Urban land intelligent classification method and system based on remote sensing technology

Abstract

The invention discloses an urban land intelligent classification method and system based on a remote sensing technology. The method comprises the following steps: acquiring remote sensing image data covering the whole region of a city; based on the remote sensing image data, extracting a ground feature feature vector set; based on the ground feature vector set, obtaining a preliminary classification result set; and according to the preliminary classification result set, introducing ground survey data as auxiliary verification information, and obtaining a final classification result. According to the method, the precision of urban land classification is remarkably improved, the classification efficiency is greatly improved, the requirement for refined land identification in urban planning and management can be better met, and powerful technical support is provided for urban sustainable development.

Description

Technical Field

[0001] The present invention relates to the field of urban planning, and in particular to a remote sensing technology-based intelligent classification method and system for urban land. Background Art

[0002] The identification of urban land use types is a crucial research area in urban planning and management, directly related to the rational allocation of urban resources and sustainable development. With the acceleration of urbanization, accurately understanding the functions and uses of different areas within cities has become key to improving the efficiency of urban governance. However, traditional identification methods often rely on ground surveys or single image data, which suffer from limited coverage, long update cycles, and insufficient classification accuracy. This makes it difficult to distinguish high-density areas and diverse land features, especially in complex urban environments.

[0003] In this context, the main limitation of current methods lies in their inadequate in-depth exploration of urban land features, particularly their weak ability to perform detailed classification in complex environments. This often leads to ambiguous recognition results, making it difficult to meet the needs of refined management. Furthermore, the surface structure and geometric characteristics of land features in urban environments vary significantly. Failure to effectively capture this characteristic information directly impacts classification accuracy. The acquisition of this characteristic information is closely related to the depth and method of signal analysis. Failure to comprehensively analyze multi-dimensional signal data will inevitably exacerbate the difficulty of distinguishing different land use types, particularly in demarcating the boundaries between high-density built-up areas and open spaces, creating a technical bottleneck that urgently needs to be overcome.

[0004] Therefore, how to fully explore the structural and geometric characteristics of the surface of the object through innovative technical means, and combine the analysis method of multi-dimensional signal data to achieve high-precision identification of urban land use types has become a key issue that needs to be urgently solved in current research. Summary of the Invention

[0005] To solve the technical problems in the above background, the present invention provides an intelligent classification method for urban land based on remote sensing technology, comprising the following steps:

[0006] Obtain remote sensing image data covering the entire city;

[0007] Extracting a ground feature vector set based on the remote sensing image data;

[0008] Based on the ground feature vector set, a preliminary classification result set is obtained;

[0009] Based on the preliminary classification result set, ground survey data is introduced as auxiliary verification information to obtain the final classification result.

[0010] Preferably, multi-source remote sensing technology is used to obtain the remote sensing image data covering the entire city; after the remote sensing image data is obtained, filtering technology is used to remove noise, and geometric correction technology is used to correct image distortion caused by the sensor angle.

[0011] Preferably, based on the remote sensing image data, the image data is segmented using preset rules to determine the regional distribution of land objects; by analyzing the regional distribution of land objects, geometric characteristic data is extracted, and the segmented image is deeply processed using a convolutional neural network to obtain feature decomposition results.

[0012] Preferably, the convolutional neural network includes an input layer, two convolutional layers, two maximum pooling layers, a fully connected layer, and an output layer; wherein the two convolutional layers respectively include 16 and 32 3×3 convolution kernels with a stride of 1; the stride of the maximum pooling layer is 2, and the fully connected layer includes 256 neurons; the activation function adopts the ReLU function;

[0013] The convolution layer captures the structural features of the ground object through the local receptive field, and the deep convolution kernel learns the high-level semantic features:

[0014]

[0015] Where X represents the input image; C represents the number of channels; W represents the convolution kernel weight; b k represents the bias term; Y i，j，k Represents the value of the output feature map at position (i, j) and the kth channel; ReLU represents the activation function; c represents the channel index; p represents the height direction index; q represents the width.

[0016] Preferably, based on the feature vector set of the land features, a random forest classification algorithm is used to perform preliminary classification of the land features, and feature matching is performed on the boundary division of high-density areas and open spaces to obtain the preliminary classification result set; when there are overlapping classification areas, the final attribution is determined by comparative analysis of texture information.

[0017] Preferably, the method for obtaining the final classification result includes:

[0018] Performing preliminary screening based on the preliminary classification result set;

[0019] According to the screened areas, corresponding survey data are obtained from the ground survey;

[0020] Matching the survey data with the initial classification results in a data combination manner to obtain a preliminary fused data set;

[0021] Correction and comparison are performed on the preliminary fusion data set to obtain the final classification result.

[0022] The present invention also provides an urban land intelligent classification system based on remote sensing technology, the system is used to implement the above method, including: a collection module, an extraction module, a construction module and a classification module;

[0023] The acquisition module is used to obtain remote sensing image data covering the entire city;

[0024] The extraction module is used to extract a ground feature vector set based on the remote sensing image data;

[0025] The construction module is used to obtain a preliminary classification result set based on the ground feature vector set;

[0026] The classification module is used to introduce ground survey data as auxiliary verification information based on the preliminary classification result set to obtain a final classification result.

[0027] Preferably, the acquisition module uses multi-source remote sensing technology to obtain the remote sensing image data covering the entire city; after the remote sensing image data is obtained, filtering technology is used to remove noise, and geometric correction technology is used to correct image distortion caused by the sensor angle.

[0028] Compared with the prior art, the present invention has the following beneficial effects:

[0029] The present invention not only significantly improves the accuracy of urban land classification, but also greatly improves the classification efficiency. It can better meet the needs of refined land identification in urban planning and management, and provide strong technical support for sustainable urban development. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solution of the present invention, the following briefly introduces the drawings required for use in the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0031] Figure 1 Schematic diagram of a method flow in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0033] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0034] Example 1

[0035] like Figure 1 FIG. 1 is a flow chart of the method of this embodiment, and the steps include:

[0036] S1. Obtain remote sensing image data covering the entire city.

[0037] Acquire high-resolution image data covering the entire city, conduct preliminary scanning of the distribution of land features in high-density areas and complex environments, and obtain remote sensing image data containing a variety of land feature information.

[0038] Using a variety of methods, including satellite remote sensing, aerial photography, and drone imagery, we acquire image data with a resolution of 0.5 meters to form an initial image set. This data covers various urban features, including buildings, roads, and green spaces, laying the foundation for subsequent analysis. Images of this resolution capture detailed features, helping to improve the accuracy of feature classification. For preprocessing of remote sensing image data, filtering techniques are used to remove noise, such as median filtering to eliminate random noise in the image. Geometric correction techniques are also used to correct image distortion caused by sensor angle.

[0039] S2. Extract feature vector sets based on remote sensing image data.

[0040] For the collected remote sensing image data, the structural features and geometric characteristics of the surface of the objects are extracted, and the convolutional neural network in the deep learning model is used to decompose the image data to obtain the feature vector set of the objects.

[0041] The system identifies structural features on the surface of objects and segments the image data using pre-set rules to determine the regional distribution of the objects. By analyzing the regional distribution of the objects, geometric characteristic data is extracted. The segmented images are then deeply processed using a convolutional neural network to obtain feature decomposition results. Based on the feature decomposition results, a preliminary description of the object's features is generated by combining structural and geometric characteristics. The feature is then determined to determine whether it meets a pre-set threshold. If the feature value is below the threshold, the image data is corrected to obtain an optimized feature description.

[0042] The convolutional neural network described above consists of an input layer, two convolutional layers, two max-pooling layers, a fully connected layer, and an output layer. The two convolutional layers contain 16 and 32 3×3 convolution kernels, respectively, with a stride of 1. The max-pooling layer has a stride of 2, and the fully connected layer has 256 neurons. The activation function is the ReLU function.

[0043] The convolution layer captures the structural features of the ground objects (such as edges and textures) through the local receptive field; the deep convolution kernel learns high-level semantic features (such as building outlines and road directions):

[0044]

[0045] Where X represents the input image; C represents the number of channels; W represents the convolution kernel weight; b k represents the bias term; Y i，j，k Represents the value of the output feature map at position (i, j) and the kth channel; ReLU represents the activation function; c represents the channel index; p represents the height direction index; q represents the width.

[0046] S3. Based on the feature vector set of the ground objects, a preliminary classification result set is obtained.

[0047] The random forest classification algorithm is used to preliminarily classify the features based on the feature vector set, and feature matching is performed on the boundary division of high-density areas and open spaces to obtain a preliminary classification result set.

[0048] Using feature features and vector sets, height and texture information are extracted from multidimensional data. A random forest classification algorithm is then used to perform a preliminary classification of features, generating an initial classification dataset. Based on the initial classification dataset, regional segmentation is performed based on the distribution characteristics of high-density areas and open spaces, generating a regional distribution layer. Using the regional distribution layer, feature matching is performed on the boundary between high-density areas and open spaces, combining height and texture information, to determine a set of boundary feature points. If the similarity of feature matching within the boundary feature point set falls below a preset threshold, a secondary extraction is performed on the multidimensional data of the boundary area to obtain a revised boundary feature dataset. The revised boundary feature dataset is then integrated with the preliminary classification results to optimize the accuracy of boundary delineation, generating an optimized classification result set. Based on the optimized classification result set, if there is overlap between the distribution characteristics of high-density areas and open spaces, texture information is compared and analyzed to determine the final classification assignment, generating a final classification layer. Using the final classification layer, feature features and multidimensional data are combined to generate a complete feature classification distribution dataset, confirming the final feature classification result.

[0049] When processing feature vectors and vector sets, height and texture information can be extracted from multidimensional data. Height information typically reflects the vertical distribution of features, while texture information reflects the roughness or regularity of the feature surface. For example, in imagery data of an urban area, height information can be obtained using LiDAR data, revealing the distribution of building heights ranging from 5 to 50 meters. Texture information, through high-resolution image analysis, can identify differences in building surface smoothness, such as the difference between a glass curtain wall and a brick wall.

[0050] When the random forest classification algorithm is used to perform preliminary classification of land objects, the land objects are divided into three categories: buildings, roads and green spaces based on height and texture features.

[0051] In one possible implementation, a random forest constructs multiple decision trees to vote on each feature. Assuming 800 of 1,000 sample points are correctly classified as buildings, the classification accuracy is high. This method is suitable for processing multidimensional data and can better adapt to the distribution of complex land features.

[0052] To segment high-density areas and open space, we generated a regional distribution layer using the initial classification dataset. We assumed that high-density areas primarily comprised urban centers, with buildings accounting for over 70%, while open space consisted of parks or vacant land, accounting for approximately 20%. Using a segmentation algorithm, we separated these two areas, creating a clear distribution layer for subsequent analysis.

[0053] In feature matching for boundary demarcation, height and texture information are combined to determine a set of boundary feature points. For example, at the junction of a high-density area and open space, a point with a sudden change in height is used as a feature point. If the matching similarity falls below a preset threshold of 80%, a secondary extraction of the boundary area data is required. If there are areas of overlapping classifications, texture information comparison and analysis are used to determine the final classification. For example, if an area is classified as both green space and road, and texture roughness analysis reveals that it more closely resembles green space characteristics, it will ultimately be classified as green space.

[0054] S4. Based on the preliminary classification result set, introduce ground survey data as auxiliary verification information to obtain the final classification result.

[0055] By obtaining a set of results from the initial classification, a preliminary screening is performed for areas with insufficient classification accuracy, identifying the complex environmental areas requiring adjustment. Based on these selected complex environmental areas, corresponding survey data is obtained from ground surveys. Using data fusion, the survey data is matched with the initial classification results to generate a preliminary fused dataset. If the classification accuracy of the preliminary fused dataset deviates from a preset threshold, feature adjustment methods are used to correct the classification features within the complex environmental area, resulting in a corrected feature dataset. Based on this corrected feature dataset, the classification results for the complex environmental area are recalculated using a support vector machine algorithm to generate an updated set of classification results. For areas within the updated set of classification results where accuracy is still insufficient, if the survey data coverage reaches a preset percentage, data fusion is performed again to generate a secondary fused dataset. Based on this secondary fused dataset, conventional comparison tools are used to verify the consistency of the classification results with the survey data, determining whether the final adjusted results meet the expected standards, and finally generating the final set of classification results. The final set of classification results is then used to verify the classification accuracy of the complex environmental area. If local deviations exist, additional survey data is supplemented through data fusion to determine the final stable classification output.

[0056] Example 2

[0057] This embodiment also provides an intelligent urban land classification system based on remote sensing technology, including: an acquisition module, an extraction module, a construction module and a classification module; the acquisition module is used to obtain remote sensing image data covering the entire city; the extraction module is used to extract a set of ground feature vectors based on the remote sensing image data; the construction module is used to obtain a preliminary classification result set based on the ground feature vector set; the classification module is used to introduce ground survey data as auxiliary verification information based on the preliminary classification result set to obtain the final classification result.

[0058] The following will describe in detail how the present invention solves technical problems in practical work in conjunction with this embodiment.

[0059] First, the acquisition module is used to obtain high-resolution image data covering the entire city, and a preliminary scan is performed on the distribution of objects in high-density areas and complex environments to obtain remote sensing image data containing a variety of object information.

[0060] Using a variety of methods, including satellite remote sensing, aerial photography, and drone imagery, we acquire image data with a resolution of 0.5 meters to form an initial image set. This data covers various urban features, including buildings, roads, and green spaces, laying the foundation for subsequent analysis. Images of this resolution capture detailed features, helping to improve the accuracy of feature classification. For preprocessing of remote sensing image data, filtering techniques are used to remove noise, such as median filtering to eliminate random noise in the image. Geometric correction techniques are also used to correct image distortion caused by sensor angle.

[0061] Then the extraction module extracts the feature vector set of the ground objects based on the remote sensing image data.

[0062] For the collected remote sensing image data, the structural features and geometric characteristics of the surface of the objects are extracted, and the convolutional neural network in the deep learning model is used to decompose the image data to obtain the feature vector set of the objects.

[0063] The system identifies structural features on the surface of objects and segments the image data using pre-set rules to determine the regional distribution of the objects. By analyzing the regional distribution of the objects, geometric characteristic data is extracted. The segmented images are then deeply processed using a convolutional neural network to obtain feature decomposition results. Based on the feature decomposition results, a preliminary description of the object's features is generated by combining structural and geometric characteristics. The feature is then determined to determine whether it meets a pre-set threshold. If the feature value is below the threshold, the image data is corrected to obtain an optimized feature description.

[0064] The convolutional neural network described above consists of an input layer, two convolutional layers, two max-pooling layers, a fully connected layer, and an output layer. The two convolutional layers contain 16 and 32 3×3 convolution kernels, respectively, with a stride of 1. The max-pooling layer has a stride of 2, and the fully connected layer has 256 neurons. The activation function is the ReLU function.

[0065] The convolution layer captures the structural features of the ground objects (such as edges and textures) through the local receptive field; the deep convolution kernel learns high-level semantic features (such as building outlines and road directions):

[0066]

[0067] Where X represents the input image; C represents the number of channels; W represents the convolution kernel weight; b k represents the bias term; Y i，j，k Represents the value of the output feature map at position (i, j) and the kth channel; ReLU represents the activation function; c represents the channel index; p represents the height direction index; q represents the width.

[0068] The construction module obtains a preliminary classification result set based on the feature vector set of the ground objects.

[0069] The random forest classification algorithm is used to preliminarily classify the features based on the feature vector set, and feature matching is performed on the boundary division of high-density areas and open spaces to obtain a preliminary classification result set.

[0070] Using feature features and vector sets, height and texture information are extracted from multidimensional data. A random forest classification algorithm is then used to perform a preliminary classification of features, generating an initial classification dataset. Based on the initial classification dataset, regional segmentation is performed based on the distribution characteristics of high-density areas and open spaces, generating a regional distribution layer. Using the regional distribution layer, feature matching is performed on the boundary between high-density areas and open spaces, combining height and texture information, to determine a set of boundary feature points. If the similarity of feature matching within the boundary feature point set falls below a preset threshold, a secondary extraction is performed on the multidimensional data of the boundary area to obtain a revised boundary feature dataset. The revised boundary feature dataset is then integrated with the preliminary classification results to optimize the accuracy of boundary delineation, generating an optimized classification result set. Based on the optimized classification result set, if there is overlap between the distribution characteristics of high-density areas and open spaces, texture information is compared and analyzed to determine the final classification assignment, generating a final classification layer. Using the final classification layer, feature features and multidimensional data are combined to generate a complete feature classification distribution dataset, confirming the final feature classification result.

[0071] When processing feature vectors and vector sets, height and texture information can be extracted from multidimensional data. Height information typically reflects the vertical distribution of features, while texture information reflects the roughness or regularity of the feature surface. For example, in imagery data of an urban area, height information can be obtained using LiDAR data, revealing the distribution of building heights ranging from 5 to 50 meters. Texture information, through high-resolution image analysis, can identify differences in building surface smoothness, such as the difference between a glass curtain wall and a brick wall.

[0072] When the random forest classification algorithm is used to perform preliminary classification of land objects, the land objects are divided into three categories: buildings, roads and green spaces based on height and texture features.

[0073] In one possible implementation, a random forest constructs multiple decision trees to vote on each feature. Assuming 800 of 1,000 sample points are correctly classified as buildings, the classification accuracy is high. This method is suitable for processing multidimensional data and can better adapt to the distribution of complex land features.

[0074] To segment high-density areas and open space, we generated a regional distribution layer using the initial classification dataset. We assumed that high-density areas primarily comprised urban centers, with buildings accounting for over 70%, while open space consisted of parks or vacant land, accounting for approximately 20%. Using a segmentation algorithm, we separated these two areas, creating a clear distribution layer for subsequent analysis.

[0075] In feature matching for boundary demarcation, height and texture information are combined to determine a set of boundary feature points. For example, at the junction of a high-density area and open space, a point with a sudden change in height is used as a feature point. If the matching similarity falls below a preset threshold of 80%, a secondary extraction of the boundary area data is required. If there are areas of overlapping classifications, texture information comparison and analysis are used to determine the final classification. For example, if an area is classified as both green space and road, and texture roughness analysis reveals that it more closely resembles green space characteristics, it will ultimately be classified as green space.

[0076] The final classification module introduces ground survey data as auxiliary verification information based on the preliminary classification result set to obtain the final classification result.

[0077] By obtaining a set of results from the initial classification, a preliminary screening is performed for areas with insufficient classification accuracy, identifying the complex environmental areas requiring adjustment. Based on these selected complex environmental areas, corresponding survey data is obtained from ground surveys. Using data fusion, the survey data is matched with the initial classification results to generate a preliminary fused dataset. If the classification accuracy of the preliminary fused dataset deviates from a preset threshold, feature adjustment methods are used to correct the classification features within the complex environmental area, resulting in a corrected feature dataset. Based on this corrected feature dataset, the classification results for the complex environmental area are recalculated using a support vector machine algorithm to generate an updated set of classification results. For areas within the updated set of classification results where accuracy is still insufficient, if the survey data coverage reaches a preset percentage, data fusion is performed again to generate a secondary fused dataset. Based on this secondary fused dataset, conventional comparison tools are used to verify the consistency of the classification results with the survey data, determining whether the final adjusted results meet the expected standards, and finally generating the final set of classification results. The final set of classification results is then used to verify the classification accuracy of the complex environmental area. If local deviations exist, additional survey data is supplemented through data fusion to determine the final stable classification output.

[0078] The embodiments described above are merely descriptions of preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by persons skilled in the art should fall within the scope of protection defined by the claims of the present invention.

Claims

1. A method for intelligent classification of urban land based on remote sensing technology, characterized in that the steps include: Obtain remote sensing image data covering the entire city; Extracting a ground feature vector set based on the remote sensing image data; Based on the ground feature vector set, a preliminary classification result set is obtained; Based on the preliminary classification result set, ground survey data is introduced as auxiliary verification information to obtain the final classification result.

2. The urban land intelligent classification method based on remote sensing technology according to claim 1 is characterized in that: Multi-source remote sensing technology is used to obtain the remote sensing image data covering the entire city; after the remote sensing image data is obtained, filtering technology is used to remove noise, and geometric correction technology is used to correct image distortion caused by sensor angle.

3. The urban land intelligent classification method based on remote sensing technology according to claim 1 is characterized in that: Based on the remote sensing image data, the image data is segmented using preset rules to determine the regional distribution of ground objects; by analyzing the regional distribution of ground objects, geometric characteristic data is extracted, and the segmented image is deeply processed using a convolutional neural network to obtain feature decomposition results.

4. The urban land intelligent classification method based on remote sensing technology according to claim 3 is characterized in that: The convolutional neural network includes an input layer, two convolutional layers, two maximum pooling layers, a fully connected layer, and an output layer; wherein the two convolutional layers respectively include 16 and 32 3×3 convolution kernels with a stride of 1; the stride of the maximum pooling layer is 2, and the fully connected layer includes 256 neurons; the activation function adopts the ReLU function; The convolution layer captures the structural features of the ground object through the local receptive field, and the deep convolution kernel learns the high-level semantic features: Where X represents the input image; C represents the number of channels; W represents the convolution kernel weight; b k represents the bias term; Y i，j，k Represents the value of the output feature map at position (i, j) and the kth channel; ReLU represents the activation function; c represents the channel index; p represents the height direction index; q represents the width.

5. The urban land intelligent classification method based on remote sensing technology according to claim 1 is characterized in that: Based on the feature vector set of the land features, the random forest classification algorithm is used to perform preliminary classification of the land features, and feature matching is performed on the boundary division of high-density areas and open spaces to obtain the preliminary classification result set; when there are overlapping classification areas, the final attribution is determined through texture information comparison and analysis.

6. The urban land intelligent classification method based on remote sensing technology according to claim 1 is characterized in that: The method for obtaining the final classification result includes: Performing preliminary screening based on the preliminary classification result set; According to the screened areas, corresponding survey data are obtained from the ground survey; Matching the survey data with the initial classification results in a data combination manner to obtain a preliminary fused data set; Correction and comparison are performed on the preliminary fusion data set to obtain the final classification result.

7. An intelligent urban land classification system based on remote sensing technology, the system being used to implement the method according to any one of claims 1 to 6, characterized in that: include: Acquisition module, extraction module, construction module and classification module; The acquisition module is used to obtain remote sensing image data covering the entire city; The extraction module is used to extract a ground feature vector set based on the remote sensing image data; The construction module is used to obtain a preliminary classification result set based on the ground feature vector set; The classification module is used to introduce ground survey data as auxiliary verification information based on the preliminary classification result set to obtain a final classification result.

8. The urban land intelligent classification system based on remote sensing technology according to claim 6 is characterized in that: The acquisition module uses multi-source remote sensing technology to obtain the remote sensing image data covering the entire city; after the remote sensing image data is obtained, filtering technology is used to remove noise, and geometric correction technology is used to correct image distortion caused by the sensor angle.

Images