A method for extracting ground object information based on single-phase remote sensing image

Abstract

The application discloses a kind of based on single period remote sensing image ground feature information extraction method, the extraction method includes: from single period multispectral remote sensing image data obtains the first characteristic index data of target ground feature;Through power function, the first characteristic index data is calculated, and second characteristic index data is obtained;The first characteristic index data or second characteristic index data is centralized and is handled and is handled, and third characteristic index data is obtained;From the feature sequence data consisting of the first characteristic index data, second characteristic index data and third characteristic index data, target ground feature information extraction is carried out.The application reduces the data amount required for extraction, reduces the difficulty of data acquisition, effectively avoids human factor interference, can realize extraction process automation, with more extensive universality.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of remote sensing image classification, and particularly relates to a method for extracting ground object information from single-period remote sensing images. BACKGROUND

[0002] The technology for identifying, analyzing, extracting and distinguishing various ground object information in remote sensing images is remote sensing image classification technology. In the prior art, the widely used and mature remote sensing classification methods mainly include supervised classification methods based on minimum distance, maximum likelihood method, spectral angle method, neural network, decision tree classification methods based on multi-time sequence remote sensing images, object-oriented classification methods and unsupervised classification methods. Although there are many commonly used remote sensing classification methods, each method has some common technical problems that need to be further solved, such as the relatively high human subjective factors in the supervised classification method, the need for multi-period remote sensing images and the determination of the threshold value of some rules in the decision tree classification method, the selection of the best threshold value in the object-oriented classification method, and the classification form and post-merging class in the unsupervised classification method. In general, the main problems faced by the prior art include difficulty in obtaining effective image data, human interference, low efficiency of extracting ground object information, and poor universality. SUMMARY

[0003] In order to overcome the defects of the prior art, the purpose of the present application is to provide a method for efficiently, accurately and automatically extracting ground object information only by using one period of remote sensing images, which reduces the amount of data and the difficulty of data acquisition, effectively avoids human factor interference, and improves the extraction efficiency, automation degree and universality.

[0004] The technical scheme of the present application is as follows:

[0005] A method for extracting ground object information from single-period remote sensing images, comprising:

[0006] S1, based on the selected single-period remote sensing image data, obtaining remote sensing feature index information of the target ground object, i.e. first feature index data;

[0007] S2, performing single or multiple raster operations on the first feature index data by using a power function to obtain second feature index data;

[0008] S3, performing centering processing on the first feature index data or the second feature index data, and performing inversion processing on the centering processed data to obtain third feature index data;

[0009] S4, composing feature sequence data of the target ground object from the first feature index data, the second feature index data and the third feature index data;

[0010] S5 extracts target ground object information from the feature sequence data.

[0011] In the above process, the processing of S2 can enlarge the difference of feature information between the target ground object and the interference ground object with similar spectrum, and meanwhile, the number of images for information extraction can be increased on the basis of the original single-period image, forming multi-period remote sensing image data about the target ground object.

[0012] In the above process, the processing of S3 can obtain new feature data completely opposite to the feature information of the target ground object, i.e. the third feature index data, which can depict the different features of the target ground object and other ground objects from a different angle, further enriching the difference information of the target ground object and other ground objects. In some specific embodiments, it is equivalent to obtaining another period of remote sensing data for identifying the target ground object.

[0013] According to some specific embodiments of the present application, the remote sensing feature index is various remote sensing vegetation indices of green vegetation, various remote sensing water indices of water body, various drought remote sensing indices of land surface, and the like, and more specifically, normalized difference vegetation index (NDVI), normalized difference water index (NDWI), temperature vegetation drought index (TVDI).

[0014] Preferably, when the remote sensing feature index is an unnormalized feature index, it is normalized first, and the normalized data is used as the first feature index data.

[0015] According to some preferred embodiments of the present application, the power function is a quadratic to fourth power function.

[0016] With the increase of the exponential constant in the power function, the effect of enlarging the feature difference between the target ground object and the interference ground object with similar spectral feature information is more significant, but an excessively large exponential constant will affect the ground object information extraction effect to some extent. The inventors have unexpectedly found that the quadratic to fourth power function has the best effect.

[0017] According to some preferred embodiments of the present application, the centering processing uses a processing model as follows:

[0018] y = x - u;

[0019] wherein x represents the first feature index data or the second feature index data, y represents the data after centering processing, and u represents the statistical average value of x.

[0020] The centering processing can centralize the data of the target ground object to a specific value range, so that the difference between the data of the target ground object and the data of the non-target ground object is more significant.

[0021] According to some embodiments of the present application, in the S5, the extraction is realized by one or more of principal component analysis, unsupervised classification, and decision tree classification, preferably by principal component analysis.

[0022] According to the above extraction method, a device for extracting ground object information from single-period remote sensing images is obtained, which comprises a storage medium storing a program, algorithm, and / or data structure for realizing any of the above extraction methods.

[0023] The present application has the following advantages:

[0024] The existing remote sensing classification methods are difficult to accurately extract target ground objects from single-period remote sensing data when the spectral information of the ground objects is similar. The extraction method of the present application can construct feature information remote sensing data of multiple target ground objects from single-period remote sensing data. Based on the constructed multiple sets of remote sensing data, other remote sensing classification methods based on multi-period data can be applied to extract target ground object information.

[0025] The extraction method of the present application can construct feature sequence remote sensing data that greatly distinguishes the target ground objects from other background ground objects and the interference ground objects with similar spectral information, and that has distinct characteristics of the target ground objects, thereby improving the extraction accuracy and reducing the extraction difficulty based on single-period remote sensing data.

[0026] The extraction method of the present application has a simple process, few human subjective factors, and universality, and can be directly extracted automatically, thereby further improving the extraction accuracy and efficiency. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 The flowchart of the method of the present application provided in the specific embodiments.

[0028] Figure 2 The winter wheat planting spatial distribution map obtained in the examples. DETAILED DESCRIPTION

[0029] The present application is described in detail below in conjunction with the examples and drawings, but it should be understood that the examples and drawings are only used to exemplarily describe the present application and cannot constitute any limitation on the protection scope of the present application. All reasonable modifications and combinations within the scope of the inventive concept of the present application fall within the protection scope of the present application.

[0030] Referring to the drawings Figure 1 According to the technical scheme of the present application, some embodiments of the ground object information extraction method from single-period remote sensing images comprise the following steps:

[0031] S1: obtaining multi-spectral remote sensing image data of a region to be extracted for ground object information, wherein the multi-spectral remote sensing image data has a medium-high spatial resolution.

[0032] S2 calculating the feature index of the target ground object according to the obtained single-period multispectral remote sensing image data to obtain remote sensing feature index data of the target ground object, i.e. first feature index data;

[0033] S3 performing single or multiple grid operations on the first feature index data by using a power function to obtain second feature index data that enhances the difference in information between the target ground object and the interference ground object;

[0034] S4 performing centralization processing and negation processing on the first feature index data or the second feature index data to obtain third feature index data in which the feature information is opposite to the feature index;

[0035] S5 composing feature sequence data of the target ground object from the first feature index data, the second feature index data and the third feature index data;

[0036] S6 extracting target ground object information from the feature sequence data.

[0037] In some embodiments, the feature index in S2 can be various remote sensing vegetation indices of green vegetation (such as normalized difference vegetation index NDVI, etc.), various remote sensing water indices of water body (such as normalized difference water index NDWI, etc.), various drought remote sensing indices of land surface (temperature vegetation drought index TVDI, etc.), etc.

[0038] In some embodiments, the power function in S3 is a quadratic function, a cubic function and a quartic function. With the increase of the number of the power function, the feature difference between the target ground object and the interference ground object with similar spectral feature information is enlarged, and the other ground objects are removed, and the area of the extracted target ground object is reduced. The remote sensing classification accuracy mainly includes user accuracy and producer accuracy. When the classification result focuses more on the producer accuracy of the target ground object, a power function with a smaller number should be selected, such as a quadratic function. When the classification result focuses more on the user accuracy, a power function with a larger number should be selected, such as a quartic function.

[0039] In some embodiments, the standardization processing in S4 adopts the method of y=x-u, x can be the first feature index data or the second feature index data, y is the centralized data, and u is the average value of the x data after statistics.

[0040] In some embodiments, the extraction in S6 preferably adopts an extraction method with less human subjective factors and simple principle and process, such as unsupervised method of clustering, principal component analysis method, etc.

[0041] Embodiment 1

[0042] As shown in the above embodiment, the remote sensing feature index data of the target ground object is obtained by calculating the feature index of the target ground object according to the obtained single-period multispectral remote sensing image data, and the feature sequence data of the target ground object is composed of the first feature index data, the second feature index data and the third feature index data. The target ground object information is extracted from the feature sequence data. Figure 1The shown single period remote sensing image ground object information extraction method extracts the winter wheat planting area and its spatial distribution information in a certain place and performs precision verification.

[0043] The remote sensing image used can be selected from medium and high resolution remote sensing images, such as SPOT, GF1, TM / ETM, etc. In this embodiment, the Sentinel 2 remote sensing image data with a spatial resolution of 10 meters is used.

[0044] The specific extraction process is as follows:

[0045] (1) Select single period remote sensing image data in the area to be extracted; in this embodiment, the Sentinel 2 multispectral data with a spatial resolution of 10 meters at the heading stage of winter wheat (April 18) is selected.

[0046] (2) Calculate the normalized difference vegetation index (NDVI) representing the characteristic information of winter wheat planting through the multispectral data obtained above, and obtain the NDVI data as the first characteristic index data.

[0047] (3) Based on the calculated NDVI data, a quadratic function, a cubic function or a quartic function is used for calculation to generate new data of winter wheat characteristic information enhancement, i.e. NDVI^2 data, NDVI^3 data or NDVI^4 data, as the second characteristic index data.

[0048] In the sample image, one point is selected at the corresponding position of the winter wheat ground object and other green vegetation with similar spectrum, and the values of the corresponding positions before and after the power function calculation in step (3) are compared, as shown in Table 1 below:

[0049] Table 1 Difference in characteristic values of winter wheat and other green vegetation with similar spectrum before and after power function calculation

[0050]

[0051] NDVI value range is -1 ~ 1, green vegetation features in the NDVI high value area tends to 1, non-green vegetation features in the NDVI low value area is negative or positive tends to 0. From table 1 can be seen: before applying power function calculation, winter wheat and spectrum similar other green vegetation interference features are in the NDVI high value area, winter wheat NDVI value is slightly higher, two similar green features NDVI data difference is 0.206, accounting for 12.35% of the sum of two features NDVI value; after power calculation two features characteristic value decreases with the power of the power function increases, two features characteristic value difference is gradually increasing, to the fourth power function can be seen that the characteristic value of winter wheat (0.771) is in the high value area tends to 1, while the spectrum similar other green features characteristic value (0.286) is in the low value area tends to 0.

[0052] (4) using the centralization processing model y = x-u to the second characteristic index data or the first characteristic index data, the processed data is negated, and new data completely opposite to the target feature NDVI data is obtained, that is, the anti-NDVI characteristic data, as the third characteristic index data.

[0053] In this embodiment, the winter wheat value in the first characteristic data NDVI is the maximum, located in the NDVI high value area, and other non-green features are located in the NDVI low value area; the third characteristic index data is obtained by centralization processing and negation processing, which not only makes the characteristics of winter wheat and other non-green features completely opposite, but also further enhances the difference characteristics between them; the third characteristic index data is similar to the NDVI data of winter wheat after harvesting in early June, which is equivalent to obtaining another very important period of remote sensing data for identifying the target feature winter wheat.

[0054] In the above process, before centralization processing, the characteristic data of the target feature winter wheat is located in the positive value area, and part of other non-green features is located in the positive value area and part is located in the negative value area. After centralization processing, the characteristic data of the target feature winter wheat is still in the positive value area, and other non-green features are all located in the negative value area.

[0055] (5) constructing the characteristic sequence data of winter wheat from the obtained first characteristic index data, second characteristic index data and third characteristic index data.

[0056] (6) performing principal component analysis on the characteristic sequence data, and selecting the principal component data containing the characteristic information of winter wheat. The part less than 0 in the principal component data is the winter wheat planting area.

[0057] In the principal component processing process, the selection rule of the principal component data containing the characteristic information of the target feature winter wheat is:

[0058] The data is inputted in the order of the first characteristic index data, the second characteristic index data and the third characteristic index data;

[0059] According to whether the area of the target ground object is greater than or equal to the area of the non-target ground object and the eigenvalue rule of the eigenvector matrix between the input data after principal component analysis and each principal component, the principal component data containing the characteristic information of the target ground object winter wheat is selected, and the specific selection rule is shown in Table 2:

[0060] Table 2 Selection rule of principal component data containing characteristic information of target ground object winter wheat

[0061]

[0062]

[0063] As shown in Table 2, when the area of the target ground object is greater than or equal to the area of the non-target ground object, if the third characteristic index data is constructed based on the first characteristic index data, the principal component whose corresponding eigenvalue of the first and third input data is positive and the corresponding eigenvalue of the second input data after power operation is negative is the characteristic information of the target ground object winter wheat; if the third characteristic index data is constructed based on the second characteristic index data, the principal component whose corresponding eigenvalue of the first and second input data is negative and the corresponding eigenvalue of the third input data is positive is the characteristic information of the target ground object winter wheat. When the area of the target ground object is less than the area of the non-target ground object, regardless of how the third characteristic index data is constructed, the principal component whose corresponding eigenvalue of the first and second input data is negative and the corresponding eigenvalue of the third input data is positive is the characteristic information of the target ground object winter wheat.

[0064] Example 2

[0065] The ground object information extraction is performed according to the process of Example 1, wherein the quadratic function is selected in step (3), and the third characteristic index data is selected for centralization and inversion processing in step (4).

[0066] The eigenvector matrix between the input data after principal component analysis and each principal component is shown in Table 3:

[0067] Table 3 Eigenvector between input data and principal component in Example 2

[0068]

[0069] In this example, the area of the target ground object winter wheat in the test area is greater than the area of the non-winter wheat ground object, the second principal component whose corresponding eigenvalue of the first and second input data is negative and the corresponding eigenvalue of the third input data is positive is selected as the characteristic information data of the winter wheat, and the part less than 0 in the second principal component data is the winter wheat, and the obtained result is shown in FIG. 2. Figure 2As shown.

[0070] The random point generation and confusion matrix method is applied to verify the precision of the winter wheat planting result extracted by the embodiment, and the verification result is shown in Table 4.

[0071] Table 4: Confusion matrix information statistics of winter wheat planting information extracted by Example 2

[0072]

[0073] From Figure 2 It can be seen from the spatial distribution map of winter wheat planting that, in addition to the obvious separation of urban agglomeration and roads, the field roads and part of the non-winter wheat plots can be clearly seen after zooming in. It can be seen from the accuracy verification result in Table 4 that the winter wheat producer accuracy reaches 95.43%, and each accuracy is above 90%, and the Kappa coefficient is 0.86. Therefore, whether the result map or the accuracy verification result shows that, based on the spatial resolution of 10 meters of remote sensing data, the method of the present application not only has less human subjective factor interference, but also requires less data, and only one period of remote sensing data can achieve good target ground object extraction result.

[0074] The above embodiments are only preferred embodiments of the present application, and the protection scope of the present application is not limited to the above embodiments. Any technical solution falling within the idea of the present application belongs to the protection scope of the present application. It should be pointed out that, for ordinary skilled persons in the art, improvements and refinements without departing from the principles of the present application should also be considered as the protection scope of the present application.

Claims

1. A method for extracting ground feature information based on single-period remote sensing imagery, characterized in that, It includes: S1 Based on single-period remote sensing image data, obtain the remote sensing feature index information of the target ground object, namely the first feature index data; S2 performs one or more raster operations on the first feature index data using a power function to obtain the second feature index data. S3 performs a centralization process on the first feature index data or the second feature index data, and then performs an inversion process on the centralized data to obtain the third feature index data. S4 consists of the first feature index data, the second feature index data, and the third feature index data, forming the feature sequence data of the target land feature; S5. Target feature information is extracted from the feature sequence data using principal component analysis. Wherein, the power function is a second to fourth power function; the centering process uses the following processing model: y=xu; where x represents the first feature index data or the second feature index data, y represents the data after centering, and u represents the statistical average of x; The remote sensing feature index is selected from the Normalized Difference Vegetation Index (NDVI), and the third feature index data is obtained by performing the centering and inversion processing on the first feature index data. Therefore, the principal component analysis method includes: Data is input in the order of the first characteristic index data, the second characteristic index data, and the third characteristic index data, which are the first to third input data respectively; Principal component analysis was performed on the first to third input data to obtain their respective principal component eigenvalues, which are the first to third eigenvalues. When the area of ​​the target feature is greater than or equal to the area of ​​the non-target feature, the principal components with positive first and third feature values ​​and negative second feature value are used as the feature information of the target feature. When the area of ​​the target feature is smaller than the area of ​​the non-target feature, the principal components with negative first and second feature values ​​and positive third feature value are used as the feature information of the target feature. Alternatively, if the remote sensing feature index is selected from the Normalized Difference Vegetation Index (NDVI), and the third feature index data is obtained by performing the centering and inversion processing on the second feature index data, then the principal component analysis method includes: Data is input in the order of the first characteristic index data, the second characteristic index data, and the third characteristic index data, which are the first to third input data respectively; Principal component analysis was performed on the first to third input data to obtain their respective principal component eigenvalues, which are the first to third eigenvalues. When the area of ​​the target feature is greater than or equal to the area of ​​the non-target feature, the principal components with negative first and second feature values ​​and positive third feature value are used as the feature information of the target feature. When the area of ​​the target feature is smaller than the area of ​​the non-target feature, the principal components with negative first and second feature values ​​and positive third feature value are used as the feature information of the target feature.

2. The method for extracting ground feature information according to claim 1, characterized in that, The extraction method is implemented through automated programs and / or devices.

Images