High-resolution fire image prediction method and device, equipment and storage medium
Through the matching and feature extraction of low-resolution fire images and high-resolution spatial geographic images, combined with prediction models and spatial smoothing algorithms, the problem of fire image acquisition with high frequency and high-detail analysis is solved, and high-precision fire monitoring is achieved.
Patent Information
- Application Number
- CN202510413182.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-04-02
- Publication Date
- 2025-07-18
AI Technical Summary
The prior art is difficult to achieve a balance between high frequency and high-detail analysis in fire monitoring, resulting in inaccurate acquisition of high-resolution fire images.
By acquiring low-resolution fire images and high-resolution spatial geographic images, pixel-level geographic location matching is performed, the features of adjacent pixels are extracted, and the pre-trained prediction model is used to predict high-resolution fire images, combining gradient enhancement tree model and spatial smoothing algorithm to establish a nonlinear mapping relationship.
It breaks through the limitation that satellite data time-space resolution cannot be obtained at the same time, accurately and at high frequency, it can obtain high-resolution fire images, improving the accuracy and efficiency of fire monitoring.
Smart Images
Figure CN120339694A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of image processing technologies, and particularly to a prediction method, apparatus, device, and storage medium for high-resolution fire images. Background Art
[0002] With the intensification of global warming, the climate has gradually become arid and the temperature has risen, leading to frequent fires. Especially forest fires, due to their characteristics such as fast occurrence rate, rapid spread, and large damage, have become the focus of fire monitoring. Therefore, it is particularly important to prevent, monitor, and promptly manage the fire situation. High-resolution images are of great significance in fire monitoring. It can provide detailed information on the fire area, which helps to efficiently monitor and evaluate fires. However, the update cycle of high-resolution data in current satellite data is relatively long, making it difficult to simultaneously meet the requirements of high-frequency monitoring and high-detail analysis.
[0003] Currently, in order to obtain high-resolution data frequently for timely fire monitoring, the commonly used method is to interpolate the low-resolution data that can be obtained frequently to obtain high-resolution data. In this way, high-resolution data can be obtained frequently, but the high-resolution data obtained by such interpolation is not accurate. Summary of the Invention
[0004] To solve the problems in the related technologies, embodiments of the present disclosure provide a prediction method, apparatus, device, and storage medium for high-resolution fire images.
[0005] In a first aspect, embodiments of the present disclosure provide a prediction method for high-resolution fire images, including:
[0006] Obtain a low-resolution fire image and a high-resolution spatial geographic image of a target area;
[0007] Perform pixel-level geographical location matching on both the low-resolution fire image and the high-resolution spatial geographic image with the high-resolution fire image to be predicted;
[0008] Based on the low-resolution fire image and the high-resolution fire image to be predicted that are pixel-level geographically located, extract the fire point image features of n first neighboring pixel points, where the n first neighboring pixel points are the n first neighboring pixel points closest to each pixel point in the high-resolution fire image to be predicted on the low-resolution fire image, and n is an integer greater than 1;
[0009] Based on a high-resolution spatial geographic image and a high-resolution fire image to be predicted that are pixel-level geographically matched, extract the spatial geographic features of n second neighboring pixel points, where the n second neighboring pixel points are the n second neighboring pixel points closest to each pixel point in the high-resolution fire image to be predicted on the high-resolution spatial geographic image;
[0010] Input the extracted fire point image features of n first neighboring pixel points and the spatial geographic features of n second neighboring pixel points into a pre-trained prediction model to obtain the high-resolution fire image predicted by the pre-trained prediction model.
[0011] In a possible implementation manner, the extracting of the fire point image features of n first neighboring pixel points based on the low-resolution fire image and the high-resolution fire image to be predicted that are pixel-level geographically matched includes:
[0012] Based on the low-resolution fire image and the high-resolution fire image to be predicted that are pixel-level geographically matched, extract the fire direction vector feature, spatial distance weight feature, and radiation brightness temperature feature of n first neighboring pixel points;
[0013] Wherein, the spatial distance weight feature is the reciprocal of the actual geographic distance between the n first neighboring pixel points and the corresponding pixel points of the high-resolution fire image to be predicted.
[0014] In a possible implementation manner, the high-resolution spatial geographic image includes a high-resolution Normalized Difference Vegetation Index (NDVI) image, a surface vegetation type image, and a Digital Elevation Model (DEM) image; the extracting of the spatial geographic features of n second neighboring pixel points based on the high-resolution spatial geographic image and the high-resolution fire image to be predicted that are pixel-level geographically matched includes:
[0015] Based on the high-resolution spatial geographic image and the high-resolution fire image to be predicted that are pixel-level geographically matched, extract the geographic location information, elevation information, vegetation distribution type, and NDVI of n second neighboring pixel points.
[0016] In a possible implementation manner, the method further includes:
[0017] Obtain sample data, where the sample data includes multiple samples, and each sample includes the actually collected high-resolution fire image of the sample area and the fire point image features of n first neighboring pixel points corresponding to each pixel point in the low-resolution fire image, and the spatial geographic features of n second neighboring pixel points corresponding to each pixel point in the high-resolution spatial geographic image;
[0018] Use the sample data to train to obtain the prediction model, and the prediction model includes a Gradient Boosting Tree (GBT) model.
[0019] In a possible implementation, training the prediction model using the sample data includes:
[0020] Using the K-fold cross-validation algorithm to perform multiple splits and evaluations on the sample for parameter optimization in the sample data, and at the same time combining grid search to optimize the hyperparameters of the prediction model, to obtain an initial prediction model composed of the optimized hyperparameters;
[0021] Using the training samples in the sample data to train the initial prediction model to obtain the prediction model.
[0022] In a possible implementation, the method further includes:
[0023] Using a spatial smoothing algorithm to perform smoothing processing on the predicted high-resolution fire image to obtain a final high-resolution fire image.
[0024] In a second aspect, an embodiment of the present disclosure provides a method for training a prediction model, including:
[0025] Obtaining a low-resolution fire image, a high-resolution spatial geographic image, and a real-acquired high-resolution fire image of a sample area;
[0026] Performing pixel-level geographic location matching on both the low-resolution fire image and the high-resolution spatial geographic image of the sample area with the real-acquired high-resolution fire image;
[0027] Based on the low-resolution fire image and the real-acquired high-resolution fire image of the sample area with pixel-level geographic location matching, extracting the fire point image features of n first neighboring pixel points, where the n first neighboring pixel points are the n first neighboring pixel points closest to each pixel point in the real-acquired high-resolution fire image on the corresponding low-resolution fire image, and n is an integer greater than 1;
[0028] Based on the high-resolution spatial geographic image and the real-acquired high-resolution fire image of the sample area with pixel-level geographic location matching, extracting the fire point image features of n second neighboring pixel points, where the n second neighboring pixel points are the n second neighboring pixel points closest to each pixel point in the real-acquired high-resolution fire image on the corresponding high-resolution spatial geographic image;
[0029] Obtaining sample data, where the samples in the sample data include the real-acquired high-resolution fire image of the sample area and the fire point image features of n first neighboring pixel points corresponding to each pixel point in the corresponding low-resolution fire image, and the spatial geographic features of n second neighboring pixel points corresponding to each pixel point in the corresponding high-resolution spatial geographic image;
[0030] Using the sample data, the prediction model is trained and obtained.
[0031] In a third aspect, an embodiment of the present disclosure provides a prediction device for high-resolution fire images, including:
[0032] An image acquisition module, configured to acquire a low-resolution fire image and a high-resolution spatial geographic image of a target area;
[0033] A matching module, configured to perform pixel-level geographic location matching on both the low-resolution fire image and the high-resolution spatial geographic image with the high-resolution fire image to be predicted;
[0034] A first feature extraction module, configured to extract fire point image features of n first neighboring pixel points based on the low-resolution fire image and the high-resolution fire image to be predicted that are pixel-level geographically matched, where the n first neighboring pixel points are the n first neighboring pixel points closest to each pixel point in the high-resolution fire image to be predicted on the low-resolution fire image, and n is an integer greater than 1;
[0035] A second feature extraction module, configured to extract spatial geographic features of n second neighboring pixel points based on the high-resolution spatial geographic image and the high-resolution fire image to be predicted that are pixel-level geographically matched, where the n second neighboring pixel points are the n second neighboring pixel points closest to each pixel point in the high-resolution fire image to be predicted on the high-resolution spatial geographic image;
[0036] A prediction module, configured to input the extracted fire point image features of n first neighboring pixel points and the spatial geographic features of n second neighboring pixel points into a pre-trained prediction model to obtain the high-resolution fire image predicted by the pre-trained prediction model.
[0037] In a fourth aspect, an embodiment of the present disclosure provides an electronic device, including a memory and a processor, where the memory is used to store one or more computer instructions, and the one or more computer instructions are executed by the processor to implement the method as described in any one of the first aspect.
[0038] In a fifth aspect, an embodiment of the present disclosure provides a computer-readable storage medium, on which computer instructions are stored, and when the computer instructions are executed by a processor, the method as described in any one of the first aspect is implemented.
[0039] According to the technical solution provided by the embodiments of the present disclosure, a low-resolution fire image and a high-resolution spatial geographic image of a target area can be obtained; both the low-resolution fire image and the high-resolution spatial geographic image are pixel-level geographically matched with the high-resolution fire image to be predicted, so that the fire image features of n first neighboring pixels of each pixel point in the high-resolution fire image to be predicted on the low-resolution fire image and the spatial geographic features of n second neighboring pixels on the high-resolution spatial geographic image can be extracted. In this way, the input features of the prediction model are obtained, and a non-linear mapping relationship is established between the input features and the high-resolution fire image through the prediction model. In this way, the fire image with high frequency and low spatial resolution can be converted into a fire image with high frequency and high spatial resolution, breaking through the limitation of "incompatibility between time and spatial resolution" in satellites, accurately obtaining high-resolution fire images with high frequency, and further improving the accuracy and efficiency of fire monitoring.
[0040] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] In combination with the accompanying drawings, through the following detailed description of non-limiting embodiments, other features, objects, and advantages of the present disclosure will become more obvious. In the drawings:
[0042] Figure 1 A flowchart of a method for predicting a high-resolution fire image provided by an embodiment of the present disclosure is shown.
[0043] Figure 2 Training data provided by this embodiment is shown.
[0044] Figure 3 The predicted high-resolution fire image and the high-resolution fire image collected on the spot are shown.
[0045] Figure 4 A flowchart of a method for training a prediction model provided by an embodiment of the present disclosure is shown.
[0046] Figure 5 A structural block diagram of a device for predicting a high-resolution fire image provided by an embodiment of the present disclosure is shown.
[0047] Figure 6 A structural block diagram of a device for training a prediction model provided by an embodiment of the present disclosure is shown.
[0048] Figure 7 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown.
[0049] Figure 8 A schematic structural diagram of a computer system suitable for implementing the method of the embodiment of the present disclosure is shown. Detailed implementation manners
[0050] In the following, exemplary embodiments of the present disclosure will be described in detail with reference to the accompanying drawings, so that those skilled in the art can easily implement them. In addition, for clarity, parts unrelated to the description of the exemplary embodiments are omitted in the drawings.
[0051] In the present disclosure, it should be understood that terms such as "including" or "having" are intended to indicate the presence of features, numbers, steps, actions, components, parts, or combinations thereof disclosed in this specification, and are not intended to exclude the possibility of the presence or addition of one or more other features, numbers, steps, actions, components, parts, or combinations thereof.
[0052] In addition, it should be noted that, without conflict, the embodiments in the present disclosure and the features in the embodiments can be combined with each other. The present disclosure will be described in detail below with reference to the drawings and in combination with the embodiments.
[0053] Figure 1 The flowchart showing a prediction method for a high-resolution fire image provided by an embodiment of the present disclosure is as follows Figure 1 As shown, the prediction method for the high-resolution fire image includes the following steps S101 - S105:
[0054] In step S101, a low-resolution fire image and a high-resolution spatial geographic image of a target area are acquired;
[0055] In step S102, both the low-resolution fire image and the high-resolution spatial geographic image are subjected to pixel-level geographical location matching with the high-resolution fire image to be predicted;
[0056] In step S103, based on the low-resolution fire image and the high-resolution fire image to be predicted that are pixel-level geographically located, fire point image features of n first neighboring pixel points are extracted, where the n first neighboring pixel points are the n first neighboring pixel points closest to each pixel point in the high-resolution fire image to be predicted on the low-resolution fire image;
[0057] In step S104, based on the high-resolution spatial geographic image and the high-resolution fire image to be predicted that are pixel-level geographically located, spatial geographic features of n second neighboring pixel points are extracted, where the n second neighboring pixel points are the n second neighboring pixel points closest to each pixel point in the high-resolution fire image to be predicted on the high-resolution spatial geographic image;
[0058] In step S105, the fire image features of the extracted n first neighboring pixel points and the spatial geographical features of the n second neighboring pixel points are input into a pre-trained prediction model to obtain a high-resolution fire image predicted by the pre-trained prediction model.
[0059] In a possible implementation manner, the prediction of the high-resolution fire image is applicable to electronic devices such as computing devices and servers that can execute the prediction of the high-resolution fire image.
[0060] In a possible implementation manner, low-resolution images can usually be acquired at a high frequency. For example, GOES (Geostationary Operational Environmental Satellites) is fixed at 75.2°W and observes the Americas. GOES-16 provides high-frequency and high-resolution Earth images in 16 different bands, and the wavelength range of these 16 bands is from 0.47μm to 13.3μm, including 2 visible light channels (0.5 - 1Km); 4 near-infrared channels (1 - 2Km); 10 infrared bands (2Km). Among them, infrared band I4 is often used to monitor fires. Although the image resolution of GOES is low, its recurrence period is about 23.9 hours, which may be more timely in fire monitoring.
[0061] In a possible implementation manner, the acquisition frequency of high-resolution images will be lower. For example, VIIRS (Visible Infrared Imaging Radiometer Suite) is a sensor carried on the NPP (National Polar-orbiting Partnership) satellite and can collect radiation images of land, atmosphere, ice sheet, and ocean in visible and infrared bands. VIIRS provides 22 bands with a wavelength range from 0.4μm to 12.5μm, including 5 high-resolution image channels (I band, 375m); 16 medium-resolution channels (M band, 750m); 1 day / night band (DNB, 750m). The fire image generated based on its data has a resolution of 750m and belongs to high-resolution data. However, its revisit period is long, about 16 days.
[0062] In a possible implementation, in order to obtain high-resolution fire images at high frequencies, high-resolution fire images can be predicted through low-resolution fire images at high frequencies and high-resolution spatial geographic images that remain unchanged for a long time. For example, the low-resolution fire images can be low-resolution fire images obtained based on the infrared band I4 images of GOES, and the low-resolution fire images here can be surface temperature image data. The high-resolution spatial geographic images can be obtained from a high-resolution satellite image database, and can be NDVI (Normalized Difference Vegetation Index) images, surface vegetation type images, DEM (Digital Elevation Model) images, etc.
[0063] Among them, NDVI is a remote sensing index used to evaluate vegetation cover and health status, and the NDVI value of each pixel point is recorded in the NDVI image; the DEM image is a digital image used to represent surface elevation information, and it can record the elevation value of each pixel point in the form of a regular grid; for the surface vegetation type image, remote sensing image classification technology can be used to classify the surface vegetation types in the target area, such as forests, grasslands, shrubs, etc., and map them to flammability levels; because different vegetation types have different combustion characteristics, affecting the flammability and spread rate of fires, a combustion rating system can be adopted to classify each surface vegetation, for example, dividing each surface vegetation into five combustion levels, etc.; the combustion level of each pixel point is recorded in the surface vegetation type image.
[0064] In a possible implementation, the low-resolution fire images, high-resolution spatial geographic images, and the high-resolution fire images to be predicted can be projected in an equal longitude and latitude manner to eliminate the spatial position deviation between the low-resolution fire images, high-resolution spatial geographic images, and the high-resolution fire images to be predicted, and achieve pixel-level geographical location matching. It should be noted here that the longitude and latitude range and spatial resolution of the high-resolution fire images to be predicted can be known in advance, but the specific content of the images needs to be predicted later, so that pixel-level geographical location matching can be performed with the low-resolution fire images and high-resolution spatial geographic images.
[0065] In a possible implementation, corresponding features can be extracted from the low-resolution fire images and high-resolution spatial geographic images.
[0066] Among them, in order to capture the local environmental information of each pixel point in the high-resolution fire image to be predicted on the low-resolution image and reflect the diffusion direction and trend of the fire, n first neighboring pixel points closest to each pixel point in the high-resolution fire image to be predicted on the low-resolution fire image can be extracted, and the fire image features of these first neighboring points can be obtained. In this way, the fire image features of the n first neighboring pixel points can be used to characterize the fire situation of the corresponding pixel point in the high-resolution fire image to be predicted. The fire image feature can be any feature that can reflect the fire situation. The prediction of the high-resolution fire image not only depends on the direct measurement of the fire intensity, but also needs to comprehensively consider the potential impact of geographical environmental factors on the fire spread. Therefore, n second neighboring pixel points closest to each pixel point in the high-resolution fire image to be predicted on the high-resolution spatial geographical image can be extracted, and the spatial geographical features of these second neighboring points can be obtained. The spatial geographical feature can be any feature that can reflect the spatial geographical situation of the neighboring area of each pixel point in the high-resolution fire image to be predicted.
[0067] In a possible implementation manner, the prediction model represents the non-linear mapping relationship between the input features and the high-resolution fire image, and is mainly used to utilize the non-linear mapping relationship between the input features and the high-resolution fire image. Based on the input features, that is, the fire image features of the n first neighboring pixel points of each pixel point in the high-resolution fire image to be predicted on the low-resolution fire image and the spatial geographical features of the n second neighboring pixel points on the high-resolution spatial geographical image. In this way, the fire image features of the n first neighboring pixel points and the spatial geographical features of the n second neighboring pixel points obtained previously can be input into the pre-trained prediction model to obtain the high-resolution fire image predicted by the pre-trained prediction model. The prediction model not only predicts the change of the fire intensity, but also maintains the spatial correlation, making the output high-resolution fire image have strong spatial continuity.
[0068] This embodiment can obtain a low-resolution fire image and a high-resolution spatial geographic image of the target area; perform pixel-level geographical location matching on both the low-resolution fire image and the high-resolution spatial geographic image with the high-resolution fire image to be predicted, so that the fire image features of n first neighboring pixels of each pixel point in the high-resolution fire image to be predicted on the low-resolution fire image and the spatial geographic features of n second neighboring pixels on the high-resolution spatial geographic image can be extracted. In this way, the input features of the prediction model are obtained, and a non-linear mapping relationship is established between the input features and the high-resolution fire image through the prediction model. In this way, the fire image with high frequency and low spatial resolution can be converted into a fire image with high frequency and high spatial resolution, breaking through the limitation of "incompatibility between time and spatial resolution" in satellites, accurately obtaining high-resolution fire images at high frequencies, and further improving the accuracy and efficiency of fire monitoring.
[0069] In a possible implementation manner, for the low-resolution fire image and the high-resolution fire image to be predicted based on pixel-level geographical location matching, extracting the fire image features of n first neighboring pixels includes:
[0070] Based on the pixel-level geographical location matching of the low-resolution fire image and the high-resolution fire image to be predicted, extracting the fire direction vector feature, spatial distance weight feature, and radiation brightness temperature feature of n first neighboring pixels;
[0071] Among them, the spatial distance weight feature is the reciprocal of the actual geographical distance between the n first neighboring pixels and the corresponding pixel of the high-resolution fire image to be predicted.
[0072] In this embodiment, the fire image feature may include at least one of the fire direction vector, spatial distance weight feature, and radiation brightness temperature feature. Preferably, the fire image feature may include the fire direction vector, spatial distance weight feature, and radiation brightness temperature feature.
[0073] In this embodiment, the fire direction vector feature of each pixel point in the low-resolution fire image can be calculated. The fire direction vector feature is used to identify the development trend of the fire. For example, the fire direction vector feature can be represented by (dx, dy), where dx and dy are the components of the fire in the horizontal and vertical directions respectively. The fire direction vectors of these n neighboring pixels extracted in the low-resolution fire image help to understand the diffusion direction and rate of the fire.
[0074] In this embodiment, the actual geographical distance between each neighboring pixel point on the low-resolution fire image and the corresponding pixel point in the high-resolution fire image to be predicted can be calculated. The reciprocal of this actual geographical distance is denoted as the spatial distance weight feature, and the smaller the actual geographical distance, the greater the weight. This spatial distance weight feature can quantify the spatial influence of fire spread.
[0075] In this embodiment, the radiation brightness temperature value corresponding to the neighboring pixel points can be extracted as the radiation brightness temperature feature, which is important temperature intensity information in fire intensity prediction.
[0076] In a possible embodiment, the high-resolution spatial geographical image includes a high-resolution Normalized Difference Vegetation Index (NDVI) image, a surface vegetation type image, and a Digital Elevation Model (DEM) image; based on the high-resolution spatial geographical image and the high-resolution fire image to be predicted that are matched at the pixel-level geographical location, the spatial geographical features of n second neighboring pixel points are extracted, including:
[0077] Based on the high-resolution spatial geographical image and the high-resolution fire image to be predicted that are matched at the pixel-level geographical location, the geographical location information, elevation information, vegetation distribution type, and NDVI of n second neighboring pixel points are extracted.
[0078] In this embodiment, the geographical location information can be the geographical coordinates (such as longitude and latitude) of each pixel point in the high-resolution fire image to be predicted among the n second neighboring pixel points in the high-resolution spatial geographical image. This geographical location information can provide spatial positioning information to help the prediction model understand the distribution pattern of the fire in the geographical space.
[0079] In this embodiment, the elevation information of each pixel point in the high-resolution fire image to be predicted among the n second neighboring pixel points in the DEM image can be obtained from the high-resolution DEM image. This elevation information can quantify the constraint of terrain undulation on the fire spread path. This elevation information can affect the spread speed and direction of the fire. For example, highlands may hinder the spread of the fire or cause changes in the wind direction.
[0080] In this embodiment, different vegetation types have different combustion characteristics, which affect the flammability and spread rate of the fire. Therefore, the vegetation distribution type of each pixel point in the high-resolution fire image to be predicted among the n second neighboring pixel points in the high-resolution surface vegetation type image, such as forest, grassland, shrub, etc., can be extracted.
[0081] In this embodiment, the NDVI of the n second neighboring pixels of each pixel of the to-be-predicted high-resolution fire image can be extracted from the high-resolution NDVI image. The high NDVI value area indicates dense vegetation, which may increase the fuel supply for the fire and result in a larger fire, while the low NDVI value area may indicate sparse vegetation and a faster fire spread speed.
[0082] In this embodiment, by introducing the fire point image features and spatial geographical features of the neighboring pixels of each pixel in the to-be-predicted high-resolution fire image as relevant variables, the complex spatial relationships in the fire area are fully captured. At the same time, geographical features such as vegetation type and terrain undulation, as well as temporal dynamic factors such as temperature and humidity, are considered, comprehensively reflecting multiple factors affecting fire intensity, fire direction, and propagation speed, thereby more accurately predicting the evolution process of the fire.
[0083] In a possible embodiment, the method further includes:
[0084] Obtaining sample data, where the sample data includes multiple samples, and each sample includes the actually collected high-resolution fire image of the sample area, the fire point image features of the n first neighboring pixels corresponding to each pixel in the low-resolution fire image, and the spatial geographical features of the n second neighboring pixels corresponding to each pixel in the high-resolution spatial geographical image;
[0085] Using the sample data to train the prediction model, where the prediction model includes a Gradient Boosting Tree (GBT) model.
[0086] In this embodiment, the low-resolution fire image, high-resolution spatial geographical image, and actually collected high-resolution fire image of the historically collected sample area can be obtained. The low-resolution fire image and high-resolution spatial geographical image of the sample area are both pixel-level geographically matched with the actually collected high-resolution fire image. Then, based on the pixel-level geographically matched low-resolution fire image and actually collected high-resolution fire image of the sample area, the fire point image features of the n first neighboring pixels are extracted, where the n neighboring pixels are the n first neighboring pixels closest to each pixel in the actually collected high-resolution fire image on the corresponding low-resolution fire image. At the same time, based on the pixel-level geographically matched high-resolution spatial geographical image and actually collected high-resolution fire image of the sample area, the fire point image features of the n second neighboring pixels are extracted, where the n neighboring pixels are the n second neighboring pixels closest to each pixel in the actually collected high-resolution fire image on the corresponding high-resolution spatial geographical image. In this way, multiple samples in the sample data are obtained.
[0087] In this embodiment, the above sample data can be used to train a GBT (Gradient Boosted Trees) model. The GBT model is an ensemble learning algorithm based on decision trees, which constructs a strong prediction model by gradually optimizing the loss function.
[0088] In this embodiment, the GBT model can effectively process high-dimensional features and capture non-linear relationships, and is suitable for complex patterns in fire intensity prediction. The theoretical basis of this prediction method is that the radiance temperature value of the pixel points in the high-resolution fire image is related to the radiance temperature value, distance and fire direction of the pixel points in the adjacent low-resolution fire image. In addition to the low-resolution fire image at the same time, the geographical location information, elevation information, vegetation distribution type, and NDVI of the high-resolution pixel points in the same area also affect their radiance temperature values. Therefore, the relationship formula is established as follows:
[0089] Y = A·X + δ;
[0090] Where, Y is the pixel matrix of the high-resolution fire image, X is the feature matrix, A represents the regression function, and δ represents the residual.
[0091] Among them, random forest is an ensemble learning method. By generating multiple decision trees and combining the prediction results of each tree, it can effectively reduce overfitting and improve the generalization ability of the model. The leaf nodes of each decision tree represent different combinations of input features and output prediction values. To further improve the prediction accuracy of the model, the gradient boosting algorithm gradually optimizes the residuals of each tree and combines the prediction results of each tree in a weighted manner, thereby improving the accuracy of the overall model.
[0092] The gradient boosting algorithm optimized model provided in this embodiment can automatically identify important features and avoid overfitting, further improving the prediction accuracy.
[0093] In a possible embodiment, the using the sample data to train the prediction model includes:
[0094] Using the K-fold cross-validation algorithm to split and evaluate the sample for parameter optimization in the sample data multiple times, and at the same time combining grid search to optimize the hyperparameters of the prediction model, obtaining an initial prediction model composed of the optimized hyperparameters;
[0095] Using the training sample in the sample data to train the initial prediction model to obtain the prediction model.
[0096] In this embodiment, the K-Fold Cross-Validation method is a commonly used model evaluation method. By splitting and evaluating the samples multiple times, it reduces the variance of the model evaluation results and improves the stability and reliability of the evaluation.
[0097] In this embodiment, Grid Search is a commonly used hyperparameter optimization method. By traversing the given hyperparameter combinations, it searches for the hyperparameter configuration that optimizes the model performance to enhance the model's prediction ability.
[0098] In this embodiment, the hyperparameters of the GBT model include the number of trees, the depth of the trees, the learning rate, and so on.
[0099] For example, during hyperparameter optimization, when using Grid Search to traverse a given set of hyperparameter combinations, K iterations can be performed. In each iteration, a subset is selected as the validation set, and the remaining K - 1 subsets are combined as the training set. The prediction model constructed with the given set of hyperparameter combinations is trained on the training set. After training, the model performance of the trained prediction model is evaluated on the validation set, and the evaluation results (such as accuracy, mean squared error, etc.) are recorded. The average value of the K evaluation results can be used as the final performance of the hyperparameter combination. After traversing all the hyperparameter combinations using Grid Search, the hyperparameter combination with the optimal performance can be obtained, thus ensuring that the model maintains a high prediction accuracy and adaptability when facing different types of fire data.
[0100] In this embodiment, after obtaining the initial prediction model composed of the optimized hyperparameters, the training samples in the sample data can be used to train the initial prediction model to obtain the above-mentioned pre-trained prediction model. The specific training process is well-known to those skilled in the art and will not be elaborated here.
[0101] In a possible embodiment, the method further includes:
[0102] Performing smoothing processing on the predicted high-resolution fire image using a spatial smoothing algorithm to obtain the final high-resolution fire image.
[0103] In this embodiment, the Spatial Smoothing Algorithm is a filtering or noise reduction technique for processing spatial data. Its core idea is to eliminate noise or outliers by performing weighted averaging or aggregation on the data within a local region, while preserving the overall structure and trend of the data. To avoid noise introduced by prediction errors, the Spatial Smoothing Algorithm can be used to smooth the predicted high-resolution fire image, ensuring the smoothness and continuity of the final high-resolution fire image, which helps to further improve the visual quality of the high-resolution fire image.
[0104] To verify the accuracy and effectiveness of the above method for predicting high-resolution fire images, Figure 2 the training data provided in this embodiment is shown. As Figure 2 shown in the figure, the images are VIIRS fire point data (high-resolution fire images with a spatial resolution of 375m) and GOES brightness temperature data (low-resolution fire images with a spatial resolution of 2KM) of the target area (latitude 43.15 to 45.15, longitude -123.45 to -125.45) on September 8, 2020 and September 11, 2020. Among them, Figure 2 the left image in the figure is the low-resolution GOES brightness temperature data, and the right image is the high-resolution VIIRS fire point data. After training the fire point data of the target area by the method provided in this embodiment, high-resolution prediction is performed on the low-resolution GOES brightness temperature data of the target area on September 9, 2020. Figure 3 The predicted high-resolution fire image and the high-resolution fire image collected on the ground are shown. The result Figure 3 the left image in the figure is the predicted high-resolution fire image, Figure 3 the right image in the figure is the high-resolution VIIRS fire point data collected on the ground in the target area on September 9, 2020. Figure 3 The abscissa in the figure is longitude and the ordinate is latitude. The color card is used to represent the brightness temperature, with the unit of K. By comparing the high-resolution VIIRS fire point data collected on the ground and the high-resolution fire image predicted by the method provided in this embodiment, it can be seen that the fire point areas and intensities of the two images are very similar, indicating that the prediction by the method provided in this embodiment is very accurate.
[0105] Figure 4 The flowchart of a prediction model training method provided by an embodiment of the present disclosure is shown. As Figure 4 shown, the prediction model training method includes the following steps S401 - S406:
[0106] In step S401, a low-resolution fire image, a high-resolution spatial geographic image, and a high-resolution fire image collected on the ground of a sample area are obtained;
[0107] In step S402, both the low-resolution fire image and the high-resolution spatial geographic image of the sample area are subjected to pixel-level geographic location matching with the actually collected high-resolution fire image;
[0108] In step S403, based on the low-resolution fire image and the actually collected high-resolution fire image of the sample area obtained by pixel-level geographic location matching, fire point image features of n first neighboring pixel points are extracted, where the n first neighboring pixel points are the n first neighboring pixel points closest to each pixel point in the actually collected high-resolution fire image on the corresponding low-resolution fire image;
[0109] In step S404, based on the high-resolution spatial geographic image and the actually collected high-resolution fire image of the sample area obtained by pixel-level geographic location matching, fire point image features of n second neighboring pixel points are extracted, where the n second neighboring pixel points are the n second neighboring pixel points closest to each pixel point in the actually collected high-resolution fire image on the corresponding high-resolution spatial geographic image;
[0110] In step S405, sample data is obtained. The samples in the sample data include the actually collected high-resolution fire image of the sample area and the fire point image features of the n first neighboring pixel points corresponding to each pixel point in the corresponding low-resolution fire image, and the spatial geographic features of the n second neighboring pixel points corresponding to each pixel point in the corresponding high-resolution spatial geographic image;
[0111] In step S406, the prediction model is trained using the sample data.
[0112] In a possible implementation manner, this prediction model training method is applicable to electronic devices such as computing devices and servers that can execute the training of the prediction model.
[0113] In a possible implementation manner, the low-resolution fire image may be a low-resolution fire image obtained based on the infrared band I4 image of GOES. Here, the low-resolution fire image may be surface temperature image data. The high-resolution spatial geographic image may be an NDVI image, a surface vegetation type image, a DEM image, etc.
[0114] Among them, NDVI is a remote sensing index used to evaluate vegetation cover and health status, and the NDVI value of each pixel is recorded in the NDVI image; the DEM image is a digital image used to represent surface elevation information, and it can record the elevation value of each pixel in the form of a regular grid; for this surface vegetation type image, remote sensing image classification technology can be used to classify the surface vegetation types in the target area, such as forests, grasslands, shrubs, etc., and map them to flammability levels; because different vegetation types have different combustion characteristics, affecting the flammability and spread rate of fires, a combustion rating system can be adopted to classify each surface vegetation, for example, each surface vegetation is divided into five combustion levels, etc.; the combustion level of each pixel is recorded in this surface vegetation type image.
[0115] In a possible implementation manner, the low-resolution fire image, the high-resolution spatial geographic image, and the actually captured high-resolution fire image can be projected in an equi-longitude and equi-latitude manner to eliminate the spatial position deviation between the low-resolution fire image, the high-resolution spatial geographic image, and the actually captured high-resolution fire image, and achieve pixel-level geographic location matching. It should be noted here that the longitude and latitude range and spatial resolution of the actually captured high-resolution fire image can be known in advance, but the specific content of its image needs to be predicted later, so that pixel-level geographic location matching can be performed with the low-resolution fire image and the high-resolution spatial geographic image.
[0116] In a possible implementation manner, corresponding features can be extracted from the low-resolution fire image and the high-resolution spatial geographic image of the sample area.
[0117] Among them, in order to capture the local environmental information of each pixel in the actually captured high-resolution fire image on the low-resolution image and reflect the spread direction and trend of the fire, n first neighboring pixels closest to each pixel in the actually captured high-resolution fire image on the low-resolution fire image can be extracted, and the fire point image features of these first neighboring points can be obtained. In this way, the fire point image features of n first neighboring pixels can be used to characterize the fire point situation of the corresponding pixel in the actually captured high-resolution fire image, and the fire point image feature can be any feature that can reflect the fire point situation. The prediction of the high-resolution fire image not only depends on the direct measurement of the fire intensity, but also needs to comprehensively consider the potential impact of geographical environmental factors on the fire spread. Therefore, n second neighboring pixels closest to each pixel in the actually captured high-resolution fire image on the high-resolution spatial geographic image can be extracted, and the spatial geographic features of these second neighboring points can be obtained. The spatial geographic feature can be any feature that can reflect the spatial geographic situation of the neighboring area of each pixel in the actually captured high-resolution fire image.
[0118] In a possible implementation, the actually captured high-resolution fire image of the sample area can be a high-resolution fire image captured in the sample area at the same time. For example, it can be the high-resolution fire image provided by VIIRS.
[0119] In a possible implementation, the sample data can be used to train the prediction model. The fire image features of n first neighboring pixels corresponding to each pixel point in the actually captured high-resolution fire image of the sample area in the low-resolution fire image and the spatial geographical features of n second neighboring pixels corresponding to each pixel point in the high-resolution spatial geographical image can be input into the prediction model to obtain the predicted high-resolution fire image output by the prediction model. The predicted high-resolution fire image is compared with the actually captured high-resolution fire image to obtain the prediction accuracy rate. Based on the prediction accuracy rate, the model parameters of the prediction model are continuously adjusted. In this way, the model parameters of the prediction model are continuously adjusted until the prediction accuracy rate reaches a predetermined threshold such as 96% or the like, and a trained prediction model is obtained.
[0120] In a possible implementation, the prediction model can be a GBT model.
[0121] In a possible implementation, the training of the prediction model using the sample data includes:
[0122] The parameter optimization sample in the sample data is split and evaluated multiple times using the K-fold cross-validation algorithm, and at the same time, grid search is combined to optimize the hyperparameters of the prediction model to obtain an initial prediction model composed of the optimized hyperparameters;
[0123] The training sample in the sample data is used to train the initial prediction model to obtain the prediction model.
[0124] The present disclosure also provides a prediction device for high-resolution fire images. Figure 5 The structural block diagram of a prediction device for high-resolution fire images provided by an embodiment of the present disclosure is shown. This device can be implemented as part or all of an electronic device through software, hardware, or a combination of both. As Figure 5 shown, the prediction device for high-resolution fire images includes:
[0125] An image acquisition module 501, configured to acquire a low-resolution fire image and a high-resolution spatial geographical image of a target area;
[0126] A matching module 502, configured to perform pixel-level geographical location matching on both the low-resolution fire image and the high-resolution spatial geographical image with the high-resolution fire image to be predicted;
[0127] The first feature extraction module 503 is configured to extract the fire point image features of n first neighboring pixel points based on the low-resolution fire image and the high-resolution fire image to be predicted that are matched at the pixel-level geographical location. Wherein, the n first neighboring pixel points are the n first neighboring pixel points that are closest to each pixel point in the high-resolution fire image to be predicted on the low-resolution fire image, and n is an integer greater than 1;
[0128] The second feature extraction module 504 is configured to extract the spatial geographical features of n second neighboring pixel points based on the high-resolution spatial geographical image and the high-resolution fire image to be predicted that are matched at the pixel-level geographical location. Wherein, the n second neighboring pixel points are the n second neighboring pixel points that are closest to each pixel point in the high-resolution fire image to be predicted on the high-resolution spatial geographical image;
[0129] The prediction module 505 is configured to input the extracted fire point image features of n first neighboring pixel points and the spatial geographical features of n second neighboring pixel points into a pre-trained prediction model to obtain the high-resolution fire image predicted by the pre-trained prediction model.
[0130] In a possible implementation manner, the first feature extraction module is configured to:
[0131] Based on the low-resolution fire image and the high-resolution fire image to be predicted that are matched at the pixel-level geographical location, extract the fire direction vector features, spatial distance weight features, and radiance temperature features of n first neighboring pixel points;
[0132] Wherein, the spatial distance weight feature is the reciprocal of the actual geographical distance between the n first neighboring pixel points and the corresponding pixel points of the high-resolution fire image to be predicted.
[0133] In a possible implementation manner, the high-resolution spatial geographical image includes a high-resolution normalized difference vegetation index (NDVI) image, a surface vegetation type image, and a digital elevation model (DEM) image; the second feature extraction module is configured to:
[0134] Based on the high-resolution spatial geographical image and the high-resolution fire image to be predicted that are matched at the pixel-level geographical location, extract the geographical location information, elevation information, vegetation distribution type, and NDVI of n second neighboring pixel points.
[0135] In a possible implementation manner, the device further includes:
[0136] A model training module, configured to obtain sample data, where the sample data includes multiple samples, and each sample includes a real-time high-resolution fire image of a sample area and fire point image features of n first neighboring pixels corresponding to each pixel point in a low-resolution fire image, and spatial geographical features of n second neighboring pixels corresponding to the high-resolution spatial geographical image; using the sample data, a prediction model is trained, and the prediction model includes a Gradient Boosting Tree (GBT) model.
[0137] In a possible implementation manner, the part in the model training module that uses the sample data to train the prediction model is configured to:
[0138] The parameter optimization sample in the sample data is split and evaluated multiple times by using the K-fold cross-validation algorithm, and at the same time, grid search is combined to optimize the hyperparameters of the prediction model, and an initial prediction model composed of the optimized hyperparameters is obtained;
[0139] The initial prediction model is trained by using the training samples in the sample data to obtain the prediction model.
[0140] In a possible implementation manner, the device further includes:
[0141] A smoothing module, configured to perform smoothing processing on the predicted high-resolution fire image by using a spatial smoothing algorithm to obtain a final high-resolution fire image.
[0142] The present disclosure also provides a prediction model training device, Figure 6 showing a structural block diagram of a prediction model training device provided by an embodiment of the present disclosure. The device can be implemented as part or all of an electronic device through software, hardware, or a combination of both. As Figure 6 shown, the prediction model training device includes:
[0143] A data acquisition module 601, configured to acquire a low-resolution fire image, a high-resolution spatial geographical image, and a real-time high-resolution fire image of a sample area;
[0144] A position matching module 602, configured to perform pixel-level geographical location matching on both the low-resolution fire image and the high-resolution spatial geographical image of the sample area with the real-time high-resolution fire image;
[0145] The first sample feature extraction module 603 is configured to extract the fire point image features of n first neighboring pixel points based on the low-resolution fire image and the actually captured high-resolution fire image of the sample area with pixel-level geographical location matching. Among them, the n neighboring pixel points are the n first neighboring pixel points closest to each pixel point in the actually captured high-resolution fire image on the corresponding low-resolution fire image, and n is an integer greater than 1;
[0146] The second sample feature extraction module 604 is configured to extract the fire point image features of n second neighboring pixel points based on the high-resolution spatial geographical image and the actually captured high-resolution fire image of the sample area with pixel-level geographical location matching. Among them, the n neighboring pixel points are the n second neighboring pixel points closest to each pixel point in the actually captured high-resolution fire image on the corresponding high-resolution spatial geographical image;
[0147] The sample data acquisition module 605 is configured to acquire sample data. The samples in the sample data include the actually captured high-resolution fire image of the sample area and the fire point image features of n first neighboring pixel points corresponding to each pixel point in the corresponding low-resolution fire image, and the spatial geographical features of n second neighboring pixel points corresponding to each pixel point in the corresponding high-resolution spatial geographical image;
[0148] The training module 606 is configured to use the sample data to train and obtain the prediction model.
[0149] The technical terms and technical features mentioned in the embodiments of this device are the same as or similar to those mentioned in the above method embodiments. For the explanations and descriptions of the technical terms and technical features involved in this device, reference can be made to the explanations and descriptions of the above method embodiments, which will not be repeated here.
[0150] The present disclosure also discloses an electronic device, Figure 7 The structural block diagram of the electronic device showing the embodiments according to the present disclosure is shown.
[0151] As Figure 7 shown, the electronic device 700 includes a memory 701 and a processor 702. Among them, the memory 701 is used to store one or more computer instructions, and the one or more computer instructions are executed by the processor 702 to implement the method according to the embodiments of the present disclosure.
[0152] Figure 8 The structural schematic diagram of the computer system suitable for implementing the method of the embodiments of the present disclosure is shown.
[0153] As Figure 8As shown in the figure, the computer system 800 includes a processing unit 801, which can perform various processes in the above embodiments according to a program stored in a read-only memory (ROM) 802 or a program loaded from a storage section 808 into a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the computer system 800 are also stored. The processing unit 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.
[0154] The following components are connected to the I / O interface 805: an input section 806 including a keyboard, a mouse, etc.; an output section 807 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc. and a speaker, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a LAN card, a modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the I / O interface 805 as needed. A removable medium 811, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is mounted on the drive 810 as needed so that a computer program read therefrom is installed into the storage section 808 as needed. Among them, the processing unit 801 can be implemented as a processing unit such as a CPU, a GPU, a TPU, an FPGA, an NPU, etc.
[0155] In particular, according to an embodiment of the present disclosure, the method described above can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes computer instructions that implement the method steps described above when executed by a processor. In such an embodiment, the computer program product can be downloaded and installed from a network via the communication section 809, and / or installed from the removable medium 811.
[0156] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a part of code, which 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 than that marked in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, and they may 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, as well as the combinations of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.
[0157] The units or modules described in the embodiments of the present disclosure can be implemented in software or in programmable hardware. The described units or modules can also be set in a processor, and the names of these units or modules do not, in some cases, constitute a limitation on the units or modules themselves.
[0158] As another aspect, the present disclosure also provides a computer-readable storage medium, which can be the computer-readable storage medium included in the electronic device or computer system in the above embodiments; or can be a computer-readable storage medium that exists separately and is not assembled into the device. The computer-readable storage medium stores one or more programs, and the programs are used by one or more processors to execute the methods described in the present disclosure.
[0159] The above description is only for the preferred embodiments of the present disclosure and the explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the present disclosure is not limited to the technical solutions formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) having similar functions disclosed in the present disclosure.
Claims
1. A prediction method for high-resolution fire images, characterized in that, Including: Obtain a low-resolution fire image and a high-resolution spatial geographic image of the target area; Perform pixel-level geographical location matching on both the low-resolution fire image and the high-resolution spatial geographic image with the high-resolution fire image to be predicted; Based on the low-resolution fire image and the high-resolution fire image to be predicted that are pixel-level geographically matched, extract the fire point image features of n first neighboring pixel points, where the n first neighboring pixel points are the n first neighboring pixel points closest to each pixel point in the high-resolution fire image to be predicted on the low-resolution fire image, and n is an integer greater than 1; Based on the high-resolution spatial geographic image and the high-resolution fire image to be predicted that are pixel-level geographically matched, extract the spatial geographic features of n second neighboring pixel points, where the n second neighboring pixel points are the n second neighboring pixel points closest to each pixel point in the high-resolution fire image to be predicted on the high-resolution spatial geographic image; Input the extracted fire point image features of n first neighboring pixel points and the spatial geographic features of n second neighboring pixel points into a pre-trained prediction model to obtain the high-resolution fire image predicted by the pre-trained prediction model.
2. The method according to claim 1, characterized in that The extracting the fire point image features of n first neighboring pixel points based on the low-resolution fire image and the high-resolution fire image to be predicted that are pixel-level geographically matched includes: Based on the low-resolution fire image and the high-resolution fire image to be predicted that are pixel-level geographically matched, extract the fire direction vector features, spatial distance weight features, and radiant brightness temperature features of n first neighboring pixel points; Wherein, the spatial distance weight feature is the reciprocal of the actual geographical distance between the n first neighboring pixel points and the corresponding pixel points of the high-resolution fire image to be predicted.
3. The method according to claim 1, wherein The high-resolution spatial geographic image includes a high-resolution normalized difference vegetation index (NDVI) image, a surface vegetation type image, and a digital elevation model (DEM) image; the extracting the spatial geographic features of n second neighboring pixel points based on the high-resolution spatial geographic image and the high-resolution fire image to be predicted that are pixel-level geographically matched includes: Based on the high-resolution spatial geographic image and the high-resolution fire image to be predicted that are pixel-level geographically matched, extract the geographical location information, elevation information, vegetation distribution type, and NDVI of n second neighboring pixel points.
4. The method according to claim 1, wherein The method further includes: Obtain sample data, the sample data includes multiple samples, and each sample includes the actually collected high-resolution fire image of the sample area and the fire point image features of n first neighboring pixel points corresponding to each pixel point in the low-resolution fire image, and the spatial geographic features of n second neighboring pixel points corresponding to each pixel point in the high-resolution spatial geographic image; Use the sample data to train the prediction model, and the prediction model includes a gradient boosting tree (GBT) model.
5. The method according to claim 4, characterized in that, The using the sample data to train the prediction model includes: The K-fold cross-validation algorithm is used to perform multiple splits and evaluations on the samples for parameter optimization in the sample data. Meanwhile, grid search is combined to optimize the hyperparameters of the prediction model, and an initial prediction model composed of the optimized hyperparameters is obtained. The initial prediction model is trained using the training samples in the sample data to obtain the prediction model.
6. The method according to claim 1, wherein The method further includes: A spatial smoothing algorithm is used to smooth the predicted high-resolution fire image to obtain the final high-resolution fire image.
7. A method for training a prediction model, characterized in that, It includes: Obtain the low-resolution fire image, high-resolution spatial geographical image, and actual high-resolution fire image of the sample area. Both the low-resolution fire image and the high-resolution spatial geographical image of the sample area are subjected to pixel-level geographical location matching with the actual high-resolution fire image. Based on the low-resolution fire image and the actual high-resolution fire image of the sample area with pixel-level geographical location matching, the fire point image features of n first neighboring pixels are extracted, where the n first neighboring pixels are the n first neighboring pixels closest to each pixel in the actual high-resolution fire image on the corresponding low-resolution fire image, and n is an integer greater than 1. Based on the high-resolution spatial geographical image and the actual high-resolution fire image of the sample area with pixel-level geographical location matching, the fire point image features of n second neighboring pixels are extracted, where the n second neighboring pixels are the n second neighboring pixels closest to each pixel in the actual high-resolution fire image on the corresponding high-resolution spatial geographical image. Obtain sample data, where the samples in the sample data include the actual high-resolution fire image of the sample area and the fire point image features of n first neighboring pixels corresponding to each pixel in the low-resolution fire image, and the spatial geographical features of n second neighboring pixels corresponding to each pixel in the high-resolution spatial geographical image. The prediction model is trained using the sample data.
8. A prediction device for high-resolution fire images, characterized in that, It includes: An image acquisition module configured to obtain the low-resolution fire image and the high-resolution spatial geographical image of the target area. A matching module configured to perform pixel-level geographical location matching on both the low-resolution fire image and the high-resolution spatial geographical image with the high-resolution fire image to be predicted. A first feature extraction module configured to extract the fire point image features of n first neighboring pixels based on the low-resolution fire image and the high-resolution fire image to be predicted with pixel-level geographical location matching, where the n first neighboring pixels are the n first neighboring pixels closest to each pixel in the high-resolution fire image to be predicted on the low-resolution fire image, and n is an integer greater than 1. A second feature extraction module configured to extract the spatial geographical features of n second neighboring pixels based on the high-resolution spatial geographical image and the high-resolution fire image to be predicted with pixel-level geographical location matching, where the n second neighboring pixels are the n second neighboring pixels closest to each pixel in the high-resolution fire image to be predicted on the high-resolution spatial geographical image. A prediction module, configured to input the fire point image features of the extracted n first neighboring pixel points and the spatial geographical features of the n second neighboring pixel points into a pre-trained prediction model, so as to obtain a high-resolution fire image predicted by the pre-trained prediction model.
9. An electronic device, characterized in that, It includes a memory and a processor, and the memory is used to store one or more computer instructions, wherein the one or more computer instructions are executed by the processor to implement the method according to any one of claims 1 to 7.
10. A readable storage medium, characterized in that, Computer instructions are stored thereon, and when the computer instructions are executed by a processor, the method according to any one of claims 1 to 7 is implemented.