Satellite remote sensing smoke identification method and device, electronic equipment and storage medium

Abstract

The invention relates to the technical field of fire detection, and provides a satellite remote sensing smoke identification method and device, electronic equipment and a storage medium, and the method comprises the steps: obtaining a satellite remote sensing image of a monitoring region, and extracting surface reflectance data from the satellite remote sensing image; performing spectral feature analysis on the surface reflectance data to obtain a plurality of spectral feature indexes; and inputting the surface reflectance data and each spectral characteristic index into a smoke identification model to obtain a smoke identification result output by the smoke identification model. According to the invention, by acquiring the satellite remote sensing image of the monitoring area, smog identification can be carried out in multiple directions. In addition, smoke recognition is performed in combination with the surface reflectance data and the spectral characteristic indexes, and real-time and high-precision smoke recognition can be realized.

Description

Technical Field

[0001] The present invention relates to the field of fire detection technology, and in particular to a satellite remote sensing smoke recognition method, device, electronic equipment and storage medium. Background Art

[0002] Smoke is one of the important phenomena produced by explosions and combustion. In natural disasters or man-made accidents, fire is one of the most common combustion phenomena. The smoke produced by fire not only contains gas and particulate matter, but also carries a large amount of heat energy and chemical components. The generation of smoke is mainly caused by material changes and chemical reactions during the combustion process. Fire smoke usually contains incompletely burned fuel, gas, solid particles, water vapor, carbon dioxide and other substances, which may also include some toxic gases and chemicals, such as carbon monoxide, nitrogen oxides, benzene, formaldehyde, etc. The diffusion and deposition of these particles and gases in the smoke in the air will not only affect the air quality, but may also have serious impacts on human health, the environment and the ecosystem.

[0003] Traditional satellite remote sensing smoke recognition methods are mainly based on image processing technology, such as threshold segmentation, texture feature extraction, morphological operations, etc. Although this method can detect smoke to a certain extent, it is prone to false detection or missed detection for complex smoke scenes, such as the overlap of smoke and clouds, the irregular shape of smoke, etc., and the computational complexity of this method is high, which makes it difficult to meet the needs of real-time smoke recognition. Summary of the invention

[0004] The present invention provides a satellite remote sensing smoke identification method, device, electronic equipment and storage medium to solve the defects in the prior art.

[0005] The present invention provides a satellite remote sensing smoke identification method, comprising the following steps: Obtaining satellite remote sensing images of the monitoring area, and extracting surface reflectivity data from the satellite remote sensing images; Performing spectral characteristic analysis on the surface reflectance data to obtain a plurality of spectral characteristic indexes; Inputting the surface reflectance data and each spectral characteristic index into a smoke recognition model to obtain a smoke recognition result output by the smoke recognition model; The smoke recognition model is trained based on sample surface reflectance data, sample spectral characteristic index and sample smoke recognition results, and the sample surface reflectance data is extracted from sample satellite remote sensing images.

[0006] According to a satellite remote sensing smoke recognition method provided by the present invention, the smoke recognition model is trained based on the following steps: Obtain sample satellite remote sensing images from multiple sources, and mark sample smoke identification results of each sample satellite remote sensing image; The corresponding sample surface reflectance data is extracted from each sample satellite remote sensing image; Perform spectral feature analysis on the surface reflectance data of each sample to obtain multiple candidate spectral feature indices; Based on the correlation between every two candidate spectral characteristic indices, determining a corresponding sample spectral characteristic index from a plurality of candidate spectral characteristic indices corresponding to each sample surface reflectance data; Inputting the surface reflectance data of each sample and the corresponding sample spectral characteristic index into the initial model to obtain the sample smoke prediction result output by the initial model; Based on the difference between the sample smoke prediction result and the sample smoke recognition result, the parameters of the initial model are optimized to obtain the smoke recognition model.

[0007] According to a satellite remote sensing smoke identification method provided by the present invention, the step of extracting corresponding sample surface reflectance data from each sample satellite remote sensing image comprises: After converting the projection coordinate system in each sample satellite remote sensing image into a standard coordinate system, remote sensing preprocessing is performed on each sample satellite remote sensing image to obtain the initial surface reflectance data of each sample; Based on the cloud mask band in the initial surface reflectance data of each sample, the cloud data in the initial surface reflectance data of each sample is removed to obtain the corresponding sample surface reflectance data.

[0008] According to a satellite remote sensing smoke identification method provided by the present invention, the remote sensing preprocessing of each sample satellite remote sensing image includes: Each sample satellite remote sensing image is sequentially subjected to geometric correction processing, radiation correction processing and other preprocessing, wherein the other preprocessing includes at least one of band extraction processing, region clipping processing, band synthesis processing and image registration processing.

[0009] According to a satellite remote sensing smoke recognition method provided by the present invention, the parameters of the initial model are optimized based on the difference between the sample smoke prediction result and the sample smoke recognition result to obtain the smoke recognition model, including: Determining the value range of each parameter in the initial model based on the difference between the sample smoke prediction result and the sample smoke recognition result; Based on the value range of each parameter, the combination of each parameter is exhaustively enumerated to determine multiple candidate parameter combinations; Testing each candidate parameter combination, and selecting an optimal parameter combination from the plurality of candidate parameter combinations based on the test results; Based on the optimal parameter combination, the parameters of the initial model are optimized to obtain the smoke recognition model.

[0010] According to a satellite remote sensing smoke identification method provided by the present invention, the test of each candidate parameter combination and the selection of the optimal parameter combination from the multiple candidate parameter combinations based on the test results include: Under each candidate parameter combination, predicting the function value of the objective function; the objective function is constructed based on the accuracy index of the initial model; Based on the function values ​​corresponding to the candidate parameter combinations, an optimal parameter combination is selected from the multiple candidate parameter combinations.

[0011] According to a satellite remote sensing smoke identification method provided by the present invention, the multiple spectral characteristic indexes include at least one of a normalized vegetation index, an enhanced vegetation index, a burned area index, a global environmental monitoring index and a normalized water body index.

[0012] The present invention also provides a satellite remote sensing smoke identification device, comprising the following modules: An acquisition unit, used to acquire satellite remote sensing images of the monitoring area and extract surface reflectivity data from the satellite remote sensing images; An analysis unit, used for performing spectral characteristic analysis on the surface reflectance data to obtain a plurality of spectral characteristic indexes; An identification unit, used for inputting the surface reflectance data and each spectral characteristic index into a smoke identification model to obtain a smoke identification result output by the smoke identification model; The smoke recognition model is trained based on sample surface reflectance data, sample spectral characteristic index and sample smoke recognition results, and the sample surface reflectance data is extracted from sample satellite remote sensing images.

[0013] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, any of the satellite remote sensing smoke identification methods described above is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium on which a computer program is stored. When the computer program is executed by a processor, the satellite remote sensing smoke identification method as described above is implemented.

[0015] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the satellite remote sensing smoke identification method as described above is implemented.

[0016] The satellite remote sensing smoke identification method, device, electronic device and storage medium provided by the present invention can perform smoke identification in multiple directions by acquiring satellite remote sensing images of the monitoring area. In addition, the embodiment of the present invention combines surface reflectivity data and various spectral characteristic indexes for smoke identification, thereby achieving real-time and high-precision smoke identification. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0018] Figure 1 This is one of the flow charts of the satellite remote sensing smoke identification method provided by the present invention.

[0019] Figure 2 This is the second flow chart of the satellite remote sensing smoke identification method provided by the present invention.

[0020] Figure 3 It is a structural schematic diagram of the satellite remote sensing smoke identification device provided by the present invention.

[0021] Figure 4 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0022] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] Traditional satellite remote sensing smoke recognition methods are mainly based on image processing technology, such as threshold segmentation, texture feature extraction, morphological operations, etc. Although this method can detect smoke to a certain extent, it is prone to false detection or missed detection for complex smoke scenes, such as the overlap of smoke and clouds, the irregular shape of smoke, etc., and the computational complexity of this method is high, which makes it difficult to meet the needs of real-time smoke recognition.

[0024] The present invention considers the use of satellite remote sensing and machine learning technology for fire smoke monitoring and identification, which can not only help to timely detect and control fires, reduce casualties and property losses caused by fires, but also provide important environmental monitoring and emergency management support. By analyzing the characteristics of fire smoke in remote sensing images, such as color, shape, texture, etc., combined with machine learning algorithms for automatic identification and classification, it is possible to better achieve effective monitoring and early warning of smoke generated by fires and explosions.

[0025] Figure 1 FIG. 1 is one of the flow charts of the satellite remote sensing smoke identification method provided by the present invention, such as Figure 1 As shown, the method includes step 110 , step 120 and step 130 .

[0026] Step 110: Acquire satellite remote sensing images of the monitoring area, and extract surface reflectivity data from the satellite remote sensing images.

[0027] Here, the monitoring area can be understood as the area where smoke identification is required. Satellite remote sensing images are images or data of the earth's surface collected by sensors on satellites. Satellite remote sensing images usually include multiple spectral bands that can capture light of different wavelengths. The reflection and transmission characteristics of smoke in different spectral bands are different from those of other substances on the ground (such as water bodies, vegetation, buildings, etc.). Therefore, smoke and other substances can be distinguished by analyzing these spectral characteristics. Among them, satellite remote sensing images can be obtained from multiple sources, such as environmental satellite remote sensing data, high-resolution satellite remote sensing data, sentinel satellite remote sensing data, land satellite remote sensing data, etc.

[0028] In addition, satellite remote sensing images can cover a wide area, so that smoke identification can be carried out over a large range. That is, even when the monitoring area is large, smoke identification can be accurately performed based on satellite remote sensing images.

[0029] The surface reflectance data is the intensity of light reflected by surface objects in a satellite remote sensing image in a specific band, which can reflect the spectral characteristics of surface objects in a certain band in the satellite remote sensing image. Optionally, the surface reflectance data can be extracted after performing radiation correction and atmospheric correction on the satellite remote sensing image in sequence.

[0030] Step 120: Perform spectral feature analysis on the surface reflectance data to obtain a plurality of spectral feature indices.

[0031] Specifically, the spectral characteristic index is a value calculated from the surface reflectance data, which is used to quantify the spectral characteristics of surface objects in different bands, thereby highlighting the surface characteristics of different surface objects.

[0032] Taking into account that smoke can scatter and absorb light, it has different effects on different spectral bands. In other words, the presence of smoke will change the spectral characteristics of satellite remote sensing images, making the reflectivity or radiation values ​​of certain bands different from normal conditions. The spectral characteristic index obtained based on the spectral characteristic analysis of the surface reflectance data can effectively distinguish between smoke and other substances, such as clouds, vegetation or water bodies.

[0033] It should be noted that, considering that different spectral feature indices utilize spectral features between different bands to highlight specific surface objects, that is, different spectral feature indices have different sensitivities to different bands, the embodiment of the present invention combines multiple different spectral feature indices to analyze spectral features from different angles, that is, different spectral feature indices provide complementary information, enhance the contrast of smoke, and thus accurately identify smoke.

[0034] For example, within a spectral range, smoke may have a more obvious change in reflectance in one band, but not in another band. By combining multiple spectral feature indices, these changes can be captured more effectively, thereby enhancing the contrast of smoke and thus improving the accuracy of smoke identification.

[0035] Step 130: input the surface reflectance data and each spectral characteristic index into the smoke recognition model to obtain the smoke recognition result output by the smoke recognition model; The smoke recognition model is trained based on sample surface reflectance data, sample spectral characteristic index and sample smoke recognition results. The sample surface reflectance data is extracted from sample satellite remote sensing images.

[0036] Specifically, the surface reflectance data provides basic spectral information of the surface, while the spectral characteristic index emphasizes the impact of smoke on spectral information by calculating the combination of specific bands. Combining the two can provide a more comprehensive perspective and thus more accurately identify smoke.

[0037] Since smoke may have a small effect on the reflectivity of some bands, especially at low concentrations, this small change may not be obvious enough in the surface reflectivity data, making it difficult to accurately distinguish the presence of smoke. In other words, relying solely on surface reflectivity data for smoke identification may lead to misidentification.

[0038] In urban area monitoring, the apparent reflectance may be affected by buildings, roads and different materials, which may mask the weak signal of smoke. If only the reflectance data of the visible light band is relied upon, it may not fully reflect the actual presence of smoke.

[0039] Therefore, the embodiment of the present invention combines the surface reflectance data and each spectral characteristic index to perform smoke recognition, that is, the surface reflectance data and each spectral characteristic index are input into the smoke recognition model to obtain the smoke recognition result output by the smoke recognition model. The smoke recognition result is used to characterize whether there is smoke in the monitoring area.

[0040] In addition, the smoke recognition model is trained based on sample surface reflectance data, sample spectral characteristic index and sample smoke recognition results, so that the smoke recognition model can learn the characteristic patterns of smoke under different spectral characteristics (such as the reflectance changes of smoke in different bands, the differences in spectral characteristics, etc.), and thus can accurately identify smoke.

[0041] Optionally, corresponding weights may be set for the surface reflectance data and each spectral characteristic index respectively, and after the weights of the surface reflectance data and each spectral characteristic index are added, they are input into the smoke recognition model to obtain the smoke recognition result output by the smoke recognition model.

[0042] The satellite remote sensing smoke identification method provided by the embodiment of the present invention can perform smoke identification in multiple directions by acquiring satellite remote sensing images of the monitoring area. In addition, the embodiment of the present invention combines surface reflectivity data and various spectral characteristic indexes for smoke identification, thereby achieving real-time and high-precision smoke identification.

[0043] Based on the above embodiment, the smoke recognition model is trained based on the following steps: Obtain sample satellite remote sensing images from multiple sources, and mark sample smoke identification results of each sample satellite remote sensing image; The corresponding sample surface reflectance data is extracted from each sample satellite remote sensing image; Perform spectral feature analysis on the surface reflectance data of each sample to obtain multiple candidate spectral feature indices; Based on the correlation between every two candidate spectral characteristic indices, determining a corresponding sample spectral characteristic index from a plurality of candidate spectral characteristic indices corresponding to each sample surface reflectance data; Input the surface reflectance data of each sample and the corresponding sample spectral characteristic index into the initial model to obtain the sample smoke prediction result output by the initial model; Based on the difference between the sample smoke prediction results and the sample smoke recognition results, the parameters of the initial model are optimized to obtain the smoke recognition model.

[0044] Specifically, the sample satellite remote sensing images from multiple sources refer to remote sensing image data obtained from different satellites or data providers, which may come from different satellite platforms, different sensors, different time points or different geographical areas. The embodiments of the present invention obtain sample satellite remote sensing images from multiple sources, which can increase the diversity and representativeness of the samples, thereby improving the robustness and accuracy of the model.

[0045] After obtaining sample satellite remote sensing images from multiple sources, the sample smoke identification result of each sample satellite remote sensing image can be marked, and the sample smoke identification result is used to characterize whether smoke actually exists in the corresponding sample satellite remote sensing image.

[0046] The sample surface reflectance data is the intensity of light reflected by the surface objects in the sample satellite remote sensing image in a specific band, which can reflect the spectral characteristics of the surface objects in the sample satellite remote sensing image in a certain band. Optionally, the sample surface reflectance data can be extracted after performing radiation correction and atmospheric correction on the sample satellite remote sensing image in sequence.

[0047] After obtaining the sample surface reflectance data, the spectral characteristics of each sample surface reflectance data are analyzed to obtain multiple candidate spectral characteristic indexes. Considering that there may be highly correlated indexes among the multiple candidate spectral characteristic indexes, the highly correlated indexes may contain repeated information (i.e., redundant information). If the model is trained based on the highly correlated indexes, the trained model will be at risk of overfitting, thereby reducing the model performance.

[0048] In this regard, the embodiment of the present invention determines the corresponding sample spectral feature index from the multiple candidate spectral feature indexes corresponding to the surface reflectance data of each sample, thereby ensuring low correlation between the sample spectral feature indexes, thereby reducing the repeated information between the sample spectral feature indexes, avoiding model overfitting, and improving model performance. Among them, the correlation between every two candidate spectral feature indexes is used to characterize the similarity between every two candidate spectral feature indexes. The higher the similarity, the more repeated information contained in the corresponding two candidate spectral feature indexes, and the higher the correlation between the two.

[0049] Optionally, the Pearson correlation coefficient between each two candidate spectral feature indices can be used to measure the correlation between each two candidate spectral feature indices. The larger the Pearson correlation coefficient, the greater the correlation between each two candidate spectral feature indices. The embodiment of the present invention can set a threshold (such as 0.8). If the Pearson correlation coefficient between each two candidate spectral feature indices is greater than 0.8, it indicates that there is a lot of repeated information between the corresponding two candidate spectral feature indices, and any one of them can be eliminated, and finally the retained candidate spectral feature index is used as the sample spectral feature index.

[0050] Next, the surface reflectance data of each sample and the corresponding sample spectral feature index are input into the initial model to obtain the sample smoke prediction result output by the initial model. The difference between the sample smoke prediction result and the sample smoke recognition result can determine the loss function of the model, and then the parameters of the initial model can be optimized based on the loss function to obtain the smoke recognition model. Among them, the initial model can be constructed based on the XGBoost (Extreme Gradient Boosting) model, which is an efficient gradient boosting tree algorithm for classification and regression tasks. It improves the accuracy of the model by building multiple decision trees and combining their predictions, and has fast training speed and strong generalization ability. XGBoost also includes support for feature selection, regularization and other technologies to reduce overfitting and improve model performance. In addition, the parameters of the initial model can be optimized using Bayesian optimization through the scikit-learn software package.

[0051] Based on any of the above embodiments, extracting corresponding sample surface reflectance data from each sample satellite remote sensing image includes: After converting the projection coordinate system in each sample satellite remote sensing image into a standard coordinate system, remote sensing preprocessing is performed on each sample satellite remote sensing image to obtain the initial surface reflectance data of each sample; Based on the cloud mask band in the initial surface reflectance data of each sample, the cloud data in the initial surface reflectance data of each sample is removed to obtain the corresponding sample surface reflectance data.

[0052] Specifically, the projected coordinate system is a coordinate system that converts the three-dimensional coordinates (longitude and latitude) on the earth's surface into a two-dimensional plane. The standard coordinate system usually refers to the geographic coordinate system (such as WGS 84), which expresses the position in longitude and latitude. The purpose of converting the projected coordinate system in each sample satellite remote sensing image to the standard coordinate system is to ensure the consistency of sample satellite remote sensing images from different sources in geographic space, so as to facilitate comparative analysis.

[0053] After the coordinate system conversion, remote sensing preprocessing is performed on each sample satellite remote sensing image to obtain the initial surface reflectance data of each sample. Considering that the initial surface reflectance data of each sample may be covered by cloud pollution, the embodiment of the present invention removes the cloud data in the initial surface reflectance data of each sample based on the cloud mask band in the initial surface reflectance data of each sample to obtain the corresponding sample surface reflectance data. Among them, the cloud mask band is a band used to identify the presence of clouds in the initial surface reflectance data of each sample. The cloud mask band is usually a binary image, and the value of each pixel in the image indicates whether the position is covered by clouds.

[0054] For example, for the sample satellite remote sensing images Landsat-8 / 9, the fourth bit of the QA (Quality Assurance) band can be used for logical operations to determine whether the surface is covered by clouds and perform cloud removal. For the sample satellite remote sensing images Sentinel-2, the QA60 band of Sentinel-2 can be used to remove clouds from the image and obtain a cloud-free image of Sentinel-2 data.

[0055] Based on any of the above embodiments, remote sensing preprocessing is performed on each sample satellite remote sensing image, including: Each sample satellite remote sensing image is sequentially subjected to geometric correction processing, radiation correction processing and other preprocessing, and the other preprocessing includes at least one of band extraction processing, region clipping processing, band synthesis processing and image registration processing.

[0056] Specifically, geometric correction ensures that the objects in the sample satellite remote sensing images are correctly aligned in the geographic coordinate system. Performing geometric correction first is the basis for subsequent analysis and can avoid analysis bias caused by position errors. Radiometric correction converts the raw data of the sample satellite remote sensing images into standard reflectivity or brightness values, which can eliminate radiation errors caused by factors such as the atmosphere and sensors, and ensure the true comparability of the sample satellite remote sensing images.

[0057] After geometric correction and radiation correction, the spatial and radiation characteristics of the sample satellite remote sensing images have been corrected. At this time, preprocessing such as band extraction, region clipping, band synthesis or image registration will be more accurate and efficient. In other words, geometric and radiation correction can ensure the accuracy and consistency of the sample satellite remote sensing images, lay a solid foundation for other subsequent preprocessing operations, and thus improve the quality and efficiency of the entire data processing process.

[0058] Among them, the band extraction processing is to extract specific bands from multi-spectral sample satellite remote sensing images, the regional cropping processing is to crop specific areas of interest from large-scale sample satellite remote sensing images, the band synthesis processing is to combine data from multiple bands into a composite image, and the image registration processing is to align multi-source sample satellite remote sensing images so that they overlap accurately in space.

[0059] Based on any of the above embodiments, based on the difference between the sample smoke prediction result and the sample smoke recognition result, the parameters of the initial model are optimized to obtain a smoke recognition model, including: Based on the difference between the sample smoke prediction result and the sample smoke recognition result, the value range of each parameter in the initial model is determined; Based on the value range of each parameter, the combination of each parameter is exhaustively enumerated to determine multiple candidate parameter combinations; Testing each candidate parameter combination, and selecting an optimal parameter combination from multiple candidate parameter combinations based on the test results; Based on the optimal parameter combination, the parameters of the initial model are optimized to obtain the smoke recognition model.

[0060] Specifically, the parameters of the model (such as learning rate, regularization parameter, tree depth, etc.) have a significant impact on the performance of the model. Considering that the initial model may not achieve the best performance, it is necessary to determine the value range of each parameter in the initial model for further optimization.

[0061] Among them, by comparing the differences between the sample smoke prediction results and the actual recognition results, it is possible to identify which parameters in the initial model need to be adjusted and the value range of each parameter.

[0062] After determining the value range of each parameter, it is necessary to test different candidate parameter combinations to find the optimal setting (i.e., the optimal parameter combination). In this case, an exhaustive search or grid search method can be used to exhaustively search the combinations of each parameter to determine multiple candidate parameter combinations.

[0063] After obtaining multiple candidate parameter combinations, since different candidate parameter combinations have different effects on model performance, it is necessary to test each candidate parameter combination, evaluate the effect of each candidate parameter combination, and select the optimal parameter combination. Among them, cross-validation (such as ten-fold cross-validation) and other techniques can be used to evaluate each candidate parameter combination, and by comparing the performance indicators of different parameter combinations (such as accuracy, recall rate, F1 score, etc.), the parameter combination with the best performance (such as the best accuracy) is selected as the optimal parameter combination.

[0064] After determining the optimal parameter combination, the optimal parameter combination can be applied to the initial model for model training and optimization, and finally an optimized smoke recognition model is obtained. Table 1 is the value range of the XGBoost model parameters and the optimal value list of Bayesian optimization. As shown in Table 1, the combination of the optimal values ​​corresponding to each parameter is the optimal parameter combination.

[0065] Table 1

[0066] Based on any of the above embodiments, each candidate parameter combination is tested, and an optimal parameter combination is selected from multiple candidate parameter combinations based on the test results, including: Under each candidate parameter combination, predict the function value of the objective function; the objective function is constructed based on the accuracy index of the initial model; Based on the function values ​​corresponding to each candidate parameter combination, an optimal parameter combination is selected from multiple candidate parameter combinations.

[0067] Specifically, the objective function is constructed based on the accuracy index of the initial model, that is, the objective function is an indicator function that measures the performance of the model. Among them, the accuracy index can include accuracy, precision, recall rate, mean square error, etc.

[0068] After determining the objective function, the function value of the objective function is predicted under each candidate parameter combination. According to the function value of the objective function, the performance of the model can be determined, and the candidate parameter combination corresponding to the best performance is selected as the optimal parameter combination. The parameters of the initial model are optimized based on the optimal parameter combination, thereby improving the performance of the initial model.

[0069] Based on any of the above embodiments, the multiple spectral characteristic indices include at least one of a normalized vegetation index, an enhanced vegetation index, a burned area index, a global environmental monitoring index, and a normalized water index.

[0070] Specifically, the Normalized Difference Vegetation Index, Enhanced Vegetation Index, Burned Area Index, Global Environmental Monitoring Index and Normalized Difference Water Index can effectively distinguish smoke, vegetation, clouds, water, non-forest land, cloud shadows and burnt areas. Among them, the Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI), Burned Area Index (BAI), Global Environmental Monitoring Index (GEMI) and Normalized Difference Water Index (NDWI) are calculated based on the following formulas:

[0071] in, represents the surface reflectivity in the blue light band, represents the surface reflectance in the green light band, represents the surface reflectivity in the red light band, Represents the surface reflectance in the near-infrared band.

[0072] Based on any of the above embodiments, Figure 2 FIG. 2 is a flow chart of the satellite remote sensing smoke identification method provided by the present invention. Figure 2 As shown, the method includes: First, obtain the satellite remote sensing images of the monitoring area, perform remote sensing preprocessing on the satellite remote sensing images, and extract the surface reflectance data. De-clouding is performed through the cloud mask band in the surface reflectance data. Specifically, it is to identify whether each pixel in the cloud mask band is a cloud pixel (that is, there is cloud in the pixel). If so, mark the pixel value Cloud=0. If not, obtain the de-clouded data.

[0073] Next, the spectral feature analysis of the de-clouding data is performed to obtain multiple spectral feature indexes. The de-clouding data and each spectral feature index are input into the smoke recognition model for smoke recognition, and the smoke recognition result output by the smoke recognition model is obtained. After the smoke recognition result is obtained, the smoke recognition result is verified. If the verification result is correct, the final recognition result is output.

[0074] The smoke recognition model is obtained by training the machine learning model based on the training data set, and the smoke recognition model is an XGBoost model. The training data set is obtained from the sample data set, which is constructed based on multi-source sample satellite remote sensing images. The sample data set includes non-smog pixels and smoke pixels. In addition, the spectral feature index includes the normalized vegetation index, enhanced vegetation index, burned area index, global environmental monitoring index and normalized water index.

[0075] The embodiment of the present invention combines the surface reflectance data of multi-source satellite remote sensing images with the remote sensing feature index, and uses the XGBoost model to perform smoke pixel-level classification, which effectively solves the problems of low accuracy and poor real-time performance of traditional smoke identification methods. The model structure is simple and has good generalization and practicality.

[0076] In addition, Table 2 is a list of smoke recognition model accuracy indicators. As shown in Table 2, the overall accuracy of the model can reach 96.16%, and the Kappa coefficient can reach 0.89.

[0077] Table 2

[0078] The above accuracy index can be determined based on the following method: Confusion Matrix :The confusion matrix method commonly used in land object classification can be used to evaluate the accuracy of sub-pixel positioning. This method is also called the error matrix. The number of correctly classified and incorrectly classified sub-pixels in the sub-pixel positioning results is filled into the square matrix according to the end members. The expression is as follows:

[0079] In the formula, c is the category code, mij represents the number of sub-pixels that belong to category i in the real situation, but the classification result belongs to category j. Therefore, when the classification is correct, the result is arranged on the main diagonal, and the misclassified sub-pixels are arranged on both sides of the main diagonal. The larger the element value on the main diagonal, the more accurate the classification and the higher the classification accuracy; the larger the element value on both sides of the main diagonal, the more serious the misclassification and the lower the classification accuracy.

[0080] Ovaerall Accuracy (OA): The overall classification accuracy is equal to the proportion of the number of correctly classified sub-pixels in the total number of sub-pixels involved in the classification. In the confusion matrix, it represents the ratio of the sum of the elements on the main diagonal to the sum of all elements in the square matrix. It is used to measure the overall classification of the model. The expression is as follows:

[0081] Producer Accuracy (PA): Producer accuracy represents the ratio of the number of correctly classified sub-pixels of a certain category to the total number of sub-pixels of that category in the actual situation. In the confusion matrix, it represents the ratio of the value of an element on the main diagonal to the sum of the elements in the row where the element is located. The expression is as follows:

[0082] User Accuracy (UA): User accuracy represents the ratio of the number of correctly classified sub-pixels of a certain category to the total number of sub-pixels belonging to this category in the sub-pixel mapping result. In the confusion matrix, it represents the ratio of a certain element value on the main diagonal to the sum of the elements in the column where the element value is located. The expression is as follows:

[0083] Kappa coefficient: The Kappa coefficient obtained based on the confusion matrix is ​​used to check the consistency between the sub-pixel mapping results and the actual situation. The value range is [−1,1], which is usually greater than 0. The larger the value, the higher the consistency. This indicator can not only evaluate the accuracy of classification, but also is an important indicator to measure the balance of the confusion matrix. It is generally believed that when the Kappa coefficient is greater than 0.6, it has a high degree of consistency. The calculation formula is as follows:

[0084] in, Represents the number of correctly classified sub-pixels and the total number of sub-pixels, that is, the overall classification accuracy; It represents the sum of the product of the number of real sub-pixels in each category and the number of classified sub-pixels in the corresponding category, divided by the square of the total number of sub-pixels involved in the classification. The expressions of the two are as follows:

[0085] The satellite remote sensing smoke identification device provided by the present invention is described below. The satellite remote sensing smoke identification device described below and the satellite remote sensing smoke identification method described above can be referenced to each other.

[0086] Based on any of the above embodiments, Figure 3 FIG. 1 is a schematic diagram of the structure of the satellite remote sensing smoke identification device provided by the present invention. Figure 3 As shown, the device comprises: An acquisition unit 310 is used to acquire satellite remote sensing images of the monitoring area and extract surface reflectivity data from the satellite remote sensing images; An analysis unit 320 is used to perform spectral characteristic analysis on the surface reflectance data to obtain a plurality of spectral characteristic indexes; The recognition unit 330 is used to input the surface reflectance data and each spectral characteristic index into the smoke recognition model to obtain the smoke recognition result output by the smoke recognition model; The smoke recognition model is trained based on sample surface reflectance data, sample spectral characteristic index and sample smoke recognition results. The sample surface reflectance data is extracted from sample satellite remote sensing images.

[0087] Based on any of the above embodiments, the smoke recognition model is trained based on the following steps: Obtain sample satellite remote sensing images from multiple sources, and mark sample smoke identification results of each sample satellite remote sensing image; The corresponding sample surface reflectance data is extracted from each sample satellite remote sensing image; Perform spectral feature analysis on the surface reflectance data of each sample to obtain multiple candidate spectral feature indices; Based on the correlation between every two candidate spectral characteristic indices, determining a corresponding sample spectral characteristic index from a plurality of candidate spectral characteristic indices corresponding to each sample surface reflectance data; Input the surface reflectance data of each sample and the corresponding sample spectral characteristic index into the initial model to obtain the sample smoke prediction result output by the initial model; Based on the difference between the sample smoke prediction results and the sample smoke recognition results, the parameters of the initial model are optimized to obtain the smoke recognition model.

[0088] Based on any of the above embodiments, extracting corresponding sample surface reflectance data from each sample satellite remote sensing image includes: After converting the projection coordinate system in each sample satellite remote sensing image into a standard coordinate system, remote sensing preprocessing is performed on each sample satellite remote sensing image to obtain the initial surface reflectance data of each sample; Based on the cloud mask band in the initial surface reflectance data of each sample, the cloud data in the initial surface reflectance data of each sample is removed to obtain the corresponding sample surface reflectance data.

[0089] Based on any of the above embodiments, remote sensing preprocessing is performed on each sample satellite remote sensing image, including: Each sample satellite remote sensing image is sequentially subjected to geometric correction processing, radiation correction processing and other preprocessing, and the other preprocessing includes at least one of band extraction processing, region clipping processing, band synthesis processing and image registration processing.

[0090] Based on any of the above embodiments, based on the difference between the sample smoke prediction result and the sample smoke recognition result, the parameters of the initial model are optimized to obtain a smoke recognition model, including: Based on the difference between the sample smoke prediction result and the sample smoke recognition result, the value range of each parameter in the initial model is determined; Based on the value range of each parameter, the combination of each parameter is exhaustively enumerated to determine multiple candidate parameter combinations; Testing each candidate parameter combination, and selecting an optimal parameter combination from multiple candidate parameter combinations based on the test results; Based on the optimal parameter combination, the parameters of the initial model are optimized to obtain the smoke recognition model.

[0091] Based on any of the above embodiments, each candidate parameter combination is tested, and an optimal parameter combination is selected from multiple candidate parameter combinations based on the test results, including: Under each candidate parameter combination, predict the function value of the objective function; the objective function is constructed based on the accuracy index of the initial model; Based on the function values ​​corresponding to each candidate parameter combination, an optimal parameter combination is selected from multiple candidate parameter combinations.

[0092] Based on any of the above embodiments, the multiple spectral characteristic indices include at least one of a normalized vegetation index, an enhanced vegetation index, a burned area index, a global environmental monitoring index, and a normalized water index.

[0093] Figure 4 is a schematic diagram of the structure of the electronic device provided by the present invention, such as Figure 4 As shown, the electronic device may include: a processor 410, a communication interface 420, a memory 430 and a communication bus 440, wherein the processor 410, the communication interface 420 and the memory 430 communicate with each other through the communication bus 440. The processor 410 may call the logic instructions in the memory 430 to execute the satellite remote sensing smoke recognition method, which includes: acquiring the satellite remote sensing image of the monitoring area, and extracting the surface reflectivity data from the satellite remote sensing image; performing spectral feature analysis on the surface reflectivity data to obtain a plurality of spectral feature indexes; inputting the surface reflectivity data and each spectral feature index into the smoke recognition model to obtain the smoke recognition result output by the smoke recognition model; the smoke recognition model is trained based on the sample surface reflectivity data, the sample spectral feature index and the sample smoke recognition result, and the sample surface reflectivity data is extracted from the sample satellite remote sensing image.

[0094] In addition, the logic instructions in the above-mentioned memory 430 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0095] On the other hand, the present invention also provides a computer program product, which includes a computer program, and the computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the satellite remote sensing smoke identification method provided by the above methods, and the method includes: obtaining satellite remote sensing images of the monitoring area, and extracting surface reflectivity data from the satellite remote sensing images; performing spectral feature analysis on the surface reflectivity data to obtain multiple spectral feature indexes; inputting the surface reflectivity data and each spectral feature index into a smoke identification model to obtain a smoke identification result output by the smoke identification model; the smoke identification model is trained based on sample surface reflectivity data, sample spectral feature indexes and sample smoke identification results, and the sample surface reflectivity data is extracted from sample satellite remote sensing images.

[0096] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the satellite remote sensing smoke identification method provided by the above-mentioned methods, the method comprising: acquiring satellite remote sensing images of the monitoring area, and extracting surface reflectivity data from the satellite remote sensing images; performing spectral feature analysis on the surface reflectivity data to obtain multiple spectral feature indexes; inputting the surface reflectivity data and each spectral feature index into a smoke identification model to obtain a smoke identification result output by the smoke identification model; the smoke identification model is trained based on sample surface reflectivity data, sample spectral feature indexes and sample smoke identification results, and the sample surface reflectivity data is extracted from sample satellite remote sensing images.

[0097] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0098] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0099] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A satellite remote sensing smoke identification method, characterized in that: include: Obtaining satellite remote sensing images of the monitoring area, and extracting surface reflectivity data from the satellite remote sensing images; Performing spectral characteristic analysis on the surface reflectance data to obtain a plurality of spectral characteristic indexes; Inputting the surface reflectance data and each spectral characteristic index into a smoke recognition model to obtain a smoke recognition result output by the smoke recognition model; The smoke recognition model is trained based on sample surface reflectance data, sample spectral characteristic index and sample smoke recognition results, and the sample surface reflectance data is extracted from sample satellite remote sensing images.

2. The satellite remote sensing smoke identification method according to claim 1, characterized in that: The smoke recognition model is trained based on the following steps: Obtain sample satellite remote sensing images from multiple sources, and mark sample smoke identification results of each sample satellite remote sensing image; The corresponding sample surface reflectance data is extracted from each sample satellite remote sensing image; Perform spectral feature analysis on the surface reflectance data of each sample to obtain multiple candidate spectral feature indices; Based on the correlation between every two candidate spectral characteristic indices, determining a corresponding sample spectral characteristic index from a plurality of candidate spectral characteristic indices corresponding to each sample surface reflectance data; Inputting the surface reflectance data of each sample and the corresponding sample spectral characteristic index into the initial model to obtain the sample smoke prediction result output by the initial model; Based on the difference between the sample smoke prediction result and the sample smoke recognition result, the parameters of the initial model are optimized to obtain the smoke recognition model.

3. The satellite remote sensing smoke identification method according to claim 2, characterized in that: The step of extracting corresponding sample surface reflectance data from each sample satellite remote sensing image includes: After converting the projection coordinate system in each sample satellite remote sensing image into a standard coordinate system, remote sensing preprocessing is performed on each sample satellite remote sensing image to obtain the initial surface reflectance data of each sample; Based on the cloud mask band in the initial surface reflectance data of each sample, the cloud data in the initial surface reflectance data of each sample is removed to obtain the corresponding sample surface reflectance data.

4. The satellite remote sensing smoke identification method according to claim 3, characterized in that: The remote sensing preprocessing of each sample satellite remote sensing image includes: Each sample satellite remote sensing image is sequentially subjected to geometric correction processing, radiation correction processing and other preprocessing, wherein the other preprocessing includes at least one of band extraction processing, region clipping processing, band synthesis processing and image registration processing.

5. The satellite remote sensing smoke identification method according to claim 2, characterized in that: The step of optimizing the parameters of the initial model based on the difference between the sample smoke prediction result and the sample smoke recognition result to obtain the smoke recognition model includes: Determining the value range of each parameter in the initial model based on the difference between the sample smoke prediction result and the sample smoke recognition result; Based on the value range of each parameter, the combination of each parameter is exhaustively enumerated to determine multiple candidate parameter combinations; Testing each candidate parameter combination, and selecting an optimal parameter combination from the plurality of candidate parameter combinations based on the test results; Based on the optimal parameter combination, the parameters of the initial model are optimized to obtain the smoke recognition model.

6. The satellite remote sensing smoke identification method according to claim 5, characterized in that: The step of testing each candidate parameter combination and selecting an optimal parameter combination from the plurality of candidate parameter combinations based on the test results includes: Under each candidate parameter combination, predicting the function value of the objective function; the objective function is constructed based on the accuracy index of the initial model; Based on the function values ​​corresponding to the candidate parameter combinations, an optimal parameter combination is selected from the multiple candidate parameter combinations.

7. The satellite remote sensing smoke identification method according to any one of claims 1 to 6, characterized in that: The multiple spectral characteristic indices include at least one of a normalized vegetation index, an enhanced vegetation index, a burned area index, a global environmental monitoring index and a normalized water body index.

8. A satellite remote sensing smoke identification device, characterized in that: include: An acquisition unit, used to acquire satellite remote sensing images of the monitoring area and extract surface reflectivity data from the satellite remote sensing images; An analysis unit, used for performing spectral characteristic analysis on the surface reflectance data to obtain a plurality of spectral characteristic indexes; An identification unit, used for inputting the surface reflectance data and each spectral characteristic index into a smoke identification model to obtain a smoke identification result output by the smoke identification model; The smoke recognition model is trained based on sample surface reflectance data, sample spectral characteristic index and sample smoke recognition results, and the sample surface reflectance data is extracted from sample satellite remote sensing images.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the satellite remote sensing smoke identification method according to any one of claims 1 to 7 is implemented.

10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the satellite remote sensing smoke identification method according to any one of claims 1 to 7 is implemented.

Images