Ground surface impervious layer identification method, device and equipment and storage medium

Abstract

The invention provides an earth surface impervious layer identification method, device and equipment and a storage medium, and relates to the technical field of remote sensing image processing, and the method comprises the steps: obtaining radar remote sensing image data, multispectral remote sensing image data, night light remote sensing image data and thermal infrared remote sensing image data of a to-be-researched region; then image feature extraction of polarization features, texture features, spectral index features, night light index features and thermal infrared index features is carried out, the image features are combined and applied, classification is carried out through a random forest model, impervious earth surfaces with different densities and other ground feature types can be effectively distinguished, and the impervious layer extraction precision is improved. And key features are screened by using ranking, so that data redundancy is reduced, and the calculation efficiency and the classification effect of the model are enhanced. A pre-trained random forest classification model ensures the stability and reliability of a classification result by means of the strong processing capability of the model on complex data.

Description

Technical Field

[0001] The present application relates to the technical field of remote sensing image processing, and in particular to a method, device, equipment and storage medium for identifying an impermeable layer on the surface. Background Art

[0002] Impervious layers refer to natural or artificial surface covers such as asphalt or cement roads, roofs, and parking lots that prevent water from seeping in. With the continuous acceleration of urbanization, urban space expansion and rapid population growth, a series of problems such as urban heat islands, environmental degradation, and traffic congestion have become increasingly serious, posing huge challenges to the ecological environment and urban development. In the process of urbanization, the existence of impervious layers not only hinders the evapotranspiration of the surface and destroys the surface thermal balance, but also leads to an imbalance in the water cycle, seriously affecting human production and life. The area and distribution of impervious layers are important factors affecting the environment caused by urbanization. Therefore, timely and accurate acquisition of impervious layer information is of great significance to urbanization and ecological and environmental protection.

[0003] In the existing technology, remote sensing technology is used to extract impervious surfaces, mainly using optical remote sensing images as the data source. However, when collecting optical remote sensing images, it is impossible to overcome the atmospheric influence and cloud cover in the image. Secondly, due to the strong complexity of urban landscapes, the spectral characteristics of some objects in urban areas are not clearly distinguished, and there is a certain degree of spectral confusion, so there will be "same object different spectrum" and "different objects same spectrum" phenomena, making it difficult to accurately estimate the spatial distribution information of the impervious layer. Summary of the invention

[0004] In view of the above problems, the present application provides a method for identifying a surface impermeable layer, comprising the following contents:

[0005] In a first aspect, the present application provides a method for identifying an impermeable layer on a surface, the method comprising:

[0006] Obtain radar remote sensing image data, multispectral remote sensing image data, night light remote sensing image data and thermal infrared remote sensing image data of the area to be studied and perform data preprocessing so that the processed data have the same resolution;

[0007] Performing feature extraction on the data after the data preprocessing to obtain polarization features, texture features, spectral index features, night light index features and thermal infrared index features; the polarization features, texture features, spectral index features, night light index features and thermal infrared index features respectively include multiple features;

[0008] Ranking the features extracted from the feature extraction, and selecting the top n features in terms of importance;

[0009] The n features are processed based on a pre-trained random forest classification model to obtain a classification result of the surface type in the area to be studied.

[0010] Optionally, ranking the features extracted from the feature extraction and screening the top n features in terms of importance includes:

[0011] In a random forest consisting of m decision trees, calculate the initial out-of-bag error and new out-of-bag error of each decision tree;

[0012] Obtaining the importance of each feature based on the initial out-of-bag error and the new out-of-bag error, and sorting the importance of each feature from high to low;

[0013] Select the first n features from the sorted results.

[0014] Optionally, the extracting features from the data after the data preprocessing to obtain polarization features includes:

[0015] Acquire VV polarization characteristic images and VH polarization characteristic images of the radar remote sensing image data, wherein the radar remote sensing image data is acquired by using a synthetic aperture radar SAR system; the VV polarization characteristic image refers to an image formed by radar echo data acquired by the SAR system in a vertical transmission and vertical reception mode; the VH polarization characteristic image refers to an image formed by radar echo data acquired by the SAR system in a vertical transmission and horizontal reception mode;

[0016] Features are extracted from the VV polarization feature image and the VH polarization feature image of the radar remote sensing image data as the polarization features, and the polarization features include the average value, standard deviation, quantile, difference between the maximum and minimum values, interquartile range, skewness and kurtosis of the pixel values ​​of the VV polarization image; the average value, standard deviation, quantile, difference between the maximum and minimum values, interquartile range, skewness and kurtosis of the pixel values ​​of the VH polarization image.

[0017] Optionally, the extracting features from the data after the data preprocessing to obtain texture features includes:

[0018] The red light band of the multispectral remote sensing image data after the data preprocessing is extracted, and the texture features are extracted using the gray level co-occurrence matrix. The texture features include mean, variance, homogeneity, dissimilarity, entropy and angular second moment.

[0019] Optionally, the extracting features of the data after the data preprocessing to obtain spectral index features includes:

[0020] Extracting the green light band and the infrared band of the multispectral remote sensing image data after the data preprocessing to construct a normalized difference water body index; extracting the near infrared band and the infrared band of the multispectral remote sensing image data after the data preprocessing to construct a normalized vegetation index; extracting the blue light band, the green light band and the red light band as well as the near infrared band and the short-wave infrared band of the multispectral remote sensing image data after the data preprocessing to construct a normalized difference impervious surface index;

[0021] The normalized difference water index, the normalized vegetation index and the normalized difference impervious surface index are used as the spectral index features.

[0022] Optionally, the extracting features of the data after the data preprocessing to obtain the night light index features includes:

[0023] Acquire the nighttime light remote sensing image data after the data preprocessing, wherein the nighttime light remote sensing image data after the data preprocessing includes data of high-gain and low-gain fusion bands;

[0024] The night light index feature is obtained based on the normalized vegetation index obtained from the near infrared band and the infrared band of the multispectral remote sensing image data after the data preprocessing, and the data of the high and low gain fusion bands.

[0025] Optionally, the extracting features of the data after the data preprocessing to obtain the thermal infrared index features includes:

[0026] The thermal infrared remote sensing image data after the data preprocessing is calibrated by radiation, the gray value is converted into a radiation brightness value, and a normalized thermal infrared index is constructed as the thermal infrared index feature.

[0027] In a second aspect, the present application provides a device for identifying a surface impermeable layer, the device comprising:

[0028] An acquisition unit is used to acquire radar remote sensing image data, multispectral remote sensing image data, night light remote sensing image data and thermal infrared remote sensing image data of the area to be studied;

[0029] A data processing unit, used for performing data preprocessing so that the processed data have the same resolution;

[0030] A feature extraction unit, used to extract features from the data after the data preprocessing to obtain polarization features, texture features, spectral index features, night light index features and thermal infrared index features;

[0031] A feature screening unit, used to screen the features obtained by the feature extraction to obtain the top n features ranked by importance;

[0032] The classification unit is used to process the n features based on a pre-trained random forest classification model to obtain a classification result of the surface type in the area to be studied.

[0033] Optionally, the feature screening unit is specifically used to calculate, in a random forest consisting of m decision trees, an initial out-of-bag error and a new out-of-bag error of each decision tree respectively;

[0034] Obtaining the importance of each feature based on the initial out-of-bag error and the new out-of-bag error, and sorting the importance of each feature from high to low;

[0035] Select the first n features from the sorted results.

[0036] Optionally, the feature extraction unit extracts features from the data after the data preprocessing to obtain polarization features, including:

[0037] Acquire VV polarization characteristic images and VH polarization characteristic images of the radar remote sensing image data, wherein the radar remote sensing image data is acquired by using a synthetic aperture radar SAR system; the VV polarization characteristic image refers to an image formed by radar echo data acquired by the SAR system in a vertical transmission and vertical reception mode; the VH polarization characteristic image refers to an image formed by radar echo data acquired by the SAR system in a vertical transmission and horizontal reception mode;

[0038] Features are extracted from the VV polarization feature image and the VH polarization feature image of the radar remote sensing image data as the polarization features, and the polarization features include the average value, standard deviation, quantile, difference between the maximum and minimum values, interquartile range, skewness and kurtosis of the pixel values ​​of the VV polarization image; the average value, standard deviation, quantile, difference between the maximum and minimum values, interquartile range, skewness and kurtosis of the pixel values ​​of the VH polarization image.

[0039] Optionally, the feature extraction unit extracts features from the data after the data preprocessing to obtain texture features including:

[0040] The red light band of the multispectral remote sensing image data after the data preprocessing is extracted, and the texture features are extracted using the gray level co-occurrence matrix. The texture features include mean, variance, homogeneity, dissimilarity, entropy and angular second moment.

[0041] Optionally, the feature extraction unit performs feature extraction on the data after the data preprocessing to obtain spectral index features including:

[0042] Extracting the green light band and the infrared band of the multispectral remote sensing image data after the data preprocessing to construct a normalized difference water body index; extracting the near infrared band and the infrared band of the multispectral remote sensing image data after the data preprocessing to construct a normalized vegetation index; extracting the blue light band, the green light band and the red light band as well as the near infrared band and the short-wave infrared band of the multispectral remote sensing image data after the data preprocessing to construct a normalized difference impervious surface index;

[0043] The normalized difference water index, the normalized vegetation index and the normalized difference impervious surface index are used as the spectral index features.

[0044] Optionally, the feature extraction unit performs feature extraction on the data after the data preprocessing to obtain night light index features including:

[0045] Acquire the nighttime light remote sensing image data after the data preprocessing, wherein the nighttime light remote sensing image data after the data preprocessing includes data of high-gain and low-gain fusion bands;

[0046] The night light index feature is obtained based on the normalized vegetation index obtained from the near infrared band and the infrared band of the multispectral remote sensing image data after the data preprocessing, and the data of the high and low gain fusion bands.

[0047] Optionally, the feature extraction unit extracts features from the data after the data preprocessing to obtain thermal infrared index features including:

[0048] The thermal infrared remote sensing image data after the data preprocessing is calibrated by radiation, the gray value is converted into a radiation brightness value, and a normalized thermal infrared index is constructed as the thermal infrared index feature.

[0049] In a third aspect, the present application provides a device comprising a memory and a processor, wherein the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes the method for identifying a surface impermeable layer introduced in any implementation of the first aspect.

[0050] In a fourth aspect, the present application provides a computer-readable storage medium having codes stored therein. When the codes are executed, the device executing the codes implements the method for identifying a surface impermeable layer as described in any implementation of the first aspect.

[0051] The present application provides a method for identifying an impermeable layer on the surface. The method obtains radar remote sensing image data, multispectral remote sensing image data, night light remote sensing image data and thermal infrared remote sensing image data of the area to be studied, and then extracts image features of polarization features, texture features, spectral index features, night light index features and thermal infrared index features and applies them in combination. It can effectively distinguish impermeable surfaces from other types of land objects and improve the accuracy of impermeable layer extraction. The use of ranking to screen key features reduces data redundancy and enhances the computational efficiency and classification effect of the model. The pre-trained random forest classification model, with its powerful processing capabilities for complex data, ensures the stability and reliability of the classification results. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the technical solutions in this embodiment or the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0053] Figure 1 A flow chart of a method for identifying an impermeable layer on a surface provided in an embodiment of the present application;

[0054] Figure 2 A schematic diagram of a feature screening result provided in an embodiment of the present application;

[0055] Figure 3 A schematic diagram of the classification results of the surface types in a region to be studied provided in an embodiment of the present application;

[0056] Figure 4 A schematic diagram of the structure of a surface impermeable layer identification device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solution and advantages of the embodiments of the present application clearer, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0058] Figure 1 A flow chart of a method for identifying an impermeable layer on the surface provided in an embodiment of the present application. Figure 1 As shown, the surface impermeable layer identification method provided in the embodiment of the present application may include:

[0059] S101, obtaining radar remote sensing image data, multispectral remote sensing image data, night light remote sensing image data and thermal infrared remote sensing image data of the area to be studied and performing data preprocessing so that the processed data have the same resolution.

[0060] Radar remote sensing image data: Radar remote sensing image data is data obtained by remote sensing observation using electromagnetic radiation in the microwave band, mainly including synthetic aperture radar (SAR) and scatterometer radar. The radar remote sensing image data collected in this application refers to C-band synthetic aperture radar data obtained in the interferometric wide-band imaging mode, and has reached the Level-1 level after ground-based multi-view processing. The image has VV and VH dual polarization characteristics, and its spatial resolution is 5×20m.

[0061] After acquiring the radar remote sensing image data, the data is preprocessed by orbit correction, thermal noise removal, radiation calibration, speckle filtering, geocoding, and impact splicing. The specific processing process includes:

[0062] Orbit correction uses high-precision orbit files to fine-tune the satellite's trajectory in order to accurately obtain the satellite's position and speed information during its operation. This step can significantly improve the accuracy of subsequent image registration and geocoding, laying a solid foundation for subsequent analysis.

[0063] Thermal noise removal, based on the noise characteristic information contained in the metadata set, identifies and eliminates the thermal noise generated inside the radar system. This is the key to removing the interference of unnatural factors in the data, which helps to improve the signal-to-noise ratio of the image.

[0064] Radiation calibration processing, since there is radiation deviation in radar data, it needs to be corrected through the radiation calibration process. The radiation calibration formula used is:

[0065]

[0066] In formula (1): DN i With A i are the gray value and calibration parameter of pixel i respectively.

[0067] Speckle noise filtering, as a common random distribution noise in radar images, needs to be suppressed by appropriate methods. In the embodiment of the present application, a Refined Lee filter is selected for speckle filtering, which can effectively reduce noise while better retaining the radiation characteristics and texture details of the image, ensuring the integrity of the image information.

[0068] Geocoding and geometric correction: Geocoding is performed to convert the image's projection coordinates into a geographic coordinate system for spatial position analysis. At the same time, a 30-meter resolution SRTM digital elevation model (DEM) is used to correct geometric deformations such as perspective contraction and overlap that may occur during radar imaging. Finally, the processed image data is output in a 10-meter spatial resolution format through resampling technology to meet the needs of different application scenarios.

[0069] Multispectral remote sensing image data: This application uses the atmospherically corrected surface reflectance L2A-level data product, which contains 13 bands, and the specific parameters are shown in the table. The multispectral remote sensing image data in the embodiment of this application uses the 10m and 20m spatial resolution bands in Table 1, and its spatial resolution is uniformly resampled to 10m.

[0070] Table 1: Multispectral satellite data band information

[0071]

[0072]

[0073] Night light remote sensing image data: The intensity and distribution data of night lights detected by the low-light imager (GIU), including panchromatic band and color band. The spatial resolution of the panchromatic band is 10m, and the spatial resolution of the three color bands of red, green and blue is 40m.

[0074] Thermal infrared remote sensing image data: The thermal infrared remote sensing image captured data in three key infrared bands, B1 (8.0 to 10.5 microns), B2 (10.60 to 11.19 microns) and B3 (10.50 to 12.51 microns) through the Thermal Infrared Imager (TIS). These data demonstrated excellent detection capabilities at a spatial resolution of 30 meters, and their noise equivalent temperature difference (NEDT) was less than 0.041K at 300K, ensuring high-precision measurements.

[0075] The data used in this application are Level 4 data products acquired by TIS and GIU sensors in collaboration. These products have undergone a rigorous processing flow to meet application standards. Specifically, Level 1 data products are converted into standardized data products based on the original Level 0 products through the implementation of relative radiation correction, precise band registration, HDR fusion technology, and RPC (Rational Polynomial Coefficients) model correction. Level 2 products are further geometrically corrected on this basis to improve the accuracy of spatial positioning. As for Level 4 products, they are based on Level 1 standard products, using precise ground control points (GCPs) and detailed digital elevation models (DEM) for orthorectification, and are output in accordance with specific format specifications to ensure the ease of use and accuracy of the data.

[0076] In order to ensure that data from different sources and with different spatial resolutions can be seamlessly connected in space, this application uniformly resamples the spatial resolution of all data to 10 meters. In addition, all data are finely geometrically aligned using multispectral images as a benchmark to ensure their consistency in spatial position. Finally, for global data exchange and application, all processed data are converted to the WGS-84UTM 50N projection system, achieving data standardization and internationalization.

[0077] S102, performing feature extraction on the data after the data preprocessing to obtain polarization features, texture features, spectral index features, night light index features and thermal infrared index features.

[0078] After preprocessing the radar remote sensing image data, multispectral remote sensing image data, night light remote sensing image data and thermal infrared remote sensing image data, feature extraction is further performed on the radar remote sensing image data, multispectral remote sensing image data, night light remote sensing image data and thermal infrared remote sensing image data to obtain polarization features, texture features, spectral index features, night light index features and thermal infrared index features, wherein the polarization features, texture features, spectral index features, night light index features and thermal infrared index features include multiple features. The process of obtaining the above features is described in detail below:

[0079] The extracting features of the data after the data preprocessing to obtain polarization features comprises:

[0080] A VV polarization feature image and a VH polarization feature image of the radar remote sensing image data are obtained, wherein the radar remote sensing image data are obtained by using a synthetic aperture radar (SAR) system; the VV polarization feature image refers to an image formed by radar echo data obtained by the SAR system in a vertical transmission and vertical reception mode; the VH polarization feature image refers to an image formed by radar echo data obtained by the SAR system in a vertical transmission and horizontal reception mode; features are extracted from the VV polarization feature image and the VH polarization feature image of the radar remote sensing image data as the polarization features, wherein the polarization features include an average value, a standard deviation, a quantile, a difference between a maximum value and a minimum value, an interquartile range, a skewness and a kurtosis of pixel values ​​of the VV polarization image; and an average value, a standard deviation, a quantile, a difference between a maximum value and a minimum value, an interquartile range, a skewness and a kurtosis of pixel values ​​of the VH polarization image.

[0081] Mean: reflects the average backscatter level across the dataset and is useful for identifying commonly occurring surface types. Standard Deviation: measures the dispersion of the dataset; a high standard deviation may indicate high heterogeneity in the surface cover. Quantiles (0%, 10%, 25%, 50%, 75%, 90%, 100%): these quantiles provide a full picture of the data distribution and help identify areas of varying backscatter strength. For example, high percentiles may correspond to the tops of buildings with high reflectivity, while low percentiles may correspond to vegetation or water bodies with low reflectivity. Range: the difference between the maximum and minimum values, representing the dynamic range of the dataset. Interquartile Range (IQR, i.e., Q3-Q1): measures the dispersion of the middle 50% of the data and is useful for identifying surface types with moderate reflectivity. Skewness: describes the asymmetry of the data distribution. Positive skewness indicates that the data is right-skewed (i.e., most values ​​are below the mean), while negative skewness indicates that the data is left-skewed (i.e., most values ​​are above the mean). In an urban environment, skewness may reflect differences in reflectivity between buildings and open space. Kurtosis: This measures how sharp or flat the data distribution is. A high kurtosis indicates that the data is concentrated around the mean, forming a peak; a low kurtosis indicates that the data distribution is relatively flat.

[0082] The feature extraction of the data after the data preprocessing to obtain the texture feature comprises:

[0083] The red light band of the multispectral remote sensing image data after the data preprocessing is extracted, and the texture features are extracted using the gray level co-occurrence matrix. The texture features include mean, variance, homogeneity, dissimilarity, entropy and angular second moment.

[0084] Based on the red light band in the multispectral data, the gray level co-occurrence matrix (GLCM) was used to calculate six texture features with weak correlation, including mean (MEA), variance (VAR), homogeneity (HOM), dissimilarity (DIS), entropy (ENT) and angular second moment (ASM).

[0085] The calculation formulas for the above features are:

[0086]

[0087]

[0088]

[0089]

[0090]

[0091]

[0092] Where N is the grayscale data; P is the N×N dimensional normalized grayscale co-occurrence matrix; P(i,j) is the standardized grayscale value of the i-th row and j-th column in the matrix. The calculation processing window is set to 7 pixels × 7 pixels.

[0093] The feature extraction of the data after the data preprocessing to obtain the spectral index feature includes:

[0094] The green light band and the infrared band of the multispectral remote sensing image data after the data preprocessing are extracted to construct a normalized difference water index; the near infrared band and the infrared band of the multispectral remote sensing image data after the data preprocessing are extracted to construct a normalized vegetation index; the blue light band, the green light band and the red light band as well as the near infrared band and the short-wave infrared band of the multispectral remote sensing image data after the data preprocessing are extracted to construct a normalized difference impervious surface index; the normalized difference water index, the normalized vegetation index and the normalized difference impervious surface index are used as the spectral index features.

[0095] In order to accurately extract water body information, the significant difference in reflectivity between water bodies and vegetation in different spectral bands can be used to design a spectral index. This index aims to enhance the display of water bodies by suppressing the reflection signal of vegetation. Although it can effectively distinguish between vegetation and water bodies, it still faces the challenge of noise interference when distinguishing between water bodies and buildings. Therefore, in the embodiment of the present application, the green light and near-infrared bands in multispectral images are combined to develop a normalized difference water index (MNDWI), which can significantly improve the contrast between water bodies and buildings, thereby more effectively masking and extracting water body information.

[0096] In addition, the Normalized Difference Vegetation Index (NDVI), as a classic indicator for measuring the growth status of surface vegetation, has been widely used in vegetation information extraction. However, when dealing with urban environments, the spectral characteristics of impervious surfaces (such as buildings) and bare soil in the visible light band are similar, which often leads to high-resolution images. It is difficult to accurately distinguish between it and bare soil when extracting impervious surface information, resulting in misclassification. The embodiments of the present application deeply explore the spectral differences between impervious surfaces and bare soil in the near-infrared, short-wave infrared 1 band and short-wave infrared 2 band, and these differences provide an important basis for distinguishing the two. Based on this, the Normalized Difference Impervious Surface Index (MNDISI) is constructed. This index can make full use of these spectral differences to achieve effective distinction between impervious surfaces and bare soil, and provides a new method for accurately extracting urban surface cover information.

[0097] The calculation process of the normalized difference water index (MNDWI), the normalized difference vegetation index (NDVI) and the normalized difference impervious surface index (MNDISI) is as follows:

[0098]

[0099]

[0100]

[0101] In the above formula, Blue, Green, and Red are the blue, green, and red light bands of multispectral remote sensing images, respectively; NIR, SWIR1, and SWIR2 are the near infrared, shortwave infrared 1, and 2 bands of multispectral remote sensing images, respectively.

[0102] The feature extraction of the data after the data preprocessing to obtain the night light index feature includes:

[0103] Acquire the night light remote sensing image data after the data preprocessing, wherein the night light remote sensing image data after the data preprocessing includes data of high-gain and low-gain fusion bands; obtain the night light index characteristics based on the normalized vegetation index obtained from the near-infrared band and the infrared band of the multispectral remote sensing image data after the data preprocessing, and the data of the high- and low-gain fusion bands.

[0104] The night light remote sensing image data is a night light panchromatic band, including low gain (PanchromaticLow, PL), high gain (Panchromatic High, PH) and high and low gain fusion (High Dynamic Range, HDR, the ground system fuses PL and PH data with 50% weight each) three types of data. The night light index feature is also the vegetation adjustment light index VANUI. VANUI is a city index that combines night light remote sensing data and multi-spectral data in near-infrared and infrared bands to reduce light saturation and increase internal differences in cities.

[0105] The embodiment of the present application uses vegetation to adjust the night light index to enhance the impervious surface information and increase the difference between different types of landforms. VANUI is calculated by the following formula:

[0106] VANUI=(1-NDVI)×NTL (11)

[0107] In the above formula, NTL is the normalized data of night light intensity of high and low gain fusion bands, and NDVI is the normalized vegetation index data calculated using the near infrared and infrared bands of multispectral images.

[0108] The feature extraction of the data after the data preprocessing to obtain the thermal infrared index feature includes: performing radiation calibration on the thermal infrared remote sensing image data after the data preprocessing, converting the grayscale value into the radiation brightness value, and constructing a normalized thermal infrared index as the thermal infrared index feature.

[0109] First, the thermal infrared remote sensing data is calibrated by radiation, and the grayscale value of the remote sensing image is converted into the radiation brightness value using formula (12):

[0110] L=DN×Gain+Bias (12)

[0111] Where L is the radiance at the sensor entrance pupil, in W / m 2 / sr / μm, DN represents the gray value of the image after relative radiation calibration. Gain is gain, and Bias is bias. The specific calibration parameter values ​​of the three thermal infrared bands refer to Table 2.

[0112] Table 2: Absolute radiation calibration coefficients for thermal infrared sensors

[0113] Band Gain Bias B1 0.003947 0.167126 B2 0.003946 0.124622 B3 0.005329 0.222530

[0114] In the embodiment of the present application, two normalized thermal infrared indices (RI1, RI2) are constructed to highlight the thermal radiation differences between the three bands B1, B2, and B3 of the ground objects, and enhance the thermal radiation intensity characteristics of impermeable surfaces of different densities. The calculation methods of the two normalized thermal infrared indices are as follows:

[0115]

[0116]

[0117] Where SBT B1 、SBT B2 、SBT B3 They are bands 1, 2, and 3 of TIS thermal infrared remote sensing data respectively.

[0118] S103: rank the features obtained by the feature extraction, and select the top n features in terms of importance.

[0119] According to the introduction in S102, after feature extraction of the data after the data preprocessing, the obtained features include VV, VH (mean, standard deviation, quantile, range, interquartile range, skewness, kurtosis, NDVI, MNDWI, MNDISI, VANUI, RI1, RI2, mean (MEA), variance (VAR), homogeneity (HOM), dissimilarity (DIS), entropy (ENT), angular second moment (ASM) and different spectral bands in multispectral remote sensing image data. In order to effectively improve the performance and efficiency of the surface classification model, it is necessary to screen the acquired features and reduce the computational complexity of the model by selecting relatively important feature variables. In the embodiment of the present application, the out-of-bag data (OOB) of the random forest is used to calculate the feature importance, and the importance is ranked according to the calculation results, and the top n features are selected from the ranking. The number of n is randomly set according to the actual situation. In the embodiment of the present application, n is 13, and the specific feature importance calculation process is as follows:

[0120] In a random forest consisting of m decision trees, the out-of-bag error e of each decision tree in the random forest is calculated based on OOB. t , and then randomly select the jth feature variable X for the out-of-bag data j Assign a value and calculate the new out-of-bag error If X j The greater the increase in out-of-bag error and the greater the decrease in accuracy caused by the change in the variable, the more important the variable is. j Importance of V(X j ) is expressed as:

[0121]

[0122] The extracted feature variables are ranked according to the OOB out-of-bag error of the random forest algorithm, and the top 13 feature variables are selected for surface classification. The results are as follows: Figure 2 As shown, Figure 2 A schematic diagram of feature screening results provided for an embodiment of the present application, showing the top 13 features ranked by importance based on feature variables.

[0123] S104: Process the n features based on a pre-trained random forest classification model to obtain a classification result of the surface type in the area to be studied.

[0124] The pre-trained random forest classification model is used to input features so that the output results include the land use / cover type identification results of the area to be studied, such as Figure 3 As shown, Figure 3 A schematic diagram of the classification results of the surface types in the area to be studied provided for an embodiment of the present application shows the classification results of the area to be studied at a resolution of 10 meters. It should be noted that when using random forest classification, it is usually necessary to set the number of growing trees and the number of feature variables input when the node splits. In an implementation method of an embodiment of the present application, after a series of experiments, it was found that when N=100, the out-of-bag error (OOB) tends to be stable. Therefore, 100 trees are selected for random forest classification, and the number of feature variables is calculated by the square root of the number of features involved in the classification.

[0125] When constructing a random forest classification model for identifying surface types, in order to obtain high-quality training samples, the ROI (Region of Interest) extraction tool in the ENVI software is first used to establish regions of interest to select samples. These regions include but are not limited to: high-density impervious surfaces, medium-density impervious surfaces, low-density impervious surfaces, cultivated land, greenhouse land, forest land, water areas, and unused land. Then the random forest algorithm is used to further extract and identify the impervious layer. The random forest algorithm has been widely used in the field of remote sensing image processing due to its powerful classification ability and good generalization performance. The basic idea of ​​the algorithm is to use the Bootstrap method to randomly extract a sample subsets with replacement from the original training set, and each subset contains about 2 / 3 of the total sample size. Based on these sample subsets, the algorithm constructs a independent decision trees to form a decision tree "forest".

[0126] In the process of building each decision tree, the algorithm randomly selects a part of the features (assuming that there are b features in the training sample data) as the candidate feature set, and splits according to the optimal features each time the node splits. This process enables each tree to learn different feature combinations and classification rules from slightly different data sets, thereby increasing the diversity of the model. Each tree will grow fully until the stopping condition is met (such as all samples in the node belong to the same class) without any pruning operation.

[0127] Finally, the generated multiple classification trees are combined into a random forest classifier. When classifying new samples, each tree will give a classification result, and then the final classification result is determined by a voting mechanism (i.e. majority rule). In addition, since the a sample subsets randomly selected each time are smaller than the total sample size, the remaining samples can be used for cross-validation to evaluate the performance and generalization ability of the model. In this way, not only can the accuracy of the random forest algorithm in impermeable layer extraction be improved, but also its adaptability to new data can be enhanced.

[0128] After the classification is completed, the confusion matrix is ​​used to verify the accuracy of the classification results. The evaluation indicators derived from the confusion matrix mainly include overall classification accuracy, producer's accuracy (PA), user's accuracy (UA) and Kappa coefficient. The verification samples are obtained based on visual interpretation, and the Confusion Matrix Using Ground Truth ROIs tool is used in ENVI software to verify the accuracy. The overall accuracy of the classification results using the classification method of this study reached 93.2%, and the Kappa coefficient reached 0.923.

[0129] Table 3: Confusion matrix of classification results accuracy evaluation

[0130]

[0131]

[0132] The accuracy evaluation results show that the overall classification accuracy of the surface impermeable layer identification method in this application reaches 93.2%, among which the classification accuracy of high, medium and low density impermeable surfaces is above 95%.

[0133] The above are some specific implementations of a method for identifying a surface impermeable layer provided in the embodiment of the present application. Based on this, the present application also provides a corresponding device. The device provided in the embodiment of the present application will be introduced from the perspective of functional modularization.

[0134] Figure 4A schematic diagram of the structure of a surface impermeable layer identification device provided in an embodiment of the present application.

[0135] Combination Figure 4 As shown, the surface impermeable layer identification device 400 provided in the embodiment of the present application includes:

[0136] An acquisition unit 410 is used to acquire radar remote sensing image data, multispectral remote sensing image data, night light remote sensing image data and thermal infrared remote sensing image data of the area to be studied;

[0137] A data processing unit 420 is used to perform data preprocessing so that the processed data have the same resolution;

[0138] A feature extraction unit 430 is used to extract features from the data after the data preprocessing to obtain polarization features, texture features, spectral index features, night light index features and thermal infrared index features;

[0139] A feature screening unit 440 is used to screen the features obtained by the feature extraction to obtain the top n features ranked by importance;

[0140] The classification unit 450 is used to process the n features based on a pre-trained random forest classification model to obtain a classification result of the surface type in the area to be studied.

[0141] In an implementation of the embodiment of the present application, the feature screening unit is specifically used to calculate the initial out-of-bag error and the new out-of-bag error of each decision tree in a random forest composed of m decision trees;

[0142] Obtaining the importance of each feature based on the initial out-of-bag error and the new out-of-bag error, and sorting the importance of each feature from high to low;

[0143] Select the first n features from the sorted results.

[0144] In an implementation manner of the embodiment of the present application, the feature extraction unit extracts features from the data after the data preprocessing to obtain polarization features, including:

[0145] Acquire VV polarization characteristic images and VH polarization characteristic images of the radar remote sensing image data, wherein the radar remote sensing image data is acquired by using a synthetic aperture radar SAR system; the VV polarization characteristic image refers to an image formed by radar echo data acquired by the SAR system in a vertical transmission and vertical reception mode; the VH polarization characteristic image refers to an image formed by radar echo data acquired by the SAR system in a vertical transmission and horizontal reception mode;

[0146] Features are extracted from the VV polarization feature image and the VH polarization feature image of the radar remote sensing image data as the polarization features, and the polarization features include the average value, standard deviation, quantile, difference between the maximum and minimum values, interquartile range, skewness and kurtosis of the pixel values ​​of the VV polarization image; the average value, standard deviation, quantile, difference between the maximum and minimum values, interquartile range, skewness and kurtosis of the pixel values ​​of the VH polarization image.

[0147] In an implementation of the embodiment of the present application, the feature extraction unit extracts features from the data after the data preprocessing, and the texture features obtained include:

[0148] The red light band of the multispectral remote sensing image data after the data preprocessing is extracted, and the texture features are extracted using the gray level co-occurrence matrix. The texture features include mean, variance, homogeneity, dissimilarity, entropy and angular second moment.

[0149] In an implementation of the embodiment of the present application, the feature extraction unit extracts features from the data after the data preprocessing, and obtains spectral index features including:

[0150] Extracting the green light band and the infrared band of the multispectral remote sensing image data after the data preprocessing to construct a normalized difference water body index; extracting the near infrared band and the infrared band of the multispectral remote sensing image data after the data preprocessing to construct a normalized vegetation index; extracting the blue light band, the green light band and the red light band as well as the near infrared band and the short-wave infrared band of the multispectral remote sensing image data after the data preprocessing to construct a normalized difference impervious surface index;

[0151] The normalized difference water index, the normalized vegetation index and the normalized difference impervious surface index are used as the spectral index features.

[0152] In an implementation of the embodiment of the present application, the feature extraction unit extracts features from the data after the data preprocessing, and obtains night light index features including:

[0153] Acquire the nighttime light remote sensing image data after the data preprocessing, wherein the nighttime light remote sensing image data after the data preprocessing includes data of high-gain and low-gain fusion bands;

[0154] The night light index feature is obtained based on the normalized vegetation index obtained from the near infrared band and the infrared band of the multispectral remote sensing image data after the data preprocessing, and the data of the high and low gain fusion bands.

[0155] In an implementation of the embodiment of the present application, the feature extraction unit extracts features from the data after the data preprocessing, and obtains thermal infrared index features including:

[0156] The thermal infrared remote sensing image data after the data preprocessing is calibrated by radiation, the gray value is converted into a radiation brightness value, and a normalized thermal infrared index is constructed as the thermal infrared index feature.

[0157] The embodiments of the present application also provide corresponding devices and computer storage media for implementing the solutions provided by the embodiments of the present application.

[0158] The device includes a memory and a processor, the memory is used to store instructions or codes, and the processor is used to execute the instructions or codes so that the device executes the method described in any embodiment of the present application.

[0159] The computer storage medium stores codes, and when the codes are executed, a device executing the codes implements the method described in any embodiment of the present application.

[0160] Through the description of the above implementation methods, it can be known that those skilled in the art can clearly understand that all or part of the steps in the above-mentioned embodiment method can be implemented by means of software plus a general hardware platform. Based on such an understanding, the technical solution of the present application can be embodied in the form of a software product, which can be stored in a storage medium, such as a read-only memory (ROM) / RAM, a magnetic 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 communication device such as a router) to execute the methods described in each embodiment of the present application or some parts of the embodiments.

[0161] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0162] It should also be noted that the various embodiments in this specification are described in a progressive manner, and the same and similar parts between the various embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and apparatus embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiments. The device and apparatus embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components indicated 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.

[0163] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for identifying an impermeable layer on a surface, characterized in that: The method comprises: Obtain radar remote sensing image data, multispectral remote sensing image data, night light remote sensing image data and thermal infrared remote sensing image data of the area to be studied and perform data preprocessing so that the processed data have the same resolution; Performing feature extraction on the data after the data preprocessing to obtain polarization features, texture features, spectral index features, night light index features and thermal infrared index features; the polarization features, texture features, spectral index features, night light index features and thermal infrared index features respectively include multiple features; Ranking the features extracted from the feature extraction, and selecting the top n features in terms of importance; The n features are processed based on a pre-trained random forest classification model to obtain a classification result of the surface type in the area to be studied.

2. The method according to claim 1, characterized in that Ranking the features extracted from the feature extraction and screening the top n features in terms of importance include: In a random forest consisting of m decision trees, calculate the initial out-of-bag error and new out-of-bag error of each decision tree; Obtaining the importance of each feature based on the initial out-of-bag error and the new out-of-bag error, and sorting the importance of each feature from high to low; Select the first n features from the sorted results.

3. The method according to claim 1, characterized in that The extracting features of the data after the data preprocessing to obtain polarization features comprises: Acquire VV polarization characteristic images and VH polarization characteristic images of the radar remote sensing image data, wherein the radar remote sensing image data is acquired by using a synthetic aperture radar SAR system; the VV polarization characteristic image refers to an image formed by radar echo data acquired by the SAR system in a vertical transmission and vertical reception mode; the VH polarization characteristic image refers to an image formed by radar echo data acquired by the SAR system in a vertical transmission and horizontal reception mode; Features are extracted from the VV polarization feature image and the VH polarization feature image of the radar remote sensing image data as the polarization features, and the polarization features include the average value, standard deviation, quantile, difference between the maximum and minimum values, interquartile range, skewness and kurtosis of the pixel values ​​of the VV polarization image; the average value, standard deviation, quantile, difference between the maximum and minimum values, interquartile range, skewness and kurtosis of the pixel values ​​of the VH polarization image.

4. The method according to claim 1, characterized in that: The feature extraction of the data after the data preprocessing to obtain the texture feature comprises: The red light band of the multispectral remote sensing image data after the data preprocessing is extracted, and the texture features are extracted using the gray level co-occurrence matrix. The texture features include mean, variance, homogeneity, dissimilarity, entropy and angular second moment.

5. The method according to claim 1, characterized in that The feature extraction of the data after the data preprocessing to obtain the spectral index feature includes: Extracting the green light band and the infrared band of the multispectral remote sensing image data after the data preprocessing to construct a normalized difference water body index; extracting the near infrared band and the infrared band of the multispectral remote sensing image data after the data preprocessing to construct a normalized vegetation index; extracting the blue light band, the green light band and the red light band as well as the near infrared band and the short-wave infrared band of the multispectral remote sensing image data after the data preprocessing to construct a normalized difference impervious surface index; The normalized difference water index, the normalized vegetation index and the normalized difference impervious surface index are used as the spectral index features.

6. The method according to claim 1, characterized in that The feature extraction of the data after the data preprocessing to obtain the night light index feature includes: Acquire the nighttime light remote sensing image data after the data preprocessing, wherein the nighttime light remote sensing image data after the data preprocessing includes data of high-gain and low-gain fusion bands; The night light index feature is obtained based on the normalized vegetation index obtained from the near infrared band and the infrared band of the multispectral remote sensing image data after the data preprocessing, and the data of the high and low gain fusion bands.

7. The method according to claim 1, characterized in that The feature extraction of the data after the data preprocessing to obtain the thermal infrared index feature includes: The thermal infrared remote sensing image data after the data preprocessing is calibrated by radiation, the gray value is converted into a radiation brightness value, and a normalized thermal infrared index is constructed as the thermal infrared index feature.

8. A device for identifying a surface impermeable layer, characterized in that: The device comprises: An acquisition unit, used to acquire radar remote sensing image data, multispectral remote sensing image data, night light remote sensing image data and thermal infrared remote sensing image data of the area to be studied; A data processing unit, used for performing data preprocessing so that the processed data have the same resolution; A feature extraction unit, used to extract features from the data after the data preprocessing to obtain polarization features, texture features, spectral index features, night light index features and thermal infrared index features; A feature screening unit, used to screen the features obtained by the feature extraction to obtain the top n features ranked by importance; The classification unit is used to process the n features based on a pre-trained random forest classification model to obtain a classification result of the surface type in the area to be studied.

9. A computing device, characterized in that The computing device includes: a memory and a processor; The memory is used to store computer programs; The processor is configured to implement the method according to any one of claims 1 to 7 when executing the computer program.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

Images