CN111062265A - Shadow detection method suitable for hyperspectral remote sensing image - Google Patents

Abstract

The invention relates to a shadow detection method suitable for a hyperspectral remote sensing image. (1) Selecting spectral wavelengths of the hyperspectral images by adopting a continuous projection algorithm, and screening out red light and near infrared characteristic wave bands; (2) analyzing spectral characteristics and differences of vegetation in shadow areas, vegetation in bright areas and water areas, particularly the differences of typical ground object pixels in near-infrared bands and normalized difference vegetation indexes; (3) rejecting non-target interference information in the image, analyzing based on the preferred characteristic wave band and spectrum difference, establishing a shadow vegetation index suitable for the hyperspectral remote sensing image and normalizing the shadow vegetation index; (4) based on a step method, the optimal threshold value of the index is set, so that the effective detection of the vegetation in the shadow area is realized. The method can effectively distinguish the vegetation in the shadow area from the vegetation in the bright area, the water body area and other ground objects in the hyperspectral remote sensing image, realize shadow detection and lay a foundation for image shadow removal and information restoration.

Description

Shadow detection method suitable for hyperspectral remote sensing image

Technical Field

The invention relates to the fields of geography, surveying and mapping science and engineering, forestry, ecology and the like, in particular to a shadow detection method suitable for a hyperspectral remote sensing image.

Background

Due to the comprehensive influence of factors such as the solar incident angle, the height of the ground object, the topographic relief and the like, two levels of brightness and shadow generally exist on the remote sensing image. The brightness value of the bright area is higher due to the irradiation of sunlight; the shadow is formed by shielding the bright light by an object higher than the ground, and is divided into an intrinsic shadow and a falling shadow. The shadow area reduces the information quantity of the target ground object represented on the remote sensing image, and greatly interferes with the development of work such as ground object identification, information extraction, quantitative algorithm construction and the like. With the continuous development of the domestic satellite technology and the continuous expansion of the application field of remote sensing images, how to effectively detect and eliminate the influence of shadows becomes a problem to be solved urgently.

The hyperspectral remote sensing image has the characteristics of multiple channels, narrow band intervals, integrated atlas and the like, and has unique advantages in classification identification and information extraction of vegetation in shadow areas. However, the hyperspectral remote sensing image has more channels, more complex spectrum composition and more low-frequency information, so that the spectrum unmixing and ground object identification in the shadow area are more difficult, and the accurate application and the rapid development of the data are limited to a certain extent. Therefore, the invention provides a shadow detection method suitable for a hyperspectral remote sensing image, which aims to provide technical reference for hyperspectral remote sensing shadow detection and provide support for shadow removal, shadow area information restoration and extraction and other work.

Disclosure of Invention

The invention aims to provide a shadow detection method suitable for a hyperspectral remote sensing image, which can meet the requirements of accurate detection and identification of shadows in the hyperspectral image.

In order to achieve the purpose, the technical scheme of the invention is as follows: a shadow detection method suitable for a hyperspectral remote sensing image comprises the following steps:

s1, performing spectral wavelength optimization on the hyperspectral image by adopting a continuous projection algorithm, and screening out red light and near infrared characteristic wave bands; the method comprises the following specific steps:

(1) selecting a spectrum column vector as a starting vector in the spectrum matrix;

(2) respectively calculating projection vectors of the residual column vectors on an orthogonal plane of the initial vector;

(3) selecting the maximum projection as the initial vector of the next projection until the number of the selected variables reaches the maximum required number;

(4) performing multiple linear regression on all the extracted wavelength combinations, wherein the combination corresponding to the minimum root mean square error is the optimal wavelength combination;

s2, analyzing spectral characteristics and differences of the vegetation in the shadow area, the vegetation in the bright area and the water body area, particularly the differences of the typical feature pixels on the near infrared band and the normalized difference vegetation index, specifically selecting a plurality of sample points on three typical features of the image, counting the mean value of each band of the image, drawing a curve, analyzing and comparing;

step S3, eliminating non-target interference information in the image, analyzing based on the optimized characteristic wave band and spectrum difference, and obtaining the formula

Establishing a shadow vegetation index suitable for the hyperspectral remote sensing image and normalizing the shadow vegetation index;

and step S4, setting the optimal threshold value of the shadow vegetation index based on the step method, thereby realizing the detection of the vegetation in the shadow area.

Compared with the prior art, the invention has the following beneficial effects: the method can effectively distinguish the vegetation in the shadow area from the vegetation in the bright area, the water body area and other ground objects in the hyperspectral remote sensing image, realize shadow detection and lay a foundation for image shadow removal and information restoration.

Drawings

FIG. 1 shows the image (RGB: 14, 6, 2) of each angle of the pseudo-color composite after the preprocessing.

Fig. 2 is a characteristic band selection based on SPA.

FIG. 3 shows the bright area vegetation, shadow area vegetation, water area all bands andNDVIand (6) comparing.

FIG. 4 shows threshold selection results of PROBA/CHRIS step-up method.

FIG. 5 shows threshold results obtained by the Zhuhai step one method.

FIG. 6 shows the results of PROBA/CHRIS 5 angular image classification.

FIG. 7 shows the classification result of the first image of Zhuhai.

Detailed Description

The technical scheme of the invention is specifically explained below with reference to the accompanying drawings.

The invention provides a shadow detection method suitable for a hyperspectral remote sensing image, which is realized by the following steps:

(1) the method comprises the following steps of adopting a continuous projection algorithm to carry out spectrum wavelength optimization on a hyperspectral image, screening out red light and near infrared characteristic wave bands, and specifically comprising the following steps:

① selecting a spectrum column vector as a starting vector in the spectrum matrix;

② calculating projection vectors of the remaining column vectors on orthogonal planes of the start vector;

③ selecting the maximum projection as the initial vector of the next projection until the number of the selected variables reaches the maximum required number;

④ performing multiple linear regression on all the extracted wavelength combinations, wherein the combination corresponding to the minimum Root Mean Square Error (RMSE) is the optimal wavelength combination.

(2) And analyzing spectral characteristics and differences of vegetation in shadow areas, vegetation in bright areas and water areas, particularly the differences of the typical ground object pixels in near-infrared bands and normalized difference vegetation indexes. Selecting a plurality of sampling points from the three ground objects of the image, and counting the mean value of each wave band of the image to draw a curve for analysis and comparison.

(3) Rejecting non-eyes in an imageInterference information, based on preferred characteristic band and spectral difference analysis, based on

Establishing a shadow vegetation index suitable for the hyperspectral remote sensing image and normalizing the shadow vegetation index;

(4) based on the step method, the optimal threshold value of the shadow vegetation index is set, so that the detection of the vegetation in the shadow area is realized.

The following are specific examples of the present invention.

The test data are hyper-spectral data PROBA/CHRIS and Zhuhai No. I. The data is not limited to the examples provided.

(1) Selecting a characteristic wave band of the hyperspectral remote sensing image: firstly, preprocessing a hyperspectral image, comprising the following steps: noise removal, atmospheric correction, ortho correction, cropping, etc. (fig. 1). Then, the wave bands are classified according to the central wavelength (table 1), sensitive features of bright-area vegetation, shadow-area vegetation and water body areas are screened from the image spectrum data by using a continuous projection algorithm (SPA), and a sensitive feature wave band is selected when RMSE is minimum (figure 2).

(2) Typical spectral feature analysis: selecting a plurality of test sample points of the vegetation in the bright area, the vegetation in the shadow area and the water body area in the hyperspectral image, respectively counting the average values of the original wave bands and the NDVI of the image, drawing a spectral curve for comparative analysis (figure 3).

(3) Construction of NSVI: firstly, eliminating other ground feature information in the image, such as a building area and the like, and then analyzing based on the optimized characteristic wave band and the spectrum difference according to the formula

Establishing a shadow vegetation index and normalizing the shadow vegetation index to obtain NSVI;

(4) evaluation of the effects: based on a step method, the optimal threshold value of the index is set, so that the effective detection of the vegetation in the shadow area is realized. The specific operation thought is as follows: randomly generating a plurality of points to perform precision evaluation on the classification results under different thresholds, wherein the precision evaluation indexes show the characteristics of a parabola which is raised first and then lowered and has a downward opening, and accordingly, the threshold with the best classification effect can be determined (figures 4-5). PROBA/CHRIS and Zhuhai No. one threshold selection and precision verification are shown in tables 2-5:

it can be seen from the above classification results that the 0 ° image has the best extraction precision, i.e., + -36 ° times, and finally + -55 °, but the classification precision is all above 94%, the Kappa coefficient is generally above 0.89 (table 4), and fig. 6 is the PROBA/CHRIS 5 angular image classification result. The reason for this is probably that the image is distorted to a certain extent along with the expansion of the zenith angle of flight, the quality is poor, the whole image is fuzzy, and the shadow detection is influenced to a certain extent. The inverse view of the first image of the pearl sea is better in image quality, the classification result is also considerable, the total accuracy is 96.33%, the Kappa coefficient is 0.9382 (table 5), and fig. 7 shows the classification result of the first image of the pearl sea. In conclusion, the NSVI constructed by the method can effectively distinguish bright-area vegetation, shadow-area vegetation and water body areas.

In summary, according to the present invention, the NSVI index is constructed in the hyperspectral image, and on the premise that reasonable thresholds are selected, the present invention has strong shadow detection capability, and can provide important support for the shadow removal and the shadow information restoration of the hyperspectral image.

The above are preferred embodiments of the present invention, and all changes made according to the technical scheme of the present invention that produce functional effects do not exceed the scope of the technical scheme of the present invention belong to the protection scope of the present invention.

Claims (1)

1. A shadow detection method suitable for a hyperspectral remote sensing image is characterized by comprising the following steps:

s1, performing spectral wavelength optimization on the hyperspectral image by adopting a continuous projection algorithm, and screening out red light and near infrared characteristic wave bands; the method comprises the following specific steps:

(1) selecting a spectrum column vector as a starting vector in the spectrum matrix;

(2) respectively calculating projection vectors of the residual column vectors on an orthogonal plane of the initial vector;

(3) selecting the maximum projection as the initial vector of the next projection until the number of the selected variables reaches the maximum required number;

(4) performing multiple linear regression on all the extracted wavelength combinations, wherein the combination corresponding to the minimum root mean square error is the optimal wavelength combination;

s2, analyzing spectral characteristics and differences of the vegetation in the shadow area, the vegetation in the bright area and the water body area, particularly the differences of the typical feature pixels on the near infrared band and the normalized difference vegetation index, specifically selecting a plurality of sample points on three typical features of the image, counting the mean value of each band of the image, drawing a curve, analyzing and comparing;

step S3, eliminating non-target interference information in the image, analyzing based on the optimized characteristic wave band and spectrum difference, and obtaining the formula

Establishing a shadow vegetation index suitable for the hyperspectral remote sensing image and normalizing the shadow vegetation index;

and step S4, setting the optimal threshold value of the shadow vegetation index based on the step method, thereby realizing the detection of the vegetation in the shadow area.

Images