Vegetation extraction method and system based on multi-source remote sensing image, and storage medium
By loading different satellite sensor data, creating segmentation scales and using convolutional neural networks for multi-level feature analysis, the problem of insufficient vegetation information extraction accuracy in the existing technology is solved, and the vegetation extraction effect with high accuracy and high adaptability is achieved.
Patent Information
- Application Number
- CN202411152013.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-21
- Publication Date
- 2024-11-15
AI Technical Summary
现有技术在植被信息提取中识别精度不够,数据损耗过大,难以将多种方法进行有效融合以获得高精度识别。
Using a method based on multi-source remote sensing image, the data of different satellite sensors are loaded for preprocessing and fusion, segmentation scale is created, and multi-level feature analysis and image mapping relationship optimization is used to achieve accurate extraction of vegetation types.
It improves the accuracy and adaptability of vegetation extraction, reduces misjudgment and misjudgment, improves the level of automation and intelligence, and can more accurately identify different vegetation types and adapt to complex environments.
Smart Images

Figure CN118968309A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of remote sensing image processing, and more particularly to a vegetation extraction method, system and storage medium based on multi-source remote sensing images. Background Art
[0002] Vegetation survey is an important application field of remote sensing. Vegetation is an important component of the environment and one of the best indicators of the regional ecological environment. It is also an interpretive indicator of soil, hydrology and other elements. Therefore, it is particularly important to interpret and extract vegetation information on remote sensing images. The extraction of vegetation information on remote sensing images includes the extraction of urban vegetation (green space) information and the extraction of non-urban (agricultural land, mountainous areas and arid and semi-arid areas) vegetation information. Urban vegetation plays a positive role in improving the urban ecological environment and the harmonious coexistence of man and nature. Extracting urban vegetation can dynamically monitor the growth and decline of urban green space, which is conducive to scientific and effective management of cities and provides a scientific basis and evaluation criteria for the planning of urban green space systems. For non-urban areas, especially arid and semi-arid areas, land desertification is relatively serious. By dynamically monitoring the coverage and changes of vegetation, a certain basis can be provided for the grasp of regional environment, hydrology, ecology and global changes. In recent years, researchers have proposed various theories and methods for extracting vegetation information from remote sensing images. The methods used can be roughly divided into two categories: the first category is to start from the spectral characteristics of the ground objects, and obtain the required information by comparing the differences in the spectral characteristics of the ground objects. This method is currently more mature; the second category is to obtain the required information in combination with external knowledge, such as using expert knowledge, neural networks, wavelet transforms, and GIS. These new methods have also gradually played a role in vegetation information extraction in recent years. However, these new methods still have various problems, such as insufficient recognition accuracy and excessive data loss. Therefore, how to integrate multiple methods and obtain higher recognition accuracy with as little data as possible is urgently needed for technicians in this field. Summary of the invention
[0003] In view of this, the present invention provides a vegetation extraction method, system and storage medium based on multi-source remote sensing images, which improves accuracy, enhances adaptability, realizes multi-level feature analysis, and utilizes deep learning optimization results to improve the level of automation and intelligence.
[0004] In order to achieve the above object, the present invention adopts the following technical solution:
[0005] A vegetation extraction method based on multi-source remote sensing images comprises the following steps:
[0006] Load multiple remote sensing image data of earth observation from different satellite sensors, pre-process and fuse multiple remote sensing image data to obtain multi-source remote sensing images;
[0007] Create a segmentation scale, preview the segmentation effect of the current segmentation scale on multi-source remote sensing images, and determine the probability that the edge pixels of the connected domain in the image to be segmented belong to the current vegetation type based on the edge features in the image to be segmented. If the probability is less than the preset probability, recreate the segmentation scale until the probability exceeds the preset probability.
[0008] The multi-source remote sensing image is segmented into N multi-source remote sensing images according to the segmentation scale, and a multi-level feature optimization combination classification rule set is established for the n-th remote sensing image, including the spectral layer, the texture layer and the spatial layer;
[0009] Extract features from different levels of the nth remote sensing image, calculate the feature extraction results using the category attribution algorithm, and obtain the matching degree between the nth remote sensing image and the current vegetation type;
[0010] The convolutional neural network is used to learn the image mapping relationship between the feature extraction results and the vegetation type, and the image mapping relationship is optimized. The image mapping relationship is used to verify the matching degree. If the image mapping relationship and the matching degree are qualified, the vegetation type is obtained and extracted.
[0011] Optionally, multiple remote sensing image data are preprocessed and fused, specifically: the geometric and radiation errors of the multiple remote sensing images are removed separately, the formats, resolutions and spatial reference systems of the multiple remote sensing images are unified, and the unified remote sensing images are fused using the temporal and spatial correlation of different images and the complementarity of information.
[0012] Optionally, based on the edge features in the image to be segmented, the probability that the edge pixel points of the connected domain in the image to be segmented belong to the current vegetation type is determined, and the specific steps are as follows:
[0013] Perform corner point detection on the edge of the image to be detected to obtain each corner point in the image to be detected, mark it as an edge inflection point, connect two edge inflection points as the mth line, and calculate the angle between the mth line and the m-1th line;
[0014] The probability that the ith edge pixel point belongs to the vegetation type is determined based on the angle between the mth link and the m-1th link of the connected domain in the image to be detected and the gradient amplitude of the ith edge pixel point.
[0015] Optionally, the multi-source remote sensing image is segmented into N multi-source remote sensing images according to the segmentation scale: a multi-scale feature extraction algorithm is used, an edge detection operator, a texture recognition kernel, and a color gradient operator are combined to extract a feature map of the image to be segmented, the feature map is segmented using the segmentation scale, and the segmentation result is mapped to the original image to be segmented to obtain a segmentation result.
[0016] Optionally, extracting the feature map of the image to be segmented is specifically: extracting key features of the difference scale image and a comprehensive feature map from a set of difference resolution images at each scale, wherein the key features of the difference scale image include: image contour edges, image surface characteristics and color gradient change data.
[0017] Optionally, according to different segmentation scales, linear iterative clustering is used to perform superpixel segmentation on multi-source remote sensing images to obtain irregular superpixel sets; feature extraction is performed on the superpixel set through a superpixel-level feature extraction module to obtain global superpixel-level features; edge pixel points in the global superpixel-level features are obtained to determine the probability that the edge pixel points in the global superpixel-level features belong to the current vegetation type.
[0018] A vegetation extraction system based on multi-source remote sensing images, comprising:
[0019] Remote sensing image fusion module: used to load a variety of remote sensing image data of earth observation from different satellite sensors, pre-process and fuse a variety of remote sensing image data to obtain multi-source remote sensing images;
[0020] Segmentation scale creation module: used to create segmentation scales, preview the segmentation effect of the current segmentation scale on multi-source remote sensing images, determine the probability that the edge pixels of the connected domain in the image to be segmented belong to the current vegetation type based on the edge features in the image to be segmented, and recreate the segmentation scale if the probability is less than the preset probability until the probability exceeds the preset probability;
[0021] Multi-level feature combination module: used to segment the multi-source remote sensing image into N multi-source remote sensing images according to the segmentation scale, and establish a multi-level feature combination classification rule set for the nth remote sensing image, including spectral layer, texture layer and spatial layer;
[0022] Remote sensing image matching module: used to extract features from different levels of the nth remote sensing image, calculate the feature extraction results using the category attribution algorithm, and obtain the matching degree between the nth remote sensing image and the current vegetation type;
[0023] Vegetation type determination and extraction module: used to use convolutional neural network to learn the image mapping relationship between feature extraction results and vegetation types, optimize the image mapping relationship, and use the image mapping relationship to verify the matching degree. If the image mapping relationship and the matching degree are qualified, the vegetation type is obtained and extracted.
[0024] A computer storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the computer program implements any one of the steps of a vegetation extraction method based on multi-source remote sensing images.
[0025] It can be seen from the above technical solutions that, compared with the prior art, the present invention provides a vegetation extraction method, system and storage medium based on multi-source remote sensing images, which have the following beneficial effects:
[0026] 1. Improve the accuracy of vegetation extraction: By fusing multiple remote sensing image data from different satellite sensors, this method can fully utilize the advantages of various sensors in different spectral ranges and spatial resolutions, thereby more comprehensively capturing the characteristic information of vegetation. This helps to reduce the information loss and errors caused by a single data source and improve the accuracy of vegetation extraction.
[0027] 2. Enhanced adaptability and flexibility: By dynamically adjusting the segmentation scale and determining the optimal segmentation scale based on the probability evaluation of edge features, this method can automatically adapt to the characteristics of different vegetation types and different regional environments, enhancing the adaptability and flexibility of processing. At the same time, through multiple iterations and adjustments, it can ensure that the segmentation effect meets the preset standards and improve the accuracy of processing.
[0028] 3. Comprehensive analysis of multi-level features: By constructing a set of classification rules for the optimal combination of multiple-level features including the spectral layer, texture layer, and spatial layer, this method can comprehensively consider the characteristic information of vegetation at different levels and achieve a more comprehensive and in-depth analysis. This helps to more accurately identify different vegetation types and reduce misjudgments and missed judgments.
[0029] 4. Optimize results using deep learning: By using convolutional neural networks to learn the image mapping relationship between feature extraction results and vegetation types and optimizing the mapping relationship, this method can automatically learn and extract higher-level feature representations from a large amount of data, further improving the accuracy and efficiency of vegetation extraction. At the same time, the powerful learning ability of convolutional neural networks also enables it to cope with more complex and changing environmental conditions.
[0030] 5. Improve the level of automation and intelligence: The entire method process is highly automated and intelligent, reducing the impact of manual intervention and subjective judgment, and improving processing efficiency and objectivity of results. This is of great significance for large-scale and long-term vegetation monitoring and extraction work. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.
[0032] Figure 1 It is a schematic diagram of the process of the present invention;
[0033] Figure 2 It is a schematic diagram of the structure of the present invention. DETAILED DESCRIPTION
[0034] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. 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 creative work are within the scope of protection of the present invention.
[0035] Example 1
[0036] The embodiment of the present invention discloses a vegetation extraction method based on multi-source remote sensing images, such as Figure 1 As shown, the following steps are included:
[0037] Step 1: Load multiple remote sensing image data of earth observation from different satellite sensors, pre-process and fuse the multiple remote sensing image data to obtain multi-source remote sensing images;
[0038] Step 2: Create a segmentation scale, preview the segmentation effect of the current segmentation scale on the multi-source remote sensing image, and determine the probability that the edge pixel points of the connected domain in the image to be segmented belong to the current vegetation type based on the edge features in the image to be segmented. If the probability is less than the preset probability, recreate the segmentation scale until the probability exceeds the preset probability.
[0039] Step 3: Segment the multi-source remote sensing image into N multi-source remote sensing images according to the segmentation scale, and establish a set of classification rules for multiple hierarchical feature optimization combinations for the nth remote sensing image, including a spectral layer, a texture layer, and a spatial layer;
[0040] Step 4: Extract features from different levels of the nth remote sensing image, calculate the feature extraction results using a category attribution algorithm, and obtain the matching degree between the nth remote sensing image and the current vegetation type;
[0041] Step 5: Use the convolutional neural network to learn the image mapping relationship between the feature extraction results and the vegetation type, optimize the image mapping relationship, and use the image mapping relationship to verify the matching degree. If the image mapping relationship and the matching degree are qualified, the vegetation type is obtained and extracted.
[0042] Furthermore, in step one, the multiple remote sensing image data are preprocessed and fused, specifically: the geometric and radiation errors of the multiple remote sensing images are removed respectively, the formats, resolutions and spatial reference systems of the multiple remote sensing images are unified, and the unified remote sensing images are fused by utilizing the temporal and spatial correlation and information complementarity of different images.
[0043] Furthermore, in step 2, the probability that the edge pixel points of the connected domain in the image to be segmented belong to the current vegetation type is determined based on the edge features in the image to be segmented. The specific steps are as follows:
[0044] Step 2.1, perform corner point detection on the edge of the image to be detected to obtain each corner point in the image to be detected, mark it as an edge inflection point, connect two edge inflection points as the mth line, and calculate the angle between the mth line and the m-1th line;
[0045] Step 2.2: Determine the probability that the ith edge pixel belongs to the vegetation type based on the angle between the mth line and the m-1th line of the connected domain in the image to be detected and the gradient amplitude of the ith edge pixel.
[0046] Furthermore, in step three, the multi-source remote sensing image is segmented into N multi-source remote sensing images according to the segmentation scale. Specifically, a multi-scale feature extraction algorithm is used, and an edge detection operator, a texture recognition kernel, and a color gradient operator are combined to extract a feature map of the image to be segmented. The feature map is segmented using the segmentation scale, and the segmentation result is mapped to the original image to be segmented to obtain a segmentation result.
[0047] Furthermore, the feature map of the image to be segmented is extracted specifically by extracting the key features of the difference scale image and the comprehensive feature map from the difference resolution image set at each scale, wherein the key features of the difference scale image include: image contour edges, image surface characteristics and color gradient change data.
[0048] Furthermore, in step five, the following parts are specifically included:
[0049] Feature extraction: First, the convolutional layer and pooling layer of CNN are used to extract multi-level features from the preprocessed and segmented multi-source remote sensing images. These features may include spectral features (such as spectral reflectance in different bands), texture features (such as roughness and directionality of vegetation surface), and spatial features (such as shape, size, and distribution of vegetation).
[0050] Image mapping relationship learning: The extracted features are then fed into the fully connected layers (or densely connected layers) of the CNN for processing, which are responsible for learning the nonlinear mapping relationship between features and vegetation types. At this stage, the CNN continuously adjusts network parameters through optimization methods such as back propagation algorithms and gradient descent to minimize the error between the predicted vegetation type and the actual vegetation type.
[0051] Image mapping relationship optimization: In order to improve the accuracy and generalization ability of the mapping relationship, some optimization strategies are usually adopted, such as increasing the amount of training data, using regularization technology to prevent overfitting, adjusting the learning rate, etc. In addition, CNN models that have been trained on other similar tasks can be used to accelerate and improve the learning process of the current task through methods such as ensemble learning and transfer learning.
[0052] Verification of matching degree: After CNN training is completed, the reliability of the learned image mapping relationship and matching degree needs to be verified using a test dataset. This usually involves inputting the remote sensing images in the test dataset into the trained CNN model, obtaining the predicted vegetation type, and comparing it with the actual vegetation type. By calculating indicators such as accuracy, recall, and F1 score, the performance of the model on unseen data can be evaluated.
[0053] Vegetation type extraction: If the image mapping relationship and matching degree meet the preset qualification standards, it can be considered that the CNN model has successfully learned the effective mapping relationship between remote sensing images and vegetation types. At this time, the model can be used to extract vegetation types from new remote sensing images to obtain accurate vegetation distribution maps or other forms of vegetation information products.
[0054] and Figure 1 Corresponding to the method shown, the present invention also discloses a vegetation extraction system based on multi-source remote sensing images for Figure 1 The implementation of the method, the specific structure is as follows Figure 2 As shown, including:
[0055] Remote sensing image fusion module: used to load a variety of remote sensing image data of earth observation from different satellite sensors, pre-process and fuse a variety of remote sensing image data to obtain multi-source remote sensing images;
[0056] Segmentation scale creation module: used to create segmentation scales, preview the segmentation effect of the current segmentation scale on multi-source remote sensing images, determine the probability that the edge pixels of the connected domain in the image to be segmented belong to the current vegetation type based on the edge features in the image to be segmented, and recreate the segmentation scale if the probability is less than the preset probability until the probability exceeds the preset probability;
[0057] Multi-level feature combination module: used to segment the multi-source remote sensing image into N multi-source remote sensing images according to the segmentation scale, and establish a multi-level feature combination classification rule set for the nth remote sensing image, including spectral layer, texture layer and spatial layer;
[0058] Remote sensing image matching module: used to extract features from different levels of the nth remote sensing image, calculate the feature extraction results using the category attribution algorithm, and obtain the matching degree between the nth remote sensing image and the current vegetation type;
[0059] Vegetation type determination and extraction module: used to use convolutional neural network to learn the image mapping relationship between feature extraction results and vegetation types, optimize the image mapping relationship, and use the image mapping relationship to verify the matching degree. If the image mapping relationship and the matching degree are qualified, the vegetation type is obtained and extracted.
[0060] In this embodiment, a computer storage medium is also disclosed. A computer program is stored on the computer storage medium. When the computer program is executed by a processor, the steps of any one of the vegetation extraction methods based on multi-source remote sensing images are implemented.
[0061] Example 2
[0062] The difference between this embodiment and embodiment 1 is only that:
[0063] According to different segmentation scales, linear iterative clustering is used to perform superpixel segmentation on multi-source remote sensing images to obtain irregular superpixel sets; the superpixel level feature extraction module is used to extract features from the superpixel set to obtain global superpixel level features; the edge pixels in the global superpixel level features are obtained to determine the probability that the edge pixels in the global superpixel level features belong to the current vegetation type.
[0064] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.
[0065] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A vegetation extraction method based on multi-source remote sensing images, characterized in that: The following steps are involved: Load multiple remote sensing image data of earth observation from different satellite sensors, pre-process and fuse the multiple remote sensing image data to obtain multi-source remote sensing images; Create a segmentation scale, preview the segmentation effect of the current segmentation scale on multi-source remote sensing images, and determine the probability that the edge pixels of the connected domain in the image to be segmented belong to the current vegetation type based on the edge features in the image to be segmented. If the probability is less than the preset probability, recreate the segmentation scale until the probability exceeds the preset probability. The multi-source remote sensing image is segmented into N multi-source remote sensing images according to the segmentation scale, and a multi-level feature optimization combination classification rule set is established for the n-th remote sensing image, including the spectral layer, the texture layer and the spatial layer; Extract features from different levels of the nth remote sensing image, calculate the feature extraction results using the category attribution algorithm, and obtain the matching degree between the nth remote sensing image and the current vegetation type; The convolutional neural network is used to learn the image mapping relationship between the feature extraction results and the vegetation type, and the image mapping relationship is optimized. The image mapping relationship is used to verify the matching degree. If the image mapping relationship and the matching degree are qualified, the vegetation type is obtained and extracted.
2. The vegetation extraction method based on multi-source remote sensing images according to claim 1 is characterized in that: Preprocess and fuse multiple remote sensing image data, specifically: remove the geometric and radiation errors of multiple remote sensing images separately, unify the format, resolution and spatial reference system of multiple remote sensing images, and fuse the unified remote sensing images by utilizing the temporal and spatial correlation and information complementarity of different images.
3. The vegetation extraction method based on multi-source remote sensing images according to claim 1 is characterized in that: Based on the edge features in the image to be segmented, determine the probability that the edge pixels of the connected domain in the image to be segmented belong to the current vegetation type. The specific steps are as follows: Perform corner point detection on the edge of the image to be detected to obtain each corner point in the image to be detected, mark it as an edge inflection point, connect two edge inflection points as the mth line, and calculate the angle between the mth line and the m-1th line; The probability that the ith edge pixel point belongs to the vegetation type is determined based on the angle between the mth link and the m-1th link of the connected domain in the image to be detected and the gradient amplitude of the ith edge pixel point.
4. The vegetation extraction method based on multi-source remote sensing images according to claim 1 is characterized in that: The multi-source remote sensing image is segmented into N multi-source remote sensing images according to the segmentation scale. Specifically, a multi-scale feature extraction algorithm is used, and an edge detection operator, a texture recognition kernel, and a color gradient operator are combined to extract a feature map of the image to be segmented. The feature map is segmented using the segmentation scale, and the segmentation result is mapped to the original image to be segmented to obtain the segmentation result.
5. The vegetation extraction method based on multi-source remote sensing images according to claim 4 is characterized in that: The feature map of the image to be segmented is specifically extracted by extracting the key features of the difference scale image and the comprehensive feature map from the difference resolution image set at each scale, wherein the key features of the difference scale image include: image contour edges, image surface characteristics and color gradient change data.
6. The vegetation extraction method based on multi-source remote sensing images according to claim 1 is characterized in that: According to different segmentation scales, linear iterative clustering is used to perform superpixel segmentation on multi-source remote sensing images to obtain irregular superpixel sets; the superpixel level feature extraction module is used to extract features from the superpixel set to obtain global superpixel level features; Obtain edge pixel points in the global super-pixel level feature, and determine the probability that the edge pixel points in the global super-pixel level feature belong to the current vegetation type.
7. A vegetation extraction system based on multi-source remote sensing images, characterized in that: include: Remote sensing image fusion module: used to load a variety of remote sensing image data of earth observation from different satellite sensors, pre-process and fuse a variety of remote sensing image data to obtain multi-source remote sensing images; Segmentation scale creation module: used to create segmentation scales, preview the segmentation effect of the current segmentation scale on multi-source remote sensing images, determine the probability that the edge pixels of the connected domain in the image to be segmented belong to the current vegetation type based on the edge features in the image to be segmented, and recreate the segmentation scale if the probability is less than the preset probability until the probability exceeds the preset probability; Multi-level feature combination module: used to segment the multi-source remote sensing image into N multi-source remote sensing images according to the segmentation scale, and establish a multi-level feature combination classification rule set for the nth remote sensing image, including spectral layer, texture layer and spatial layer; Remote sensing image matching module: used to extract features from different levels of the nth remote sensing image, calculate the feature extraction results using the category attribution algorithm, and obtain the matching degree between the nth remote sensing image and the current vegetation type; Vegetation type determination and extraction module: used to use convolutional neural network to learn the image mapping relationship between feature extraction results and vegetation types, optimize the image mapping relationship, and use the image mapping relationship to verify the matching degree. If the image mapping relationship and the matching degree are qualified, the vegetation type is obtained and extracted.
8. A computer storage medium, characterized in that: The computer storage medium stores a computer program, and when the computer program is executed by the processor, the steps of the vegetation extraction method based on multi-source remote sensing images as described in any one of claims 1 to 6 are implemented.
