those skilled in the art. The general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to these embodiments herein, but rather will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting wildfires based on on-board processing, characterized in that: Applied to a satellite platform, the method comprises: Acquire satellite remote sensing image data; The satellite remote sensing image data is detected by a target detection model preset in the satellite platform to obtain initial wildfire data included in the satellite remote sensing image data, wherein the initial wildfire data is used to indicate an area suspected of containing wildfires determined by the target detection model; Performing multi-dimensional feature extraction on the initial wildfire data to obtain target features, wherein the target features include at least one of the following: color features, texture features, and shape features; The target features are detected by a decision tree classification model preset in the satellite platform to determine target wildfire data, where the target wildfire data is used to indicate an area containing wildfires determined by the decision tree classification model.

2. The method according to claim 1, characterized in that The method further comprises: Acquire multiple multispectral remote sensing image sample data; The plurality of multispectral remote sensing image sample data are respectively labeled to obtain label data corresponding to each multispectral remote sensing image sample data, wherein the label data at least includes: a wildfire label, a background label, and a fire point label; The target detection model is obtained by training the initial detection model in the satellite platform using the multiple multispectral remote sensing image sample data and the label data corresponding to each multispectral remote sensing image sample data.

3. The method according to claim 2, characterized in that After the plurality of multispectral remote sensing image sample data are respectively labeled to obtain label data corresponding to each multispectral remote sensing image sample data, the method further includes: Selecting wildfire sample data from the plurality of multispectral remote sensing image sample data, the wildfire sample data being data for which the wildfire tag indicates the presence of a wildfire; Performing feature extraction on the wildfire sample data to obtain sample features; The initial decision tree model in the satellite platform is trained using the sample features and label data corresponding to the wildfire sample data to obtain the decision tree classification model.

4. The method according to claim 3, characterized in that The method further comprises: Dividing the plurality of multispectral remote sensing image sample data into a training data set, a validation data set and a test data set; After obtaining the decision tree classification model, the method further includes: Verifying the decision tree classification model through the verification data set, and optimizing the decision tree classification model through the verification result to obtain an optimized decision tree classification model; The optimized decision tree classification model is tested by using the test data set to obtain a test result, and the optimized decision tree classification model is evaluated according to the test result.

5. The method according to claim 1, characterized in that The obtaining of satellite remote sensing image data comprises: Get initial satellite data; Preprocessing the initial satellite data to obtain the satellite remote sensing image data; Wherein, the preprocessing at least includes: radiation correction, geometric correction and cloud removal.

6. The method according to claim 1, characterized in that The detecting the satellite remote sensing image data by using a target detection model preset in the satellite platform to obtain the initial wildfire data included in the satellite remote sensing image data includes: Detecting the satellite remote sensing image data by using the target detection model to obtain the fire point confidence corresponding to the satellite remote sensing image data; The satellite remote sensing image data is screened according to a preset confidence threshold to obtain the initial wildfire data, and the fire point confidence corresponding to the initial wildfire data is greater than the preset confidence threshold.

7. The method according to claim 1, characterized in that The detecting the target features by using a decision tree classification model preset in the satellite platform to determine the target wildfire data includes: Classify the target feature by using the decision tree classification model to obtain the classification probability corresponding to the target feature; The target features are screened according to a preset classification threshold to obtain the target wildfire data, and the classification probability corresponding to the target wildfire data is greater than the preset classification threshold.

8. A mountain fire detection device based on on-board processing, characterized in that: Applied to a satellite platform, the device comprises: Acquisition module, used to acquire satellite remote sensing image data; A processing module, configured to detect the satellite remote sensing image data through a target detection model preset in the satellite platform to obtain initial wildfire data included in the satellite remote sensing image data, wherein the initial wildfire data is used to indicate an area suspected of containing a wildfire determined by the target detection model; The processing module is further used to perform multi-dimensional feature extraction on the initial wildfire data to obtain target features, wherein the target features include at least one of the following: color features, texture features, and shape features; The processing module is also used to detect the target features through a decision tree classification model preset in the satellite platform to determine target wildfire data, and the target wildfire data is used to indicate an area containing wildfires determined by the decision tree classification model.

9. The device according to claim 8, characterized in that The acquisition module is also used to acquire a plurality of multispectral remote sensing image sample data; The processing module is also used to mark the plurality of multispectral remote sensing image sample data respectively to obtain label data corresponding to each multispectral remote sensing image sample data, wherein the label data at least includes: a wildfire label, a background label, and a fire point label; The processing module is also used to perform model training on the initial detection model in the satellite platform through multiple multispectral remote sensing image sample data and label data corresponding to each multispectral remote sensing image sample data to obtain a target detection model.

10. The device according to claim 9, characterized in that The processing module is further used to select wildfire sample data from the plurality of multispectral remote sensing image sample data, where the wildfire sample data is data where a wildfire tag indicates the presence of a wildfire; The processing module is also used to extract features from wildfire sample data to obtain sample features; The processing module is also used to train the initial decision tree model in the satellite platform through the sample features and the label data corresponding to the wildfire sample data to obtain a decision tree classification model.

11. The device according to claim 10, characterized in that The processing module is further used to divide the plurality of multispectral remote sensing image sample data into a training data set, a verification data set and a test data set; The processing module is also used to verify the decision tree classification model through the verification data set, and optimize the decision tree classification model through the verification result to obtain the optimized decision tree classification model; The processing module is also used to test the optimized decision tree classification model through a test data set to obtain a test result, and evaluate the optimized decision tree classification model according to the test result.

12. The device according to claim 8, characterized in that The acquisition module is specifically used to acquire initial satellite data; The processing module is specifically used to pre-process the initial satellite data to obtain satellite remote sensing image data; The preprocessing includes at least radiation correction, geometric correction and cloud removal.

13. The device according to claim 8, characterized in that The processing module is specifically used to detect the satellite remote sensing image data through the target detection model to obtain the fire point confidence corresponding to the satellite remote sensing image data; The processing module is specifically used to screen the satellite remote sensing image data according to a preset confidence threshold to obtain initial wildfire data, and the fire point confidence corresponding to the initial wildfire data is greater than the preset confidence threshold.

14. The device according to claim 8, characterized in that The processing module is specifically used to classify the target features through a decision tree classification model to obtain the classification probability corresponding to the target features; The processing module is specifically used to screen the target features according to a preset classification threshold to obtain target wildfire data, and the classification probability corresponding to the target wildfire data is greater than the preset classification threshold.

15. An electronic device, characterized in that: The electronic device comprises: A memory storing executable program code; and a processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the mountain fire detection method based on on-board processing as described in any one of claims 1 to 7.

16. A computer-readable storage medium, characterized in that: include: The computer-readable storage medium stores computer instructions, and when the computer instructions are executed by the processor, the mountain fire detection method based on on-board processing according to any one of claims 1 to 7 is implemented.

Images