Straw incineration flue gas remote sensing image recognition method and system
A remote sensing image recognition method using a cascade classifier with RF and DT models addresses the challenge of segmenting straw burning smoke from background and clouds, ensuring efficient and accurate identification for fire and air quality monitoring.
Patent Information
- Application Number
- CN202411134384.6
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-19
- Publication Date
- 2024-12-03
AI Technical Summary
Existing remote sensing technologies struggle to accurately and efficiently segment straw burning smoke from background and cloud interference, which is crucial for fire monitoring and air quality assessment due to the difficulty in distinguishing these elements from high-resolution satellite images.
A remote sensing image recognition method utilizing a cascade classifier combining a Random Forest (RF) model and a Decision Tree (DT) model, trained on spectral reflectance and mean texture features, to identify straw burning smoke by preprocessing images, extracting optimal band combinations, and employing a gray-level co-occurrence matrix for texture feature calculation.
The method achieves rapid and accurate segmentation of straw burning smoke, effectively removing background and cloud interference, providing valuable information for fire emergency management and air quality assessment.
Smart Images

Figure CN119068333A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of remote sensing images, and relates to a remote sensing image recognition method and system for straw burning flue gas. Background Art
[0002] Using high-resolution satellites to monitor changes in land objects is one of the main research directions of modern remote sensing. For monitoring emergencies, remote sensing has irreplaceable advantages in monitoring range and efficiency, especially in fire monitoring. Due to the suddenness, pollution and destructiveness of fires, accurate and timely monitoring of fires is particularly important.
[0003] Compared with conventional forest fires, fires caused by straw burning are often overlooked. Since mountain fires and forest fires caused by the spread of straw burning often occur, severe PM2.5 pollution often occurs during periods of concentrated straw burning. Therefore, it is necessary to monitor such uncontrolled straw burning incidents.
[0004] Compared with fire point monitoring, which is difficult to judge the fire intensity, the monitoring of smoke generated by combustion can better judge the size and spread of the fire, and the monitoring of smoke can also provide guidance for the local air quality evaluation. Compared with ground monitoring, although remote sensing has a wide monitoring range, the interference of ground objects and clouds on monitoring is a difficult problem to be solved. Some highlighted backgrounds and all clouds are difficult to distinguish at the remote sensing level. Therefore, how to achieve fast and accurate segmentation of straw burning smoke from background and clouds is an urgent problem to be solved. Summary of the invention
[0005] The purpose of the present invention is to provide a remote sensing image recognition method and system for straw burning smoke, which can accurately identify straw burning smoke through remote sensing images and efficiently remove interfering ground objects and clouds.
[0006] In order to achieve the above object, the basic scheme of the present invention is: a remote sensing image recognition method for straw burning flue gas, comprising the following steps:
[0007] Obtain remote sensing images of straw burning, remote sensing images without smoke interference background and remote sensing images with cloud interference, and perform preprocessing to obtain the spectral reflectance of each band of the image;
[0008] Using the gray-level co-occurrence matrix, the mean texture features corresponding to each band of the image are calculated;
[0009] Extracting cloud, smoke, background pixels, smoke objects and cloud objects from the preprocessed images;
[0010] Using the selected cloud, smoke, and background pixels, the spectral reflectance and mean texture features of the corresponding optimal band are extracted; using the selected smoke and cloud objects, the spectral reflectance and mean texture features of the corresponding optimal band are extracted;
[0011] The spectral reflectance and mean texture features extracted from cloud, smoke, and background pixels are used to train the RF model, and the spectral reflectance and mean texture features extracted from smoke objects and cloud objects are used to train the DT model;
[0012] The RF model and the DT model are combined into a cascade classifier to identify the straw burning remote sensing image and obtain the mask of straw burning smoke.
[0013] The working principle and beneficial effects of this basic scheme are: this technical scheme extracts the spectral reflectance characteristics of smoke, background, and clouds from the preprocessed images, screens out the best band combination, and uses texture features to supplement the image feature information, which is more helpful for the model to accurately and quickly segment straw burning smoke, background, and clouds.
[0014] The two models are used to construct a cascade classifier, which simplifies the classification steps of the model and enables each model to play a greater role. It realizes the automatic recognition of remote sensing images of straw burning smoke, accurately and quickly identifies the background and cloud layer of straw burning smoke segmentation, and provides valuable information for fire emergency management and air quality assessment.
[0015] Furthermore, the method for preprocessing the straw burning remote sensing images, the smoke-free background remote sensing images and the cloud interference remote sensing images is: performing radiation calibration, atmospheric correction and geometric correction on the remote sensing images.
[0016] Pre-process the straw burning remote sensing images, smoke-free background remote sensing images and cloud interference remote sensing images for subsequent use.
[0017] Furthermore, the gray-level co-occurrence matrix is used to calculate the mean texture features corresponding to each band of the image, which is:
[0018]
[0019] Among them, M and N refer to the image size, i and j refer to the pixel positions, d refers to the calculation step size, and θ refers to the calculation direction; Mean refers to the mean texture feature, f(i,j,d,θ) refers to the spatial feature of each pixel, and is a function of i,j,d,θ.
[0020] The gray-level co-occurrence matrix is used to calculate the mean texture features corresponding to each band of the image, which is easy to operate.
[0021] Furthermore, the method for extracting the optimal band combination is:
[0022]
[0023] Among them, SU(X,Y) is the symmetric uncertainty of variables X and Y; I(X;Y) is the mutual information of variables X and Y; H(X) and H(Y) are the information entropy of variables X and Y; M(S) is the evaluation index of the best band combination, the larger the value, the better the combination; k is the feature subset size, is the average symmetric uncertainty of features and categories, is the average symmetric uncertainty between features;
[0024] The optimal band combination for cloud, smoke, and background pixels is the Coastal band, the Blue band, and the Green band.
[0025] The optimal band combination for smoke and cloud objects is shortwave infrared 1SWIR1, shortwave infrared 2SWIR2, and thermal infrared 1TIRS1.
[0026] The feature selection algorithm (CFS) is combined with the symmetric uncertainty (SU) to perform feature screening and select the best band combination for easy use.
[0027] Furthermore, the steps to extract cloud, smoke, background pixels, smoke objects and cloud objects from the preprocessed image are as follows:
[0028] Use the point function in the ROI tool in the remote sensing processing software ENVI to select cloud, smoke, and background pixel points to make a point shp file;
[0029] Use the polygon function in the ROI tool to draw a surface shp file along the edge of the smoke and cloud, treat each closed surface as an object, and select the smoke object and cloud object.
[0030] Cloud, smoke, background pixels, as well as smoke objects and cloud objects are extracted from the preprocessed images for subsequent use.
[0031] Furthermore, the method of extracting the spectral reflectance and band texture features of Coastal, Blue, and Green bands using the selected cloud, smoke, and background pixels is as follows:
[0032] Use the point function in the ROI tool in the ENVI software to select cloud, smoke, and background pixel points to create a point shp file;
[0033] The point shp file is used to extract the spectral reflectance and band texture features of Coastal, Blue, and Green bands.
[0034] The selected cloud, smoke and background pixels are used to extract the spectral reflectance and band texture features of Coastal, Blue and Green bands, which is conducive to module training.
[0035] Furthermore, the method of extracting the spectral reflectance of SWIR1 and SWIR2 bands, the brightness temperature of TIRS1 band and the texture features of SWIR1, SWIR2 and TIRS1 bands by using the selected smoke objects and cloud objects is as follows:
[0036] Use the polygon function in the ROI tool in the ENVI software to draw a surface shp file along the edge of the smoke and cloud;
[0037] Each closed surface is regarded as an object, and the surface shp file is used to extract the spectral reflectance characteristics and band texture characteristics of each object.
[0038] The selected smoke objects and cloud objects are used to extract the spectral reflectance of SWIR1 and SWIR2 bands, the brightness temperature of TIRS1 band, and the texture features of SWIR1, SWIR2, and TIRS1 bands, which is conducive to module training.
[0039] Furthermore, the steps of combining the RF model and the DT model into a cascade classifier are:
[0040] The RF model and the DT model are connected using the cv2.findContours() function in the opencv library under the python3.7 environment. The cv2.findContours() function is responsible for obtaining the contours of each smoke cloud object in the output results of the smoke cloud pixel and background segmentation RF model, and using the contours to extract features used as input into the smoke object and cloud object classification DT model to form a connection.
[0041] The two trained models are combined into a cascade classifier for the recognition of straw burning remote sensing images.
[0042] The present invention also provides a remote sensing image recognition system for straw burning smoke, comprising an image acquisition module, an image recognition module and a storage module;
[0043] The image acquisition module is used to acquire remote sensing images of straw burning, remote sensing images with smoke-free background interference, and remote sensing images with cloud interference;
[0044] The input end of the image recognition module is connected to the output end of the image acquisition module, and the image recognition module executes the method of the present invention to recognize straw burning smoke;
[0045] The storage module is connected to the image acquisition module and the image recognition module respectively, and is used to store images and recognition results.
[0046] This system uses the cooperation of various modules to realize image collection and analysis processing, identify straw burning smoke, and store the collected data and analysis results for easy retrieval and viewing. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] Figure 1 It is a schematic diagram of the process of the remote sensing image recognition method of straw burning smoke of the present invention;
[0048] Figure 2 It is a bar diagram of the smoke, cloud and background pixel feature screening results of the straw burning smoke remote sensing image recognition method of the present invention;
[0049] Figure 3 It is a bar diagram of the results of screening features of smoke objects and cloud objects in the straw burning smoke remote sensing image recognition method of the present invention. DETAILED DESCRIPTION
[0050] Embodiments of the present invention are described in detail below, examples of which are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and cannot be understood as limiting the present invention.
[0051] In the description of the present invention, it is necessary to understand that the terms "longitudinal", "lateral", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", etc., indicating the orientation or position relationship, are based on the orientation or position relationship shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention.
[0052] In the description of the present invention, unless otherwise specified and limited, it should be noted that the terms "installed", "connected" and "connected" should be understood in a broad sense. For example, it can be a mechanical connection or an electrical connection, or it can be the internal connection between two components. It can be a direct connection or an indirect connection through an intermediate medium. For ordinary technicians in this field, the specific meanings of the above terms can be understood according to the specific circumstances.
[0053] The present invention discloses a remote sensing image recognition method for straw burning smoke, which can accurately identify straw burning smoke through remote sensing images and efficiently remove interfering ground objects and clouds. Figure 1 As shown, the straw burning smoke remote sensing image recognition method includes the following steps:
[0054] Obtain remote sensing images of straw burning, remote sensing images without smoke interference background, and remote sensing images with cloud interference (which can be downloaded from the Landsat-8 public data set), and preprocess them to obtain the spectral reflectance corresponding to each band of the image (spectral reflectance refers to the value obtained after radiation calibration, atmospheric correction, and geometric correction of the original image, that is, the spectral feature; spectral reflectance is the surface reflectance value after eliminating interference, which can reflect the characteristics of clouds, smoke, and background, and is subsequently used for model training); the calculated values of Coastal, Blue, Green, Red, NIR, SWIR1, and SWIR2 bands are subsequently referred to as spectral reflectance, and the calculated value of TIRS1 band is subsequently referred to as brightness temperature;
[0055] Using the gray-level co-occurrence matrix, the mean texture features corresponding to each band of the image are calculated;
[0056] Extracting cloud, smoke, background pixels, smoke objects and cloud objects from the preprocessed images;
[0057] Using the selected cloud, smoke, and background pixels, the spectral reflectance and mean texture features of the corresponding optimal band are extracted; using the selected smoke and cloud objects, the spectral reflectance and mean texture features of the corresponding optimal band are extracted;
[0058] The spectral reflectance and mean texture features extracted from cloud, smoke, and background pixels are used to train the RF model, and the spectral reflectance and mean texture features extracted from smoke and cloud objects are used to train the DT model. Both the RF model and the DT model can be built using the scikit-learn package in the Python 3.7 environment.
[0059] The random forest (RF) and decision tree (DT) models are combined into a cascade classifier to identify straw burning remote sensing images (straw burning smoke in remote sensing images) and obtain the mask of straw burning smoke.
[0060] In a preferred embodiment of the present invention, the method for preprocessing the straw burning remote sensing image, the smoke-free background remote sensing image and the cloud interference remote sensing image is as follows: performing radiation calibration, atmospheric correction and geometric correction on the remote sensing image to reduce image noise and restore the true surface reflectivity of each band;
[0061] Remote sensing images can be cropped to at least 256*256 size.
[0062] In a preferred embodiment of the present invention, the gray level co-occurrence matrix (GLCM) is used to calculate the mean texture features corresponding to each band of the image, which is:
[0063]
[0064] Among them, M and N refer to the image size, i and j refer to the pixel positions, d refers to the calculation step size, and θ refers to the calculation direction; Mean refers to the mean texture feature, f(i,j,d,θ) refers to the spatial feature of each pixel, and is a function of i,j,d,θ.
[0065] In a preferred embodiment of the present invention, the correlation-based feature selection algorithm (CFS) is combined with symmetric uncertainty (SU) to minimize information redundancy while ensuring high correlation. The use of this band combination can more efficiently identify smoke and segment the background and clouds. After screening by the CFS-SU algorithm, the information redundancy of the retained features is minimized while retaining high correlation. The method for extracting the optimal band combination is:
[0066]
[0067] Among them, SU(X,Y) is the symmetric uncertainty of variables X and Y (the best band combination is selected from Coastal, Blue, Green, Red, NIR, SWIR1, SWIR2, and TIRS1 bands, X refers to the spectral reflectance or brightness temperature of one of the above bands, and Y refers to the corresponding category label, i.e., smoke, cloud, and background); I(X; Y) is the mutual information of variables X and Y; H(X) and H(Y) are the information entropy of variables X and Y; M(S) is the evaluation index of the best band combination, the larger the value, the better the combination; k is the feature subset size, is the average symmetric uncertainty of features and categories, is the average symmetric uncertainty between features;
[0068] The optimal band combination for cloud, smoke, and background pixels is the Coastal band, the Blue band, and the Green band. Figure 2 As shown;
[0069] The optimal band combination for smoke and cloud objects is shortwave infrared 1 SWIR1, shortwave infrared 2 SWIR2, and thermal infrared 1 TIRS1 bands, such as Figure 3 The SU calculation results are shown in Table 1:
[0070] Table 1 SU calculation results
[0071]
[0072] In a preferred embodiment of the present invention, the steps of extracting cloud, smoke, background pixels, smoke objects and cloud objects from the preprocessed image are as follows:
[0073] Use the point function in the ROI tool in the remote sensing processing software ENVI to select cloud, smoke, and background pixel points to make a point shp file;
[0074] Use the polygon function in the ROI tool to draw a surface shp file along the edge of the smoke and cloud, treat each closed surface as an object, and select the smoke object and cloud object.
[0075] A simple data encoding scheme can be used, with 0 representing background, 1 representing smoke, and 2 representing cloud, to create labels for machine learning.
[0076] In a preferred embodiment of the present invention, the method for extracting the spectral reflectance and band texture features of Coastal, Blue and Green bands by using the selected cloud, smoke and background pixels is as follows:
[0077] Use the point function in the ROI tool in the ENVI software to select cloud, smoke, and background pixel points to create a point shp file;
[0078] The point shp file is used to extract the spectral reflectance and band texture features of the Coastal, Blue, and Green bands. Specifically, in the remote sensing image processing software ENVI, the prepared point SHP file is used to export the spectral reflectance and band texture features in the corresponding image into a CSV format file for subsequent model training.
[0079] In a preferred embodiment of the present invention, the method for extracting the spectral reflectance of SWIR1 and SWIR2 bands, the brightness temperature of TIRS1 band and the texture features of SWIR1, SWIR2 and TIRS1 bands by using the selected smoke objects and cloud objects is as follows:
[0080] Use the polygon function in the ROI tool in the ENVI software to draw a surface shp file along the edge of the smoke and cloud;
[0081] Each closed surface is regarded as an object, and the surface shp file is used to extract the spectral reflectance characteristics and band texture characteristics of each object. Specifically, in the remote sensing image processing software ENVI, the prepared surface SHP file is used to export the spectral reflectance and band texture characteristics of the corresponding image into a CSV format file for subsequent model training.
[0082] In a preferred embodiment of the present invention, the steps of combining the RF model and the DT model into a cascade classifier are:
[0083] The RF model and the DT model are connected using the cv2.findContours() function in the opencv library under the python3.7 environment. The cv2.findContours() function is responsible for obtaining the contours of each smoke cloud object in the output results of the smoke cloud pixel and background segmentation RF model, and using the contours to extract features used as input into the smoke object and cloud object classification DT model to form a connection.
[0084] For example, 36 additional straw burning smoke images, 40 smoke-free background images, and 40 cloud-interference images that were collected and not used for model training were used to verify the remote sensing image recognition capability of the model of the present invention.
[0085] The evaluation criteria are missed alarm rate and false alarm rate. The calculation criteria are: if the model mistakenly judges the background and cloud as smoke, it is considered a false alarm; if the model mistakenly judges smoke as background and cloud, it is considered a missed alarm. The calculation results are shown in Table 2:
[0086] Table 2 Image recognition verification results
[0087]
[0088] The results show that the method of combining spectral features and texture features used in the present invention to construct a cascade classifier to identify straw burning smoke images has a good effect and can effectively segment the background and clouds while identifying straw burning smoke.
[0089] The present invention also provides a remote sensing image recognition system for straw burning smoke, comprising an image acquisition module, an image recognition module and a storage module. The image acquisition module is used to acquire remote sensing images of straw burning, remote sensing images with smoke-free background interference and remote sensing images with cloud interference. The image acquisition module can use the Landsat-8 land imager OLI and the thermal infrared sensor TIRS.
[0090] The input end of the image recognition module is electrically connected to the output end of the image acquisition module, and the image recognition module executes the method of the present invention to recognize straw burning smoke. The storage module is electrically connected to the image acquisition module and the image recognition module respectively, and is used to store images and recognition results.
[0091] This system uses the cooperation of various modules to realize image collection and analysis processing, identify straw burning smoke, and store the collected data and analysis results for easy retrieval and viewing.
[0092] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.
[0093] Although the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the claims and their equivalents.
Claims
1. A remote sensing image recognition method for straw burning flue gas, characterized in that: The steps include: Obtain remote sensing images of straw burning, remote sensing images without smoke interference background and remote sensing images with cloud interference, and perform preprocessing to obtain the spectral reflectance of each band of the image; Using the gray-level co-occurrence matrix, the mean texture features corresponding to each band of the image are calculated; Extracting cloud, smoke, background pixels, smoke objects and cloud objects from the preprocessed images; Using the selected cloud, smoke, and background pixels, the spectral reflectance and mean texture features of the corresponding optimal band are extracted; using the selected smoke and cloud objects, the spectral reflectance and mean texture features of the corresponding optimal band are extracted; The spectral reflectance and mean texture features extracted from cloud, smoke, and background pixels are used to train the RF model, and the spectral reflectance and mean texture features extracted from smoke objects and cloud objects are used to train the DT model; The RF model and the DT model are combined into a cascade classifier to identify the straw burning remote sensing image and obtain the mask of straw burning smoke.
2. The remote sensing image recognition method for straw burning flue gas according to claim 1, characterized in that: The method for preprocessing the remote sensing images of straw burning, remote sensing images without smoke interference background and remote sensing images with cloud interference is: radiometric calibration, atmospheric correction and geometric correction of the remote sensing images.
3. The remote sensing image recognition method for straw burning flue gas according to claim 1, characterized in that: Using the gray-level co-occurrence matrix, the mean texture features corresponding to each band of the image are calculated as follows: Among them, M and N refer to the image size, i and j refer to the pixel positions, d refers to the calculation step size, and θ refers to the calculation direction; Mean refers to the mean texture feature, f(i,j,d,θ) refers to the spatial feature of each pixel, and is a function of i,j,d,θ.
4. The remote sensing image recognition method for straw burning flue gas according to claim 1, characterized in that: The method to extract the optimal band combination is: Among them, SU(X,Y) is the symmetric uncertainty of variables X and Y; I(X;Y) is the mutual information of variables X and Y; H(X) and H(Y) are the information entropy of variables X and Y; M(S) is the evaluation index of the best band combination, the larger the value, the better the combination; k is the feature subset size, is the average symmetric uncertainty of features and categories, is the average symmetric uncertainty between features; The optimal band combination for cloud, smoke, and background pixels is the Coastal band, the Blue band, and the Green band. The optimal band combination for smoke and cloud objects is shortwave infrared 1SWIR1, shortwave infrared 2SWIR2, and thermal infrared 1TIRS1.
5. The remote sensing image recognition method for straw burning flue gas according to claim 1, characterized in that: The steps to extract cloud, smoke, background pixels, smoke objects and cloud objects from the preprocessed image are as follows: Use the point function in the ROI tool in the remote sensing processing software ENVI to select cloud, smoke, and background pixel points to make a point shp file; Use the polygon function in the ROI tool to draw a surface shp file along the edge of the smoke and cloud, treat each closed surface as an object, and select the smoke object and cloud object.
6. The remote sensing image recognition method for straw burning flue gas according to claim 1, characterized in that: The method of extracting the spectral reflectance and band texture features of Coastal, Blue, and Green bands using the selected cloud, smoke, and background pixels is as follows: Use the point function in the ROI tool in the ENVI software to select cloud, smoke, and background pixel points to create a point shp file; The point shp file is used to extract the spectral reflectance and band texture features of Coastal, Blue, and Green bands.
7. The remote sensing image recognition method for straw burning flue gas according to claim 1, characterized in that: The method for extracting the spectral reflectance of SWIR1 and SWIR2 bands, the brightness temperature of TIRS1 band and the texture features of SWIR1, SWIR2 and TIRS1 bands by using the selected smoke objects and cloud objects is as follows: Use the polygon function in the ROI tool in the ENVI software to draw a surface shp file along the edge of the smoke and cloud; Each closed surface is regarded as an object, and the surface shp file is used to extract the spectral reflectance characteristics and band texture characteristics of each object.
8. The remote sensing image recognition method for straw burning flue gas according to claim 1, characterized in that: The steps to combine the RF model and the DT model into a cascade classifier are: The RF model and the DT model are connected using the cv2.findContours() function in the opencv library under the python3.7 environment. The cv2.findContours() function is responsible for obtaining the contours of each smoke cloud object in the output results of the smoke cloud pixel and background segmentation RF model, and using the contours to extract features used as input into the smoke object and cloud object classification DT model to form a connection.
9. A remote sensing image recognition system for straw burning smoke, characterized in that: It includes an image acquisition module, an image recognition module and a storage module; The image acquisition module is used to acquire remote sensing images of straw burning, remote sensing images with smoke-free background interference, and remote sensing images with cloud interference; The input end of the image recognition module is connected to the output end of the image acquisition module, and the image recognition module executes the method according to any one of claims 1 to 8 to recognize straw burning smoke; The storage module is connected to the image acquisition module and the image recognition module respectively, and is used to store images and recognition results.
