A remote sensing image classification and extraction method and system based on remote sensing index

Abstract

The present invention discloses a remote sensing image classification and extraction method and system based on remote sensing index, which belongs to the field of remote sensing image extraction technology. After preprocessing the remote sensing image, multiple remote sensing indexes are calculated, and the image is finely classified based on these indexes, which significantly improves the diversity and accuracy of ground object identification. This method not only independently uses each remote sensing index to obtain preliminary classification results, but also accurately identifies classification difference areas by comparing these results, and performs secondary classification on such areas, thereby further optimizing the classification effect. In addition, this method, by virtue of its characteristics of combined application of multiple remote sensing indices, demonstrates strong adaptability and flexibility, and can effectively improve the accuracy and robustness of ground object extraction in different geographical environments and application scenarios, providing solid technical support for the widespread application of remote sensing technology.

Description

Technical Field

[0001] The present invention relates to the technical field of remote sensing image extraction, and more particularly to a remote sensing image classification extraction method and system based on remote sensing index. Background Art

[0002] Remote sensing images contain rich detail information, such as edge contours, gradient changes, and tiny targets. However, due to the limitation of image resolution, these subtle features are often difficult to be fully captured and accurately classified in the process of object classification. This limitation not only leads to a significant loss of detail information in the classification results, but also makes the edges of objects blurry and distorted, ultimately affecting the overall accuracy of object classification. An ideal classification system should be able to extract higher-resolution information from remote sensing images to meet the needs of detail classification. At the same time, for the rapid identification of conventional objects, the image must have a larger coverage width, and the resolution requirement is relatively loose. However, due to the use of diverse sensor technologies, the fusion process of remote sensing images of different resolutions faces many challenges.

[0003] In recent years, deep learning technology has made significant progress in the field of remote sensing image classification, especially the rise of two major methods: picture-level classification and pixel-level classification. The picture-level classification method focuses on feature learning for the entire image and identifies the dominant object category in the image through a convolutional neural network. However, this method cannot provide fine classification results at the pixel level and has insufficient ability to capture detail information. In contrast, the pixel-level classification method adopts a fully convolutional network architecture. By removing the fully connected layer and introducing a 1x1 convolutional layer, it achieves direct classification from pixel to pixel, effectively retains the spatial information of the image, breaks through the limitation of the input image size, and significantly improves the classification efficiency and accuracy. However, although pixel-level classification has shown technical advantages, in order to capture multi-scale regional features when dealing with specific tasks such as remote sensing image farmland extraction, deep convolutional networks often need to downsample high-resolution images. In this process, some image details, such as edge sharpness and gradient changes, will inevitably be lost, which will affect the edge clarity and detail richness of the classification results, and ultimately limit the further improvement of classification accuracy.

[0004] Therefore, how to provide an efficient and high-precision remote sensing image classification and extraction method and system is a problem that technical personnel in this field need to solve urgently. Summary of the invention

[0005] In view of this, the present invention provides a remote sensing image classification and extraction method and system based on remote sensing index, which can realize efficient and high-precision classification and extraction of remote sensing images.

[0006] In order to achieve the above object, the present invention adopts the following technical solution:

[0007] In one aspect, the present invention provides a remote sensing image classification and extraction method based on remote sensing index, comprising:

[0008] Acquire a remote sensing image of the area to be extracted, and preprocess the remote sensing image;

[0009] Calculating multiple remote sensing indexes of the remote sensing image based on the preprocessed remote sensing image;

[0010] Classifying the remote sensing images based on the remote sensing indexes to obtain classification results;

[0011] Comparing the classification results to obtain non-overlapping areas in the classification results;

[0012] The non-overlapping areas are further classified to obtain a final classification result of the remote sensing image.

[0013] Preferably, preprocessing the remote sensing image includes:

[0014] Perform orthorectification operations on remote sensing images;

[0015] The preset calibration parameters and spectral response function are used to perform radiometric calibration on the orthorectified remote sensing image to obtain the surface reflectance image.

[0016] The reference image is used to geometrically correct the surface reflectance image;

[0017] The principal component transformation fusion method is used to fuse the geometrically corrected reference images.

[0018] Preferably, the remote sensing images are classified based on the remote sensing indexes to obtain classification results, including:

[0019] The remote sensing indexes are combined by the XGBoost model to obtain multiple groups of classified remote sensing index groups, wherein all classified targets can be obtained based on the remote sensing indexes in the classified remote sensing index groups;

[0020] The remote sensing indices in the classified remote sensing index group are classified by a target classifier to obtain a classification result.

[0021] Preferably, the target classifier is constructed as follows:

[0022] Constructing a training sample data set and a sample test data set based on the remote sensing index of the remote sensing image, and adding a sample category label to the test remote sensing index in the sample test data set;

[0023] Determining the local accuracy of multiple different basic classifiers according to the training sample data set, the sample category label and the value of parameter k in the preset K-nearest neighbor algorithm;

[0024] Determining target category labels corresponding to training remote sensing indices in the training sample data set according to the local accuracy of each of the basic classifiers;

[0025] Acquire a category label corresponding to a remote sensing index; and determine a target classifier from the plurality of basic classifiers according to the category label corresponding to the remote sensing index and the target category label.

[0026] Preferably, the remote sensing index includes: normalized difference vegetation index, soil adjusted vegetation index normalized vegetation index, enhanced vegetation index, soil index, improved normalized water index, normalized difference water index, soil brightness index, and building index.

[0027] Preferably, comparing the classification results to obtain non-overlapping areas in the classification results includes:

[0028] Acquire the area and perimeter of the connected region of the classification result, and determine the image overlap of the classification result based on the area and perimeter;

[0029] Determining whether there is an overlapping area in the classification results according to the area and area threshold of the connection area and the image overlap and overlap threshold;

[0030] If there is no overlapping area, the area to be extracted is a non-overlapping area;

[0031] If there is an overlapping area, obtaining the intersection point of the overlapping area;

[0032] Extracting the contour intersection points of the overlapping area based on the intersection points;

[0033] Extracting the contour line of the overlapping area based on the contour intersection point to obtain the overlapping area;

[0034] The non-overlap is obtained based on the overlapping area and the remote sensing image.

[0035] On the other hand, the present invention provides a system for implementing the above-mentioned remote sensing image classification and extraction method based on remote sensing index, the system comprising:

[0036] A data acquisition module is used to obtain remote sensing images of the area to be extracted and pre-process the remote sensing images;

[0037] A remote sensing index calculation module, used for calculating a plurality of remote sensing indexes of the remote sensing image based on the preprocessed remote sensing image;

[0038] A preliminary classification module, used for classifying the remote sensing images based on the remote sensing indexes to obtain classification results;

[0039] A result analysis unit, used for comparing the classification results to obtain non-overlapping areas in the classification results;

[0040] The classification result determination unit is used to further classify the non-overlapping areas to obtain the final classification result of the remote sensing image.

[0041] Preferably, the data acquisition module includes:

[0042] An acquisition unit, used for acquiring a remote sensing image of a region to be extracted;

[0043] An orthorectification unit is used to perform orthorectification operations on remote sensing images;

[0044] A radiation unit, used to perform radiation calibration on the orthorectified remote sensing image using preset calibration parameters and spectral response functions to obtain a surface reflectance image;

[0045] A geometric correction unit, used for geometrically correcting the surface reflectance image using a reference image;

[0046] The fusion unit is used to fuse the reference image after geometric correction by using the principal component transformation fusion method.

[0047] Through the above technical solutions, it can be known that compared with the prior art, the present invention discloses a remote sensing image classification and extraction method and system based on remote sensing index, which classifies remote sensing images based on multiple remote sensing indexes, obtains various types of land objects in remote sensing images, and improves the accuracy of land object classification; further, the present invention compares and analyzes the types of land objects obtained by multi-remote sensing index classification, obtains different areas in each classification result, and performs secondary recognition on the different areas, which improves the accuracy of the recognition result. Furthermore, the present invention combines multiple remote sensing indexes to perform remote sensing image recognition, which has good universality in practical applications, can effectively improve the accuracy of land object extraction, and improve the robustness and universality of remote sensing image extraction. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] 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.

[0049] Figure 1 A flow chart of the method provided by the present invention;

[0050] Figure 2 This is a system framework diagram provided by the present invention. DETAILED DESCRIPTION

[0051] 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.

[0052] The embodiment of the present invention discloses a remote sensing image classification extraction method based on remote sensing index, referring to Figure 1 ,include:

[0053] Step 1: Obtain a remote sensing image of the area to be extracted and preprocess the remote sensing image.

[0054] Firstly, the experimental research area was determined, and a Shenzhen DJI Phantom 4 Multispectral Edition quad-rotor drone was used to capture images of the experimental research area, and the captured images were stored in the camera's internal SD card.

[0055] Step 2: Calculate multiple remote sensing indexes of the remote sensing image based on the preprocessed remote sensing image.

[0056] Step 3: Classify the remote sensing images based on the remote sensing indexes to obtain classification results.

[0057] Step 4: Compare the classification results to obtain non-overlapping areas in the classification results.

[0058] Step 5: Further classify the non-overlapping areas to obtain the final classification result of the remote sensing image.

[0059] Furthermore, after the drone acquires images of the test site, the images are stitched, geometrically corrected, and radiometrically corrected using Pix4DMapper software to obtain an orthophoto of the test site. The images are cropped using ENVI5.3 software to obtain the target area. Specifically, the preprocessing of the remote sensing images includes:

[0060] Perform orthorectification operations on remote sensing images;

[0061] The preset calibration parameters and spectral response function are used to perform radiometric calibration on the orthorectified remote sensing image to obtain the surface reflectance image.

[0062] The reference image is used to geometrically correct the surface reflectance image;

[0063] The principal component transformation fusion method is used to fuse the geometrically corrected reference images.

[0064] Specifically, step 3: classifying the remote sensing images based on the remote sensing indexes to obtain classification results, including:

[0065] Step 3.1: Combine the remote sensing indices through the XGBoost model to obtain multiple groups of classified remote sensing index groups, wherein all classified targets can be obtained based on the remote sensing indices in the classified remote sensing index groups; specifically, establish the XGBoost model using the python package, determine the remote sensing indices in the classified remote sensing index group according to the type of remote sensing index and the classification target of the ground object, and ensure that the remote sensing index group can identify all types of ground objects in the remote sensing image.

[0066] Step 3.2: Classify the remote sensing indices in the classified remote sensing index group through a target classifier to obtain a classification result.

[0067] In another embodiment, the target classifier is constructed as follows:

[0068] Constructing a training sample data set and a sample test data set based on the remote sensing index of the remote sensing image, and adding a sample category label to the test remote sensing index in the sample test data set;

[0069] Determining the local accuracy of multiple different basic classifiers according to the training sample data set, the sample category label and the value of parameter k in the preset K-nearest neighbor algorithm;

[0070] Determining the target category label corresponding to the training remote sensing index in the training sample data set according to the local accuracy of each of the basic classifiers;

[0071] Acquire a category label corresponding to a remote sensing index; and determine a target classifier from the plurality of basic classifiers according to the category label corresponding to the remote sensing index and the target category label.

[0072] Furthermore, the remote sensing index includes: normalized difference vegetation index, soil adjusted vegetation index normalized vegetation index, enhanced vegetation index, soil index, improved normalized water index, normalized difference water index, soil brightness index, and building index.

[0073] Further, step 4: comparing the classification results to obtain non-overlapping areas in the classification results, including:

[0074] Step 4.1: Obtain the area and perimeter of the connected area of ​​the classification result, and determine the image overlap of the classification result based on the area and perimeter; specifically, in this embodiment, the image overlap calculation formula is as follows:

[0075]

[0076] Where SF represents the image overlap, A represents the area of ​​the connected region, and C represents the perimeter of the connected region.

[0077] Step 4.2: Determine whether there is an overlapping area in the classification result according to the area and area threshold of the connection area and the image overlap and overlap threshold.

[0078] Since the overlapping areas have more complex boundaries, the boundary complexity formula can be used to determine whether there is overlap between the areas in the classification results. Among them, A and C are obtained based on the number of pixels. When there are multiple or partial overlapping areas, the boundaries become more complex due to the presence of depressions, and in the case of the same area, the overlapping target images have a large perimeter and a small shape factor. The area of ​​each connected area in the classification result combined with the overlap can be used as the basis for judging whether there is overlap. Specifically: If SF i >SF0 and A i max When , it means that there is no overlapping area, otherwise there is an overlapping area; where SF0 is the overlap threshold, A max is the area of ​​each connected region in the classification result, SF i is the image overlap of the i-th connected area, A i is the area of ​​the i-th connected region.

[0079] If there is no overlapping area, the area to be extracted is a non-overlapping area;

[0080] If there is an overlapping area, obtaining the intersection point of the overlapping area;

[0081] Extracting the contour intersection points of the overlapping area based on the intersection points;

[0082] Extracting the contour line of the overlapping area based on the contour intersection point to obtain the overlapping area;

[0083] The non-overlap is obtained based on the overlapping area and the remote sensing image.

[0084] On the other hand, the present invention provides a system for implementing the above-mentioned remote sensing image classification and extraction method based on remote sensing index, such as Figure 2 As shown, the system includes:

[0085] A data acquisition module is used to obtain remote sensing images of the area to be extracted and pre-process the remote sensing images;

[0086] A remote sensing index calculation module, used for calculating a plurality of remote sensing indexes of the remote sensing image based on the preprocessed remote sensing image;

[0087] ​A preliminary classification module, used for classifying the remote sensing images based on the remote sensing indexes to obtain classification results;

[0088] A result analysis unit, used for comparing the classification results to obtain non-overlapping areas in the classification results;

[0089] The classification result determination unit is used to further classify the non-overlapping areas to obtain the final classification result of the remote sensing image.

[0090] Preferably, the data acquisition module includes:

[0091] An acquisition unit, used for acquiring a remote sensing image of a region to be extracted;

[0092] An orthorectification unit is used to perform orthorectification operations on remote sensing images;

[0093] A radiation unit, used to perform radiation calibration on the orthorectified remote sensing image using preset calibration parameters and spectral response functions to obtain a surface reflectance image;

[0094] A geometric correction unit, used for geometrically correcting the surface reflectance image using a reference image;

[0095] The fusion unit is used to fuse the reference image after geometric correction by using the principal component transformation fusion method.

[0096] 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.

[0097] 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 remote sensing image classification and extraction method based on remote sensing index, characterized in that: include: Acquire a remote sensing image of the area to be extracted, and preprocess the remote sensing image; Calculating multiple remote sensing indexes of the remote sensing image based on the preprocessed remote sensing image; Classifying the remote sensing images based on the remote sensing indexes to obtain classification results; Comparing the classification results to obtain non-overlapping areas in the classification results; Further classifying the non-overlapping areas to obtain a final classification result of the remote sensing image; Preprocessing the remote sensing image includes: Perform orthorectification operations on remote sensing images; The preset calibration parameters and spectral response function are used to perform radiometric calibration on the orthorectified remote sensing image to obtain the surface reflectance image. The reference image is used to geometrically correct the surface reflectance image; The principal component transformation fusion method is used to fuse the geometrically corrected reference images. The remote sensing images are classified based on the remote sensing indexes to obtain classification results, including: The remote sensing indexes are combined by the XGBoost model to obtain multiple groups of classified remote sensing index groups, wherein all classified targets can be obtained based on the remote sensing indexes in the classified remote sensing index groups; Classifying the remote sensing indices in the classified remote sensing index group by a target classifier to obtain a classification result; Comparing the classification results to obtain non-overlapping areas in the classification results includes: Acquire the area and perimeter of the connected region of the classification result, and determine the image overlap of the classification result based on the area and perimeter; Determining whether there is an overlapping area in the classification results according to the area and area threshold of the connection area and the image overlap and overlap threshold; If there is no overlapping area, the area to be extracted is a non-overlapping area; If there is an overlapping area, obtaining the intersection point of the overlapping area; Extracting the contour intersection points of the overlapping area based on the intersection points; Extracting the contour line of the overlapping area based on the contour intersection point to obtain the overlapping area; A non-overlapping area is obtained based on the overlapping area and the remote sensing image.

2. The remote sensing image classification and extraction method based on remote sensing index according to claim 1 is characterized in that: The construction process of the target classifier is as follows: Constructing a training sample data set and a sample test data set based on the remote sensing index of the remote sensing image, and adding a sample category label to the test remote sensing index in the sample test data set; Determining the local accuracy of multiple different basic classifiers according to the training sample data set, the sample category label and the value of parameter k in the preset K-nearest neighbor algorithm; Determining target category labels corresponding to training remote sensing indices in the training sample data set according to the local accuracy of each of the basic classifiers; Acquire a category label corresponding to a remote sensing index; and determine a target classifier from the plurality of basic classifiers according to the category label corresponding to the remote sensing index and the target category label.

3. The remote sensing image classification and extraction method based on remote sensing index according to claim 1 is characterized in that: The remote sensing index includes: normalized difference vegetation index, soil adjusted vegetation index normalized vegetation index, enhanced vegetation index, soil index, improved normalized water index, normalized difference water index, soil brightness index, and building index.

4. A system for implementing a remote sensing image classification and extraction method based on remote sensing index as described in any one of claims 1 to 3, characterized in that: The system comprises: A data acquisition module is used to obtain remote sensing images of the area to be extracted and pre-process the remote sensing images; A remote sensing index calculation module, used for calculating a plurality of remote sensing indexes of the remote sensing image based on the preprocessed remote sensing image; A preliminary classification module, used for classifying the remote sensing images based on the remote sensing indexes to obtain classification results; A result analysis unit, used for comparing the classification results to obtain non-overlapping areas in the classification results; The classification result determination unit is used to further classify the non-overlapping areas to obtain the final classification result of the remote sensing image.

5. The remote sensing image classification and extraction system based on remote sensing index according to claim 4 is characterized in that: The data acquisition module comprises: An acquisition unit, used for acquiring a remote sensing image of a region to be extracted; An orthorectification unit is used to perform orthorectification operations on remote sensing images; A radiation unit, used to perform radiation calibration on the orthorectified remote sensing image using preset calibration parameters and spectral response functions to obtain a surface reflectance image; A geometric correction unit, used for geometrically correcting the surface reflectance image using a reference image; The fusion unit is used to fuse the reference image after geometric correction by using the principal component transformation fusion method.

Images