Urban land use automatic classification method and system

Abstract

The application provides a kind of urban land use automatic classification method and system, method includes: based on open source crowd sourcing geographic data production wide range land use training sample;Uncertainty perception context deep learning semantic segmentation network of end-to-end is constructed, and pixel-level land use classification product is obtained;Based on the segmentation of the remote sensing image to be classified by the bottom-up multi-scale segmentation algorithm, the pixel-level land use classification result is fused with the object-level segmentation boundary based on the mode vote, and the object-level land use classification product is obtained.The application is based on deep learning to automatically classify land use types, improve the classification accuracy, and save the labor cost.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of remote sensing data processing, and particularly relates to a method and system for automatically classifying urban land use. BACKGROUND

[0002] Urban land use refers to the way humans use and manage the Earth's surface, covering various land use patterns such as agriculture, urban construction, industrial development, and recreational activities. It is the result of the interaction between human social activities and the natural environment, reflecting how humans allocate and utilize land resources to meet their social, economic, and cultural needs. Land use information is crucial for understanding human impact on the environment, planning urban and rural development, managing natural resources, and developing sustainable development strategies. Therefore, the classification and continuous updating of land use information are of great significance for promoting social progress and protecting the natural environment.

[0003] Traditional land use classification techniques require a large amount of manual annotation, which is time-consuming and labor-intensive. With the advent of OpenStreetMap (OSM), detailed geographic location semantic labels are provided for large-scale land use classification. However, deep learning networks are highly dependent on the accuracy of reference labels, and since OSM is provided by volunteers, its labels and ranges are not accurate. Therefore, a deep network that can learn effective land use knowledge from noisy labels is needed. In addition, due to the large number of pooling operations in deep convolutional neural networks, the pixel-level classification results are not accurate at the edges, and the classification results are fragmented, which cannot reflect the true spatial pattern of land use. SUMMARY

[0004] To overcome the problems of existing land use classification methods, the present application proposes a method for automatically classifying urban land use, which solves the problem that existing deep networks cannot effectively handle noisy labels from OpenStreetMap (OSM) and the problem of inaccurate pixel-level classification results at the edges and fragmented classification results.

[0005] According to a first aspect of an embodiment of the present application, a land use type extraction method is provided, comprising:

[0006] Obtain Sentinel-2 remote sensing image data, and make corresponding OSM land use type training samples. The training samples are cropped and data enhanced;

[0007] Construct a deep convolutional semantic segmentation network model, and train the deep learning semantic segmentation network model using the land use type training samples;

[0008] Establish a deep convolutional network semantic segmentation model with an encoding-decoding structure, and introduce an uncertainty perception module to learn useful land use knowledge from noisy samples;

[0009] Establishing a bottom-up multi-scale segmentation to extract accurate boundary information from remote sensing images.

[0010] Fusing a deep learning semantic segmentation network model with multi-scale resolution segmentation to generate an end-to-end object-level land use classification model;

[0011] Inputting the remote sensing image data to be classified into the end-to-end land use classification model to obtain an object-level land use classification product.

[0012] Preferably, the Sentinel-2 remote sensing image data is acquired, and corresponding OSM land use type training samples are made. The training samples are cropped and data enhanced, specifically:

[0013] The Sentinel-2 data processed by cloud and radiometrically corrected is downloaded from Google Earth Engine, and the vector data is downloaded from OSM. After label cleaning, the label vector data related to land use attributes is left. Then, according to the one-to-one correspondence between the label and the land use category, the vector data is rasterized to obtain raster sample data;

[0014] The remote sensing data image and its corresponding land use sample are sliced to produce standard size land use type training samples;

[0015] The standard size land use sample data and remote sensing image are image enhanced, including rotation, cropping, contrast enhancement, etc.

[0016] Preferably, the deep convolution semantic segmentation network includes a feature encoder, a context feature extraction, and an uncertainty-aware decoder part.

[0017] The feature encoder part adopts a pre-trained ResNet-34, which contains two convolution layers and one maximum pooling layer.

[0018] The context feature extraction part includes a dense atrous convolution block and a residual multi-kernel pooling block, which is responsible for extracting context semantic information and generating higher-level feature maps.

[0019] The uncertainty-aware decoder part contains two encoders, which are the main classifier and the auxiliary classifier. Compared with the main classifier, the receptive field size of the auxiliary classifier is halved. Given a 3D input sample X∈R H×W×C , two decoders receive the same deep features F e from the encoder θ e , and then generate predicted classification probabilities P prim and F aux by the main decoder P and the auxiliary decoder F respectively.

[0020]

[0021] To solve the label noise problem, the prediction variance is calculated to model the uncertainty. The Kullback-Leibler (KL) divergence is approximated as the variance of the noisy label and the true label, which is used to measure the distance between two classifier outputs P prim and P aux . The formula is as follows:

[0022]

[0023] where p t is the ground-truth class probability of the target data. In the loss function, a variance regularization term is used to correct the result learned from the noisy label, as shown below:

[0024]

[0025] where L ce represents the CEloss. If the OSM label annotation is inaccurate, the two classifiers will produce different results, resulting in an increase in D kl . This ensures that noisy labels are not learned preferentially. When Var(p t ) is very small, i.e., the prediction results of the primary classifier and the auxiliary classifier are very close, the model will learn normally. By adding an auxiliary classifier to model uncertainty, the network can learn to distinguish between clean samples and noisy samples and assign different weights to their predictions. This helps the model to reduce the impact of noisy labels and focus more on reliable information.

[0026] Preferably, the bottom-up multi-scale segmentation module specifically comprises:

[0027] Initialize to generate superpixels, and use simple linear iterative clustering to locally aggregate pixels into extremely small image objects. The pixels inside each superpixel have similar color, brightness, or texture features. The formula for calculating the distance of a pixel to the surrounding seed points in local clustering is as follows:

[0028]

[0029] d c，j，i = (l j -l i ) 2 +(a j -a i ) 2 +(b j -b i ) 2 (5)

[0030] ds，j，i = (x j - x i ) 2 + (y j - y i ) 2 (6)

[0031] where dj,i denotes the distance between the jth pixel in the 2Sx2S neighborhood and the ith seed point, M is used to adjust the weight between spectral distance and spatial distance, generally taken as 10, d c，j，i , d s，j，i denote the spectral distance and spatial distance, l j , a j , b j , x j , y j denote the spectral value and spatial position of the jth pixel in the 2Sx2S neighborhood in CIELAB color space, l i , a i , b i , x i , y i denote the spectral value and spatial position of the ith pixel in CIELAB color space. Finally, the label of the jth pixel is assigned as label j = argmin({d j，i |i∈[0, K]}).

[0032] According to the spectral, texture, brightness and other features of the superpixels, the computational heterogeneity h between two objects is calculated as follows:

[0033] h = ω color h color + (1 - ω color )h shape (7)

[0034]

[0035] h shape = ω comlpact h compact + (1 - ω compact )h smooth (9)

[0036] h smooth = n m s m - (n obj1 s obj1 + n obj2 s obj2 ) (10)

[0037] h compact = nm c m -(n obj1 c obj1 +n obj2 c obj2 ) (11)

[0038]

[0039]

[0040] In the formula, h color and h shape represent the spectral heterogeneity and shape heterogeneity between two objects, ω color represents the spectral heterogeneity weight, B represents the band number of the image, ω b represents the weight of each band participating in segmentation, n m and σ b，m represent the pixel number and spectral standard deviation of the merged object, n obj1 , σ b，obj1 , n obj1 , σ b，obj2 respectively represent the pixel number and spectral standard deviation of the two objects before merging, h compact and h smooth represent the compactness heterogeneity and smoothness heterogeneity between two objects, ω compact represents the compactness heterogeneity weight, s m , s obj1 , s obj2 represent the smoothness of the merged object and the two objects before merging, c m , c obj1 , c obj2 represent the compactness of the merged object and the two objects before merging, l and b respectively represent the perimeter of the object and the shortest side length of the minimum bounding rectangle of the object. In the multi-resolution segmentation algorithm, 3 parameters need to be set, namely the segmentation scale scale, the spectral heterogeneity weight w color , and the compactness heterogeneity weight w compact ;

[0041] The self-downward superpixel merging is performed based on the heterogeneity to obtain a segmented object;

[0042] Preferably, the deep learning semantic segmentation network model and the multi-scale resolution segmentation fusion module are specifically as follows:

[0043] The pixel-level classification result and the object-level segmentation result are voted by mode, and on the basis of the object-level segmentation, the number of each class in each object is counted.

[0044] In the second aspect of the present application, the automatic classification system of urban land use comprises:

[0045] Sample making module: acquire Sentinel-2 remote sensing image data and corresponding OSM-based land use type training sample;

[0046] Deep convolutional network semantic segmentation module: construct a deep learning semantic segmentation network model, and train the deep learning semantic segmentation network model through the land use type training sample;

[0047] Multi-scale segmentation module: establish a bottom-up multi-scale segmentation to extract accurate boundary information from remote sensing images.

[0048] Model fusion module: fuse the deep learning semantic segmentation network model and the multi-scale resolution segmentation to generate an end-to-end object-level land use classification model;

[0049] Classification and identification module: input the remote sensing image data to be classified into the end-to-end land use classification model to obtain an object-level land use classification product.

[0050] In addition, the application also provides a storage medium, which comprises a stored program, wherein the device where the storage medium is located executes the method when the program runs.

[0051] A processor for running a program, wherein the method is executed when the program runs.

[0052] The application has the following beneficial effects relative to the prior art:

[0053] 1) The application constructs a process for automatically extracting land use samples, and the land use samples can be obtained for free based on OSM crowdsourced geographic data, without manual annotation, which is time-consuming and laborious.

[0054] 2) The application constructs a deep convolutional neural network based on uncertainty perception, improves the accuracy and robustness of urban land use extraction in the presence of label noise by modeling the uncertainty of the sample.

[0055] 3) The application combines the pixel-level deep semantic segmentation result with the image-based segmentation boundary to form an object-level land use classification result, which improves the accuracy of the land use classification result at the edge. BRIEF DESCRIPTION OF DRAWINGS

[0056] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0057] Figure 1 This is a schematic diagram of the overall process of the land use type classification method provided in an embodiment of the present invention;

[0058] Figure 2 A schematic diagram of an uncertainty-aware deep convolutional semantic segmentation network provided in an embodiment of the present invention;

[0059] Figure 3 A schematic diagram of object-oriented multi-scale bottom-up segmentation;

[0060] Figure 4 This is a schematic diagram illustrating the fusion of object-level segmentation and pixel-level classification based on majority voting. Detailed Implementation

[0061] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0062] One embodiment of the present invention provides a method for extracting land use types. Figure 1 This is a schematic diagram of the overall process of the land use type extraction method provided in an embodiment of the present invention. The method includes:

[0063] S1, acquire remote sensing image data and its corresponding OSM land use data, and create standard-sized land use samples and image pairs, specifically;

[0064] Download Sentinel-2 data after cloud processing and radiometric correction from Google Earth Engine, download vector data from OSM, clean the labels to retain the label vector data related to land use attributes, and then rasterize the vector data according to the one-to-one correspondence between the labels and land use categories to obtain raster sample data.

[0065] The remote sensing data images and their corresponding land use samples are sliced ​​to generate standard-sized land use type training samples.

[0066] Image enhancement is performed on standard-sized land use sample data and remote sensing images, including rotation, cropping, and contrast enhancement.

[0067] S2, a deep convolutional semantic segmentation network, includes a feature encoder, contextual feature extraction, and an uncertainty-aware decoder. Figure 2 Specifically;

[0068] The feature encoder part uses a pre-trained ResNet-34 and contains two convolutional layers and one max pooling layer.

[0069] The context feature extraction section, including densely dilated convolutional blocks and residual multi-kernel pooling blocks, is responsible for extracting contextual semantic information and generating higher-level feature maps. The calculation formula for the densely dilated convolutional block is as follows:

[0070]

[0071] A convolution operation is performed on the input feature map x and the filter w to generate the output y. In this formula, x[i+rk]w[k] represents the sampling points of the input feature map, w[k] is the weight in the convolution kernel, and r is the atrous rate, which corresponds to the stride of the sampled input signal. Dense dilated convolution enhances the ability to extract features by using dilated convolutions of different scales, thus enabling more effective capture of objects of different sizes and their features in an image.

[0072] The uncertainty-aware decoder comprises two encoders: a main classifier and an auxiliary classifier. The receptive field size of the auxiliary classifier is halved compared to the main classifier. Given a 3D input sample X∈R... H×W×C Two decoders from encoder θ e Receive the same depth feature F e Then, respectively by the main decoder P prim and auxiliary decoder F aux Generate predicted classification probabilities and It can be formally described as:

[0073]

[0074] To address the label noise problem, the prediction variance is calculated to model the uncertainty. The Kullback-Leibler (KL) divergence is approximated as the variance between the noisy labels and the true labels, and is used to measure the output P of the two classifiers. prim and P aux The distance between them is calculated using the following formula:

[0075]

[0076] Where, p t This represents the ground truth class probability of the target data. In the loss function, a variance regularization term is used to correct the results learned from noisy labels, as shown below:

[0077]

[0078] Where L ceThis represents CE loss. If the OSM label annotation is inaccurate, the two classifiers will produce different results, leading to D... kl The value increases. This ensures that noise labels are not learned preferentially. When Var(p) t When the uncertainty threshold is very small, meaning the predictions of the main classifier and the auxiliary classifier are very close, the model's learning will proceed normally. By incorporating an auxiliary classifier to model uncertainty, the network can learn to distinguish between clean and noisy samples and assign different weights to their predictions. This helps the model reduce the influence of noisy labels and focus more on reliable information.

[0079] S3, the bottom-up multi-scale segmentation module specifically includes:

[0080] Superpixels are initially generated, and simple linear iterative clustering is used to locally aggregate pixels into very small image objects. Pixels within each superpixel have similar color, brightness, or texture features. The formula for calculating the distance from a pixel to its surrounding seed points using local clustering is as follows:

[0081]

[0082] d c，j，i =(l j -l i ) 2 +(a j -a i ) 2 +(b j -b i ) 2 (5)

[0083] d s，j，i =(x j -x i ) 2 +(y j -y i ) 2 (6)

[0084] In the formula, represents the distance from the i-th pixel to the i-th seed point within a 2×2 neighborhood, used to adjust the weight between spectral distance and spatial distance, typically taken as 10, and d c，j，i d s，j，i Indicating spectral distance and spatial distance, l j a j b j x j y j This represents the spectral value and position of the i-th pixel in the CIELAB color space within a 2×2 neighborhood. i a i b i xi y i This represents the spectral value and position of the i-th pixel in the CIELAB color space. Finally, the label for the i-th pixel is assigned as `label`. j =argmin({d j，i |i∈[0,K)});

[0085] The heterogeneity h between two objects is calculated based on the spectral, texture, and brightness features of the superpixels. The calculation method is as follows:

[0086] h = ω color h color +(1-ω color )h shape (7)

[0087]

[0088] h shape =ω compact h compact +(1-ωc ompact )h smooth (9)

[0089] h smooth =n m s m -(n obj1 s obj1 +n obj2 s obj2 (10)

[0090] h compact =n m c m -(n obj1 c obj1 +n obj2 c obj2 (11)

[0091]

[0092] In the formula, h color and h shape ω represents the spectral heterogeneity and shape heterogeneity between two objects. color ω represents the weight of spectral heterogeneity, and ω represents the number of bands in the image. b n represents the weight of each band involved in the segmentation. m and σ b，m n represents the number of pixels and spectral standard deviation of the merged object. obj1 , σ b，obj1 n obj1 , σ b，obj2h represents the pixel count and spectral standard deviation of the two objects before merging, respectively. compact and h smooth ω represents the compactness heterogeneity and smoothness heterogeneity between two objects. compact The weights for compactness heterogeneity are represented by s. m s obj1 s obj2 c represents the smoothness of the merged object compared to the two objects before the merge. m c obj1 c obj2 This represents the compactness of the merged object compared to the two objects before merging, where represents the perimeter of the object and the shortest side length of the object's minimum bounding rectangle, respectively. In multi-resolution segmentation algorithms, three parameters need to be set: the segmentation scale, the spectral heterogeneity weight w, and so on. color Compactness heterogeneity weight w compact ;

[0093] Based on heterogeneity, bottom-up superpixel merging is performed to obtain segmented objects;

[0094] S4, the deep learning semantic segmentation network model and the multi-scale resolution segmentation fusion module are specifically as follows:

[0095] The pixel-level classification results and object-level segmentation results are subjected to a majority vote. Based on the object-level segmentation, the number of each category in each object is counted.

[0096] Corresponding to the above method embodiments, the present invention also discloses an object-oriented deep convolutional neural network land use classification system, the system comprising:

[0097] Sample creation module: Acquire Sentinel-2 remote sensing image data and corresponding OSM-based land use type training samples;

[0098] Deep Convolutional Network Semantic Segmentation Module: Constructs a deep learning semantic segmentation network model and trains the deep learning semantic segmentation network model using land use type training samples;

[0099] Multi-scale segmentation module: Establishes bottom-up multi-scale segmentation to extract accurate boundary information from remote sensing images.

[0100] Model fusion module: Fusion of deep learning semantic segmentation network model with multi-scale resolution segmentation to generate end-to-end object-level land use classification model;

[0101] Classification and recognition module: Input the remote sensing image data to be classified into the end-to-end land use classification model to obtain object-level land use classification products.

[0102] In addition, the present invention also provides a storage medium including a stored program, wherein the program controls the device where the storage medium is located to execute the method when it is running.

[0103] A processor for running a program, wherein the program executes the method during runtime.

[0104] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium includes various media that can store program code, such as ROM, RAM, magnetic disk, or optical disk.

[0105] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0106] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An automatic classification method for urban land use, characterized in that, Includes the following steps: (1) Based on OpenStreet Map (OSM) open crowdsourced geographic data, obtain sample data related to land use, and download land use type training samples and corresponding Sentinel-2 image data. (2) Construct a deep learning semantic segmentation network model and train the corresponding deep learning semantic segmentation network model using land use type training samples; (3) The image is segmented based on a bottom-up multi-scale segmentation algorithm to generate object-level segmentation results; (4) The pixel-level land use classification results obtained by the deep learning semantic segmentation network are fused with the object-level segmentation results, and the object-level land use classification results are obtained by using the majority voting mechanism.

2. The automatic urban land use classification method according to claim 1, characterized in that, Step (1) specifically includes the following sub-steps: (1.1) The urban built-up area is obtained based on urban boundary data and then cropped. Then, OSM vector data and corresponding Sentinel-2 image slices within the cropped area are obtained. The OSM vector data is processed into vector rasterization to obtain corresponding sample labels with land use attributes. (1.2) Slice the remote sensing image data and its corresponding land use sample labels to generate standard-scale land use type training samples.

3. The automatic urban land use classification method according to claim 1, characterized in that, The network model described in step (2) includes an encoding part, a context extraction module, and a decoding part; In the coding section, an uncertainty perception module is introduced to improve the accuracy and robustness of urban land use classification in the presence of label noise; The uncertainty perception module in the feature decoder consists of two different decoders: a main classifier and an auxiliary classifier. Compared to the main classifier, the receptive field size of the auxiliary classifier is halved; Given a 3D input sample X∈R H ×W×C Two decoders from encoder θ e Receive the same depth feature F e Then, respectively by the main decoder P prim and auxiliary decoder F aux Generate predicted classification probabilities fd prim and fd aux It is formally described as: To address the label noise problem, the prediction variance is calculated to model the uncertainty; the Kullback-Leibler divergence is approximated as the variance between the noisy labels and the true labels, and used to measure the output P of the two classifiers. prim and P aux The distance between them is calculated using the following formula: Where, p t This represents the ground truth class probability of the target data; in the loss function, a variance regularization term is used to correct the results learned from noisy labels, as shown below: Where L ce This represents CE loss; if the OSM label annotation is inaccurate, the two classifiers will produce different results, leading to D... kl The value increases.

4. The automatic urban land use classification method according to claim 1, characterized in that, Step (4) specifically includes: Superpixels are initialized and generated. Simple linear iterative clustering is used to locally aggregate pixels into image objects with extremely small scales. Pixels within each superpixel have similar color, brightness, or texture features. The formula for calculating the distance from a pixel to its surrounding seed points using local clustering is as follows: d c,j,i =(l j -l i ) 2 +(a j -a i ) 2 +(b j -b i ) 2 (5) d s，j，i =(x j -x i ) 2 +(y j -y i ) 2 (6) In the formula, d j，i This represents the distance from the j-th pixel to the i-th seed point within a 2S×2S neighborhood. M is used to adjust the weights between spectral distance and spatial distance, and is set to 10. c,j,i d s,j,i Indicating spectral distance and spatial distance, l j a j b j x j y j This represents the spectral value and spatial position of the j-th pixel within a 2S×2S neighborhood in the CIELAB color space. i a i b i x i y i This represents the spectral value and position of the i-th pixel in the CIELAB color space; finally, the label of the j-th pixel is assigned as `label`. j =argmin({d j，i |i∈[0,K)}); The heterogeneity h between two objects is calculated based on the spectral, texture, and brightness features of the superpixels. The calculation method is as follows: h=ω color h color +(1-ω color )h shape (7) h shape =ω compact h compact +(1-ω compact )h smooth (9) h smooth =n m s m -(n obj1 s obj1 +n obj2 s obj2 ) (10) h compact =n m c m -(n obj1 c obj1 +n obj2 c obj2 ) (11) In the formula, h color and h shape ω represents the spectral heterogeneity and shape heterogeneity between two objects. color The weights represent spectral heterogeneity, B represents the number of image bands, and ω represents the spectral heterogeneity weights. b n represents the weight of each band involved in the segmentation. m and σ b,m n represents the number of pixels and spectral standard deviation of the merged object. obj1 , σ b，obj1 n obj2 , σ b，obj2 h represents the pixel count and spectral standard deviation of the two objects before merging, respectively. compact and h smooth ω represents the compactness heterogeneity and smoothness heterogeneity between two objects. compact The weights for compactness heterogeneity are represented by s. m s obj1 s obj2 c represents the smoothness of the merged object compared to the two objects before the merge. m c obj1 c obj2 The compactness of the merged object compared to the two objects before merging is represented by l and b, respectively, which represent the perimeter of the object and the shortest side length of the object's minimum bounding rectangle. In the multi-resolution segmentation algorithm, three parameters are set: the segmentation scale (scale), the spectral heterogeneity weight (w), and the spectral scale weight (w). color Compactness heterogeneity weight w compact ; Based on heterogeneity, a bottom-up superpixel merging process is performed to obtain the object-level boundary.

5. The automatic urban land use classification method according to claim 3 or 4, characterized in that, The pixel-level classification results and object-level segmentation results are subjected to a majority vote to count the number of each category in each object. The category with the most pixels is the category of that object.

6. An automatic urban land use classification system, characterized in that, The system includes: Sample creation module: Acquire Sentinel-2 remote sensing image data and corresponding OSM-based land use type training samples; Deep Convolutional Network Semantic Segmentation Module: Constructs a deep learning semantic segmentation network model and trains the deep learning semantic segmentation network model using land use type training samples; Multi-scale segmentation module: Establishes bottom-up multi-scale segmentation to extract accurate boundary information from remote sensing images; Model fusion module: Fusion of deep learning semantic segmentation network model with multi-scale resolution segmentation to generate end-to-end object-level land use classification model; Classification and recognition module: Input the remote sensing image data to be classified into the end-to-end land use classification model to obtain object-level land use classification products.

7. A storage medium, characterized in that, The storage medium includes a stored program, wherein, when the program is executed, it controls the device on which the storage medium is located to perform the method according to any one of claims 1 to 5.

8. A processor, characterized in that, The processor is used to run a program, wherein the program, when running, performs the method according to any one of claims 1 to 5.

Images