A local climate zone classification method based on resource three remote sensing image

Abstract

The application relates to a local climate zone classification method based on ZY-3 remote sensing images, which comprises the following steps: acquiring ZY-3 image data, calculating the spectral characteristics, spatial texture characteristics and multi-angle stereo characteristics of the image, and constructing a plane-stereo characteristic set for LCZ classification; selecting samples based on visual interpretation of ZY-3 images and Google Earth images, verifying the samples through field sampling, and constructing a sample set for LCZ classification; training a random forest classifier based on the characteristic set and the sample set of the ZY-3 image, so as to obtain a classification model; inputting the image into the optimal classification model to perform LCZ classification, and obtaining LCZ classification mapping results. The application designs multi-angle stereo characteristics based on ZY-3 forward, normal and rear images, proposes four kinds of angle characteristic indexes, constructs a plane-stereo characteristic set, and is used for improving the recognition accuracy of different LCZ categories.

Description

Technical Field

[0001] This invention belongs to the interdisciplinary field of remote sensing technology and machine learning technology, and relates to a local climate zone classification method based on ZY-3 high-resolution multi-view remote sensing imagery. Background Technology

[0002] Local Climate Zones (LCZs) are a detailed land use classification scheme originally proposed for urban heat island research. They include 10 urban built environment types (LCZ1-10): compact high-rise buildings, compact mid-rise buildings, compact low-rise buildings, open high-rise buildings, open mid-rise buildings, open low-rise buildings, light low-rise buildings, large low-rise buildings, sparse buildings, and heavy industrial areas. In addition, they include 7 natural environment types (LCZA-G): dense trees, sparse trees, shrubs, low vegetation, paved surfaces, bare sand, and water. The LCZ classification scheme comprehensively considers urban building classification, functional zoning, and climate-related representations, while also possessing broad operability. Accurate and reliable LCZ classification mapping has significant application value in urban planning and management, disaster risk assessment, air quality and environmental monitoring, and climate change research.

[0003] Existing LCZ classification methods based on remote sensing technology mainly use medium-resolution satellite imagery such as Landsat and Sentinel-2. These imagery types have limited spatial resolution and are all single-view images, failing to consider the three-dimensional features of ground features. They often struggle to adequately distinguish different LCZ types, especially those meticulously subdivided based on building height and density. In complex urban scenarios, detailed, accurate, and large-scale LCZ mapping remains a significant challenge, and no effective solution currently exists. ZY-3, my country's civilian high-resolution stereo mapping satellite, can simultaneously acquire multispectral imagery and stereo images from forward-looking, frontal, and rear-looking angles. Currently, satellites 01-03 have been launched, forming a multi-satellite network with global stereo imaging capabilities. ZY-3 imagery provides clearer details and three-dimensional structural information of ground features; therefore, LCZ classification methods based on ZY-3 imagery have significant advantages over traditional methods and possess the potential for widespread application. Summary of the Invention

[0004] In view of this, the purpose of this invention is to provide a local climate zone classification method based on ZY-3 remote sensing imagery, thereby solving the technical problems existing in the background art.

[0005] To achieve the above objectives, the present invention provides the following technical solution:

[0006] A local climate zone classification method based on ZY-3 remote sensing imagery, comprising the following steps:

[0007] S1: Acquire ZY-3 image data, including multispectral, forward-looking, emphyseal, and rear-looking images, and perform preprocessing;

[0008] S2: Extract the pixel values ​​of each band from the preprocessed image in S1, calculate the spectral features, spatial texture features, and multi-angle stereo features of the image, and construct a planar-stereo feature set for LCZ classification.

[0009] S3: Based on the preprocessed images in S1, select samples by visual interpretation of Google Earth images, conduct on-site sampling to verify the samples, and construct a sample set for LCZ classification;

[0010] S4: Based on the feature set and sample set of ZY-3 imagery, a random forest classifier is used to train a classification model. The model parameters are then adjusted according to the accuracy evaluation results to obtain the optimal classification model.

[0011] S5: Input the entire image into the optimal classification model for LCZ classification to obtain the LCZ classification mapping results.

[0012] Furthermore, in S1, the ZY-3 image data includes panchromatic forward-looking, panchromatic backward-looking, panchromatic emphyseal, and multispectral emphyseal data.

[0013] Furthermore, in S1, the preprocessing of the acquired ZY-3 image data specifically includes: orthorectification, image registration, and image fusion of the panchromatic frontal, back-view, and multispectral images. The image registration involves geometrically registering the panchromatic images from the frontal, back-view, and back-view angles. The image fusion involves fusing the relatively low-resolution ZY-3 multispectral image with the relatively high-resolution ZY-3 panchromatic frontal single-band image to form a higher-resolution multispectral remote sensing image.

[0014] Furthermore, in S2, the pixel values ​​of each band include the four bands of blue, green, red, and near-infrared of the ZY-3 multispectral image, as well as the three bands of forward-looking, backward-looking, and front-looking of the ZY-3 multi-view image.

[0015] Furthermore, in S2, the spectral features of the image are selected from the Normalized Difference Vegetation Index (NDVI) and the Normalized Difference Water Index (NDWI); the spatial texture features of the image are selected from the homogeneity of the Gray-Level Co-occurrence Matrix (GLCM) and the Morphological Building Index (MBI).

[0016] Furthermore, in S2, the following are defined for the multi-angle stereo feature selection of the image: Difference Angle Index (DAI), Normalized Angle Index (NAI), Ratio Angle Index (RAI), and Maximum Normalized Angle Index (MNAI).

[0017] Furthermore, in S2, constructing the feature set for LCZ classification includes image fusion of ZY-3 multispectral image, spectral features, spatial texture features, and multi-angle stereo features to obtain the LCZ classification feature set.

[0018] Furthermore, in step S3, constructing an LCZ classification sample set based on ZY-3 high-resolution multi-view remote sensing imagery specifically includes the following steps:

[0019] S31: Divide the preprocessed multispectral frontal image into grids: the grid size is 125×125 meters, which corresponds to 50×50 pixels on the image. For each grid, calculate the average, variance, maximum, minimum, skewness and kurtosis of all pixels in it to extract features. Each grid yields a 6-dimensional feature data.

[0020] S32: Sample selection: A random sampling method is used, ensuring that each sample is at least 250 meters apart. All selected samples are divided into training and test sets in an 8:2 ratio. The training set is used to train the classifier, while the test set is used to evaluate the classifier's performance.

[0021] S33: Determination of Sample Labels: This is based on visual interpretation experience of ZY-3 multispectral imagery and Google Earth imagery; and with reference to the building outline vector data from Gaode Maps, the LCZ category of the grid is determined by the building height and density within the grid. Specifically, the height attributes (low-rise, mid-rise, high-rise) and density data (compact, open) of the buildings within the grid are calculated based on the building vector data. The specific determination process is as follows: Figure 2 As shown.

[0022] S34: Manual annotation: Mark the LCZ category on the selected sample grid and record the sample category;

[0023] S35: Field sampling: Select a portion of the study area to collect samples, record the LCZ category of the samples, and try to include all LCZ categories in the collected samples;

[0024] S36: Random verification: Randomly select samples collected in the field and compare them with samples manually labeled by visual interpretation to ensure the authenticity of the samples;

[0025] S37: Multi-person verification: Different people visually interpret and manually annotate the same sample to ensure sample objectivity;

[0026] Furthermore, in S4, the feature set includes: multispectral bands, spectral features, spatial texture features, and multi-angle stereo features. The sample set includes LCZ category labels for ZY-3 images of different cities that have been randomly selected and verified.

[0027] Furthermore, in step S4, the parameters n_estimators = 500, bootstrap = True, and max_features = 'sqrt' are set in the random forest classifier.

[0028] Furthermore, in S4, a confusion matrix is ​​used to evaluate the accuracy of the classification model, wherein the confusion matrix calculates the accuracy evaluation index by comparing the consistency between the model's prediction results and the actual values.

[0029] Furthermore, in S5, the optimal classification model refers to the model with the best overall accuracy in the accuracy evaluation results, representing the overall classification performance of the model.

[0030] The beneficial effects of this invention are as follows:

[0031] First, this invention makes full use of ZY-3 multi-view imagery, which can provide clearer details of ground features and three-dimensional structural information. The LCZ classification method based on ZY-3 imagery has more advantages than traditional methods and has the potential for wide application.

[0032] Secondly, the multi-angle stereo features proposed in this invention, including DAI, NAI, RAI, and MNAI, are simple and fast to calculate and can reflect the three-dimensional structural features of the city. This helps to reduce the problems of missed or incorrect classification caused by building occlusion, low brightness or contrast, shadows, etc. In addition, the introduction of three-dimensional features plays an important role in distinguishing building types and natural environment types of different heights, thereby effectively improving the accuracy of the model for LCZ classification.

[0033] Third, by comprehensively utilizing multi-angle stereo features and planar features (including spectral features and spatial texture features), this invention can capture information from remote sensing images more comprehensively, thereby improving the accuracy of LCZ classification. This planar-stereo feature fusion method is more advantageous than methods that rely solely on a single feature.

[0034] Other advantages, objectives, and features of the invention will be set forth in part in the description which follows, and in part will be apparent to those skilled in the art from the following examination, or may be learned from practice of the invention. The objectives and other advantages of the invention can be realized and obtained through the following description. Attached Figure Description

[0035] To make the objectives, technical solutions, and advantages of the present invention clearer, the preferred embodiments of the present invention will be described in detail below with reference to the accompanying drawings, wherein:

[0036] Figure 1 This is a flowchart of the present invention;

[0037] Figure 2 Flowchart for determining the LCZ category of a mesh based on building outline vector data;

[0038] Figure 3 A schematic diagram of the LCZ classification results (Guangzhou); Detailed Implementation

[0039] The following specific examples illustrate the implementation of the present invention. Those skilled in the art can easily understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the illustrations provided in the following embodiments are only schematic representations of the basic concept of the present invention. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0040] The accompanying drawings are for illustrative purposes only and are schematic diagrams, not actual pictures. They should not be construed as limiting the invention. To better illustrate the embodiments of the invention, some parts in the drawings may be omitted, enlarged, or reduced, and do not represent the actual product dimensions. It is understandable to those skilled in the art that some well-known structures and their descriptions may be omitted in the drawings.

[0041] In the accompanying drawings of the embodiments of the present invention, the same or similar reference numerals correspond to the same or similar components. In the description of the present invention, it should be understood that if terms such as "upper," "lower," "left," "right," "front," and "rear" indicate the orientation or positional relationship based on the orientation or positional relationship shown in the drawings, they are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, the terms used to describe positional relationships in the drawings are only for illustrative purposes and should not be construed as limiting the present invention. For those skilled in the art, the specific meaning of the above terms can be understood according to the specific circumstances.

[0042] Please see Figures 1-2 This is a local climate zone classification method based on ZY-3 remote sensing imagery.

[0043] This embodiment proposes a local climate zone classification method based on ZY-3 remote sensing imagery. The functions implemented by this method can be achieved by calling program code in a computer device's processor, mainly using Python programming language.

[0044] This embodiment proposes a classification method for multi-view remote sensing images, taking five Chinese cities—Beijing, Chongqing, Guangzhou, Kunming, and Wuhan—with different geographical conditions and social environments as the study area. Figure 1This is a schematic diagram illustrating the implementation process of a local climate zone classification method based on ZY-3 remote sensing imagery according to an embodiment of the present invention. Figure 1 As shown, the method includes:

[0045] Step S1: Acquire ZY-3 image data and then perform image preprocessing.

[0046] In this embodiment, ZY-3 image data refers to ground observation image data acquired by satellite sensors, which is transmitted to the ground receiving system through a certain transmission medium and then undergoes preliminary processing to become usable image data.

[0047] In this embodiment, the ZY-3 image data includes four types of images: multispectral frontal view, panchromatic frontal view, panchromatic forward view, and panchromatic back view.

[0048] In this embodiment, image preprocessing includes orthorectification, image registration, and image fusion. Orthorectification refers to orthorectifying panchromatic, front-view, back-view, and multispectral images to more accurately locate them within the Earth's geographic coordinate system. Image registration involves registering images to ensure that remote sensing data from different sources, at different times, and at different resolutions are aligned within the same geographic coordinate system, thus providing a consistent geospatial reference for various applications and analyses. Image fusion involves resampling relatively low-resolution ZY-3 panchromatic multispectral images with relatively high-resolution ZY-3 panchromatic single-band images to generate higher-resolution multispectral remote sensing images.

[0049] Step S2: Extract the pixel values ​​of each band from the preprocessed image in S1, calculate the spectral features, spatial texture features, and multi-angle stereo features of the image, and construct the feature set for LCZ classification.

[0050] In this embodiment, the pixel values ​​of each band include the four bands of blue, green, red and near-infrared of the ZY-3 multispectral image, as well as the three bands of forward view, backward view and front view of the ZY-3 multi-view image.

[0051] In this embodiment, spectral features include the Normalized Difference Vegetation Index (NDVI) and the Normalized Difference Water Index (NDWI), used to enhance the representation of typical categories such as vegetation and water bodies in the image. Spatial texture features include homogeneity calculated by the Gray-Level Co-occurrence Matrix (GLCM) and the Morphological Building Index (MBI), used to enhance the spatial texture distinction of ground features and architectural characteristics.

[0052] In this embodiment, the formula for calculating the normalized vegetation index is as follows:

[0053]

[0054] The formula for calculating the normalized water index is as follows:

[0055]

[0056] Wherein, NIR is the reflectance in the near-infrared band, R is the reflectance in the red band, and G is the reflectance in the green band.

[0057] In this embodiment, GLCM can derive various texture features. This embodiment selects homogeneity as the texture volume feature, and the calculation formula is as follows:

[0058]

[0059] Where N is the number of gray levels in the GLCM, and p(i,j) is the value of the element in the i-th row and j-th column of the GLCM matrix.

[0060] The formula for calculating the morphological architectural index is as follows:

[0061]

[0062] Where DMP stands for Differential Morphology Profile, WTH is the white hat operation, s and d represent size / scale and orientation, and Ns and N d These represent the total number of scales and directions, respectively.

[0063] In this embodiment, the multi-angle stereo features are defined as follows: Difference Angular Index (DAI), Normalized Angular Index (NAI), Ratio Angular Index (RAI), and Maximum Normalized Angular Index (MNAI). These four index features are all calculated based on the ZY-3 forward-looking, front-looking, and rear-looking images.

[0064] In this embodiment, the formula for calculating the difference angle index (DAI) is as follows:

[0065] DAI = max / min / mean / var{|x n -x f |,|x n -x b |,|x f -x b |} (5)

[0066] The formula for calculating the Normalized Angle Index (NAI) is as follows:

[0067]

[0068] The formula for calculating the Ratio Angle Index (RAI) is as follows:

[0069]

[0070] The formula for calculating the maximum normalized angle index MNAI is as follows:

[0071]

[0072] Where, x n x represents the pixel value of the frontal image. f x represents the pixel value of the forward-looking image. b This represents the pixel value of the rear-view image.

[0073] In this embodiment, the planar-stereo feature set of LCZ classification includes ZY-3 multispectral bands, spectral features, spatial texture features, and multi-angle stereo features.

[0074] Step S3: Based on the preprocessed images in S1, select samples using visual interpretation of Google Earth images, conduct on-site sampling to verify the samples, and construct a sample set for LCZ classification.

[0075] In this embodiment, the LCZ classification sample set based on ZY-3 high-resolution multi-view remote sensing imagery is constructed, specifically including the following steps:

[0076] S31: Divide the preprocessed multispectral frontal image into grids: the grid size is 125×125 meters, which corresponds to 50×50 pixels on the image. For each grid, calculate the average, variance, maximum, minimum, skewness and kurtosis of all pixels in it to extract features. Each grid yields a 6-dimensional feature data.

[0077] S32: Sample selection: A random sampling method is used, ensuring that each sample is at least 250 meters apart. All selected samples are divided into training and test sets in an 8:2 ratio. The training set is used to train the classifier, while the test set is used to evaluate the classifier's performance.

[0078] S33: Determination of Sample Labels: This is based on visual interpretation experience of ZY-3 multispectral imagery and Google Earth imagery; and with reference to the building outline vector data from Gaode Maps, the LCZ category of the grid is determined by the building height and density within the grid. Specifically, the height attributes (low-rise, mid-rise, high-rise) and density data (compact, open) of the buildings within the grid are calculated based on the building vector data. The specific determination process is as follows: Figure 2 As shown.

[0079] S34: Manual annotation: Mark the LCZ category on the selected sample grid and record the sample category;

[0080] S35: Field sampling: Select a portion of the study area to collect samples, record the LCZ category of the samples, and try to include all LCZ categories in the collected samples;

[0081] S36: Random verification: Randomly select samples collected in the field and compare them with samples manually labeled by visual interpretation to ensure the authenticity of the samples;

[0082] S37: Multi-person verification: Different people visually interpret and manually annotate the same sample to ensure sample objectivity;

[0083] Step S4: Based on the feature set and sample set of ZY-3 imagery, a random forest classifier is used to train the classification model. The model parameters are then adjusted according to the accuracy evaluation results to obtain the optimal classification model.

[0084] In this embodiment, the feature set includes: multispectral bands, spectral features, spatial texture features, and multi-angle stereo features. The sample set includes LCZ category labels for ZY-3 images of different cities that have been randomly selected and verified.

[0085] In this embodiment, the random forest classifier includes constructing multiple decision trees, each trained on a randomly sampled subset, and randomly selecting features at each node. Finally, the prediction results of multiple trees are combined through voting or averaging.

[0086] In this embodiment, the random forest classifier includes feature importance calculation, which comprehensively considers the contribution of each decision tree to each feature, as well as the impact of the feature on the model's predictive performance, and analyzes feature importance. This allows for feature selection, model interpretation, and identification of key factors.

[0087] In this embodiment, the random forest classifier is configured with parameters n_estimators = 500; bootstrap = True; max_features = 'sqrt'. Here, n_estimators determines the number of trees in the forest; bootstrap determines that each tree in the forest will be trained on bootstrap samples of the training data (sampling with replacement); and max_features determines the square root of the total number of features to be considered when searching for the optimal split.

[0088] In this embodiment, accuracy evaluation includes calculating accuracy evaluation metrics by comparing the consistency between the model's predictions and actual values ​​using the confusion matrix. These accuracy evaluation metrics include overall accuracy (OA), cartographic accuracy (PA), user accuracy (UA), and F1-Score.

[0089] In this embodiment, the formula for calculating the overall accuracy is as follows:

[0090]

[0091] The formula for calculating cartographic accuracy is as follows:

[0092]

[0093] The formula for calculating user accuracy is shown below:

[0094]

[0095] The formula for calculating F1-Score is as follows:

[0096]

[0097] Where P is the total number of samples, P ii P ij P ji It is the value in the i-th row and j-th column of the confusion matrix.

[0098] Step S5: Input the entire image into the optimal classification model for LCZ classification to obtain the LCZ classification mapping result.

[0099] In this embodiment, the optimal classification model includes the model with the best overall accuracy in the accuracy evaluation results, representing the overall classification performance of the model.

[0100] In this embodiment, the LCZ classification results of the image are predicted to obtain the final city LCZ classification mapping result, such as... Figure 3 As shown.

[0101] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for classifying local climate zones based on ZY-3 remote sensing imagery, characterized in that, Includes the following steps: S1: Acquire ZY-3 image data, which includes panchromatic forward-looking image, panchromatic backward-looking image, panchromatic frontal image, and multispectral frontal image. Perform orthorectification on the panchromatic frontal image, panchromatic forward-looking image, panchromatic backward-looking image, and multispectral frontal image. Perform geometric registration on the panchromatic frontal image, panchromatic forward-looking image, and panchromatic backward-looking image. Then, fuse the relatively low-resolution ZY-3 multispectral frontal image with the relatively high-resolution ZY-3 panchromatic frontal single-band image to generate a multispectral image with a spatial resolution of 2.5 meters. S2: Extract the pixel values ​​of each band in the preprocessed image, calculate the spectral features, spatial texture features and multi-angle stereo features of the image, and fuse the ZY-3 multispectral image, the spectral features, the spatial texture features and the multi-angle stereo features to construct a planar-stereo feature set for local climate zone LCZ classification. The multi-angle stereo features include the difference angle index DAI, the normalized angle index NAI, the ratio angle index RAI, and the maximum value normalized angle index MNAI. make This represents the pixel value of the frontal image. This represents the pixel value of the forward-looking image. Indicates the pixel value of the rearview image; The Difference Angle Index (DAI) includes the maximum, minimum, mean, and variance calculated for the following three angle differences: ; The Normalized Angle Index (NAI) includes the maximum, minimum, mean, and variance calculated for the following three normalized angle differences: ; The Ratio Angle Index (RAI) includes the maximum, minimum, mean, and variance calculated for the following six angle ratios: ; The maximum normalized angle index MNAI includes the maximum value, minimum value, mean, and variance calculated from the following three maximum normalized angle differences: ; S3: Construct a sample set for LCZ classification based on preprocessed ZY-3 images, specifically including: S31: Divide the preprocessed multispectral frontal image into a 125×125 meter grid, with each grid corresponding to 50×50 pixels. Calculate the average, variance, maximum, minimum, skewness, and kurtosis of all pixel values ​​in each grid, so that each grid obtains a six-dimensional feature data. S32: Select samples using a random sampling method, ensuring that adjacent samples are at least 250 meters apart, and divide the selected samples into training and test sets in an 8:2 ratio; S33: Visually interpret the ZY-3 multispectral imagery and Google Earth imagery, and determine sample labels by referring to building outline vector data; correlate the building outline vector data with the grid, and count the number of pixels corresponding to low-rise, mid-rise, and high-rise buildings based on the floor information of the buildings within the grid, and determine the height attribute of the grid by the building height category corresponding to the building with the largest number of pixels; count the number of pixels containing buildings within the grid. s and the total number of pixels in the grid total ,according to m = s / total Calculate building density m And according to the building density m Determine the density attribute of the grid; combine the height attribute and the density attribute, and combine them with the visual interpretation results to determine the LCZ category to which the grid belongs, wherein the height attribute is low-level, mid-level or high-level, and the density attribute is compact or open; S34: Mark the LCZ category on the selected sample grid and record the sample category; S35: Conduct field sampling in the study area and record the LCZ category to which the field-collected samples belong, so that the field-collected samples include all LCZ categories; S36: Randomly select field-collected samples and compare them with manually labeled samples through visual interpretation for verification; S37: Visual interpretation and manual annotation of the same sample by different personnel to verify the sample category; S4: Based on the planar-stereo feature set and the sample set, a random forest classifier is used for training to obtain a classification model. The parameters of the classification model are adjusted according to the accuracy evaluation results to obtain the optimal classification model. S5: Input the ZY-3 image to be classified into the optimal classification model to perform LCZ classification and obtain the LCZ classification mapping result.

2. The method for classifying local climate zones based on ZY-3 remote sensing imagery according to claim 1, characterized in that: In S2, the pixel values ​​of each band include the blue band, green band, red band and near-infrared band of the ZY-3 multispectral image, as well as the forward-looking band, backward-looking band and front-looking band of the ZY-3 multi-view image.

3. The method for classifying local climate zones based on ZY-3 remote sensing imagery according to claim 1, characterized in that: In S2, the spectral features include the Normalized Difference Vegetation Index (NDVI) and the Normalized Difference Water Index (NDWI), and the spatial texture features include the homogeneity of the Gray-Level Co-occurrence Matrix (GLCM) and the Morphological Building Index (MBI).

4. The method for classifying local climate zones based on ZY-3 remote sensing imagery according to claim 1, characterized in that: In step S4, the random forest classifier is configured with parameters n_estimators=500, bootstrap=True, and max_features='sqrt'. Here, n_estimators is used to determine the number of decision trees in the random forest, bootstrap is used to determine that each decision tree is trained using samples with replacement from the training data, and max_features is used to determine that the number of features considered when finding the optimal split is the square root of the total number of features.

5. The method for classifying local climate zones based on ZY-3 remote sensing imagery according to claim 1, characterized in that: In S4, the sample set includes LCZ category labels for ZY-3 images of different cities that have been randomly selected and verified.

6. The method for classifying local climate zones based on ZY-3 remote sensing imagery according to claim 1, characterized in that: In step S4, a confusion matrix is ​​used to evaluate the accuracy of the classification model. The confusion matrix calculates an accuracy evaluation index by comparing the consistency between the prediction results of the classification model and the actual category labels. The accuracy evaluation index includes overall accuracy OA, mapping accuracy PA, user accuracy UA, and F1-Score.

7. The method for classifying local climate zones based on ZY-3 remote sensing imagery according to claim 1, characterized in that: In step S4, the random forest classifier calculates the feature importance of each feature in the planar-stereo feature set, and determines the importance of each feature by combining the contribution of each feature in each decision tree and the impact of each feature on the prediction performance of the classification model.

8. The method for classifying local climate zones based on ZY-3 remote sensing imagery according to claim 1, characterized in that: In S5, the optimal classification model is the classification model with the highest overall accuracy (OA) in the accuracy evaluation results.

Images