WO2018076137A1

WO2018076137A1 - Method and device for obtaining hyper-spectral image feature descriptor

Info

Publication number: WO2018076137A1
Application number: PCT/CN2016/103069
Authority: WO
Inventors: 李岩山; 范雷东; 刘鹏; 夏荣杰; 谢维信; 黄庆华
Original assignee: 深圳大学
Priority date: 2016-10-24
Filing date: 2016-10-24
Publication date: 2018-05-03

Abstract

A method for obtaining a hyper-spectral image feature descriptor, comprising: obtaining direction angles of various image elements within a pre-set neighbourhood of a point of interest in a spatial spectral domain of a hyper-spectral image (S101); establishing a direction angle statistical histogram using the direction angle of each of the image elements, and obtaining an angle corresponding to the peak value in the direction angle statistical histogram (S102); by taking the angle corresponding to the peak value in the direction angle statistical histogram as a main direction, rotating the hyper-spectral image according to the main direction to obtain a rotated hyper-spectral image (S103); and acquiring a feature vector of the rotated hyper-spectral image and saving the feature vector as a descriptor of the hyper-spectral image (S104). A descriptor obtained according to the method still has good robustness and uniqueness where rotation transformation, illumination transformation and noise interference occurs to a hyper-spectral image.

Description

Method and device for acquiring hyperspectral image feature descriptor

Technical field

The invention belongs to the field of computers, and in particular relates to a method and a device for acquiring descriptors of interest points of a hyperspectral image in a hyperspectral image based on a spectral angle.

Background technique

At present, hyperspectral images have been widely used in civil and military fields such as remote sensing, atmospheric, agricultural monitoring, medical spectral imaging diagnosis, food detection, and safety. Hyperspectral image processing technology based on spectral analysis has been successfully applied to target detection, fine feature classification, mineral mapping and other fields.

With the improvement of remote sensing technology, hyperspectral image data has the characteristics of high spatial resolution, large number of bands and large amount of data. Traditional spectral curve-based hyperspectral image classification technology has been difficult to process high spatial resolution hyperspectral images. . Therefore, in recent years, there have been many methods for studying hyperspectral images using points of interest. In the paper 'Interest Points for Hyperspectral Image Data', Mukherjee combines the SIFT algorithm (Scale-invariant feature transform) with PCA technology (principal components analysis) and applies it to hyperspectral images. The detection of points of interest. Adding PCA dimensionality reduction technology to SIFT algorithm greatly reduces the computational complexity of interest point search, but also loses spectral information. Later generations improved it. In the paper 'A Vector SIFT Detector for Interest Point Detection in Hyperspectral Imagery', Leidy P.

Taking full account of the vector characteristics of spectral image pixels, considering it as a vector image, the number of points of interest obtained is more, and the robustness and uniqueness are stronger. However, when describing the points of interest, both methods directly use the spectral curve of the point of interest as its descriptor. In the image recognition process, it is important to use a compact and complete feature description. In the SIFT algorithm of two-dimensional image, Lowe performs the histogram statistical calculation descriptor on the gradient of the pixel in the neighborhood of the interest point, so that the interest point has certain rotation invariance, illumination invariance and affine invariance.

In the prior art, when describing the points of interest of hyperspectral images, the spectral curve of the point of interest is directly used as its descriptor. This method cannot guarantee the rotation transformation, illumination transformation and noise interference of the hyperspectral image. The descriptor has certain robustness and uniqueness.

Summary of the invention

The technical problem to be solved by the present invention is to provide a method and a device for acquiring a hyperspectral image feature descriptor, which are intended to solve the problem in the prior art that in the case of a hyperspectral image rotation transformation, illumination transformation, and noise interference, the description cannot be guaranteed. The child has certain problems of robustness and uniqueness.

The present invention is implemented in such a manner that a method for acquiring a hyperspectral image feature descriptor includes:

Obtaining a direction angle of each pixel in a preset neighborhood of the hyperspectral image null spectrum domain of interest;

Generating a direction angle statistical histogram according to a direction angle of each of the pixels, and acquiring an angle corresponding to a peak in the direction angle statistical histogram;

Calculating, according to the direction angle, an angle corresponding to a peak in the histogram as a main direction, and rotating the hyperspectral image according to the main direction to obtain a rotated hyperspectral image;

Obtaining a feature vector of the rotated hyperspectral image, and saving the feature vector as a descriptor of the hyperspectral image.

Further, the obtaining the direction angle of each pixel in the preset neighborhood of the hyperspectral image null spectrum domain interest point comprises:

Obtaining a circular neighborhood of the point of interest by using the hyperspectral image spatial spectral domain interest point as a dot and a preset distance as a radius;

Determining a square neighborhood of the sampling point according to a preset side length by using each pixel in the circular neighborhood as a sampling point;

Calculating a spectral angle of each pixel in the square neighborhood and the sampling point, and determining a pixel having the largest spectral angle in the square neighborhood;

Calculating an angle between a pixel having the largest spectral angle in the square neighborhood and the sampling point, and using the angle as the The direction angle of the sampling point.

Further, the establishing a direction angle statistical histogram according to the direction angle of each of the pixels includes:

The direction angle of the sampling point is the horizontal axis, and the amplitude accumulated value corresponding to the direction angle of the sampling point after the weight processing is the vertical axis, and the direction angle statistical histogram is established.

Further, the acquiring the feature vector of the rotated hyperspectral image includes:

Dividing the rotated hyperspectral image into equal intervals into a plurality of sub-regions;

Obtaining feature vectors of each of the sub-regions;

The feature vectors of the sub-regions are spliced to obtain feature vectors of the rotated hyperspectral image.

Further, the acquiring feature vectors of each of the sub-regions includes:

Calculating a direction angle of each pixel in the sub-area, and establishing a direction angle statistical histogram of the sub-area according to a direction angle of each pixel in the sub-area;

The component of the feature vector of the sub-region is determined by the direction angle of the sub-region, and the component of the feature vector of the sub-region is determined.

The invention also provides a device for acquiring a hyperspectral image feature descriptor, comprising:

a direction angle acquiring unit, configured to acquire a direction angle of each pixel in a preset neighborhood of the high-spectrum image null spectrum domain interest point;

a main direction angle acquiring unit, configured to establish a direction angle statistical histogram according to a direction angle of each of the pixels, and obtain an angle corresponding to a peak in the direction angle statistical histogram;

An image rotation unit configured to calculate an angle corresponding to a peak in the histogram in the direction direction as a main direction, and rotate the hyperspectral image according to the main direction to obtain a rotated hyperspectral image;

a descriptor acquisition unit is configured to acquire a feature vector of the rotated hyperspectral image, and save the feature vector as a descriptor of the hyperspectral image.

Further, the direction angle acquiring unit includes:

a circular neighborhood determining module, configured to obtain a circular neighborhood of the point of interest by using the hyperspectral image spatial domain point of interest as a circle and using a preset distance as a radius;

a square neighborhood determining module, configured to determine, by using each pixel in the circular neighborhood as a sampling point, a square neighborhood of the sampling point according to a preset side length;

a spectral angle calculation module, configured to calculate a spectral angle of each pixel in the square neighborhood and the sampling point, and determine a pixel with the largest spectral angle in the square neighborhood;

An angle calculation module is configured to calculate an angle between a pixel having the largest spectral angle in the square neighborhood and the sampling point, and the angle is used as a direction angle of the sampling point.

Further, the main direction angle acquiring unit is specifically configured to:

Further, the descriptor obtaining unit includes:

An image dividing module, configured to divide the rotated hyperspectral image into several sub-regions at equal intervals;

And a vector acquiring module, configured to acquire a feature vector of each of the sub-regions, and splicing the feature vectors of the sub-regions to obtain a feature vector of the rotated hyperspectral image.

Further, the vector acquisition module includes:

a direction angle calculation sub-module, configured to calculate a direction angle of each picture element in the sub-area, and establish a direction angle statistical histogram of the sub-area according to a direction angle of each picture element in the sub-area;

The vector determining sub-module is configured to calculate, according to a direction angle of the sub-region, a component of a feature vector corresponding to each column in the histogram, and a feature vector of the sub-region.

Compared with the prior art, the present invention has the beneficial effects that the embodiment of the present invention calculates the rotated hyperspectral image by acquiring the main direction of the neighborhood of the interest point of the hyperspectral image in the optical spectrum domain and rotating according to the main direction. The feature vector is represented by the feature vector as a descriptor of the hyperspectral image. This embodiment generates a description using the spatial information and spectral information of the point of interest. Sub-, with certain robustness and uniqueness, while using the main direction of the neighborhood of the point of interest to perform hyperspectral image rotation, obtain the rotated feature vector, and use the feature vector as a descriptor to ensure that the descriptor does not rotate. Denaturation and robustness, the descriptors still have good robustness and uniqueness in the case of hyperspectral image rotation transformation, illumination transformation and noise interference.

DRAWINGS

FIG. 1 is a flowchart of a method for acquiring a hyperspectral image feature descriptor according to a first embodiment of the present invention.

FIG. 2 is a flowchart of a method for acquiring a hyperspectral image feature descriptor according to a second embodiment of the present invention.

FIG. 3 is a schematic diagram of a neighborhood of a point of interest provided by a second embodiment of the present invention.

4a is a schematic diagram showing a direction angle of each pixel in a circular neighborhood of a point of interest according to a second embodiment of the present invention.

4b is a histogram of the direction angle statistics of the circular neighborhood provided by the second embodiment of the present invention.

FIG. 5a is a schematic diagram of a sub-area of a circular neighborhood of a point of interest of a rotated hyperspectral image according to a second embodiment of the present invention.

FIG. 5b is a directional statistical histogram of a sub-area provided by a second embodiment of the present invention.

FIG. 6 is a schematic structural diagram of an apparatus for acquiring a hyperspectral image feature descriptor according to a third embodiment of the present invention.

FIG. 7 is a schematic structural diagram of a direction angle acquiring unit according to a fourth embodiment of the present invention.

FIG. 8 is a schematic structural diagram of a description sub-acquisition unit according to a fifth embodiment of the present invention.

detailed description

The present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It is understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

In image processing, an ideal descriptor should have high robustness and uniqueness. Robustness mainly refers to the ability of the descriptor to work stably under the conditions of image affine transformation, density transformation and noise interference. Uniqueness means description The sub-capture has the ability to capture and reflect this change as the local image structure of the point of interest changes.

The invention relates to determining a direction reference according to a partial image structure of a point of interest, a main direction, thereby realizing rotation invariance of the image, and generating a descriptor using the pixel direction angle information of the rotated image. The SIFT algorithm uses the gradient information of the two-dimensional image to obtain the main direction of the local structure. After completing the image gradient calculation of the neighborhood of the interest point, the histogram is used to calculate the gradient direction and amplitude of the pixels in the neighborhood. Different from the method of using image gradient in two-dimensional image, the method uses the distribution feature of the spectral angle of the neighborhood pixel of interest point to determine the main direction.

Based on the foregoing description, the present invention provides a first embodiment, as shown in FIG. 1, a method for acquiring a hyperspectral image feature descriptor, including:

S101. Acquire a direction angle of each pixel in a preset neighborhood of the high-spectral image null spectrum domain interest point.

In this step, the acquiring device acquires the direction angle of each pixel in the preset neighborhood of the high-spectral image interest point, which is convenient for calculation. In this step, the preset neighborhood is a circular neighborhood.

S102. Establish a directional angle histogram according to a direction angle of each of the pixels, and obtain an angle corresponding to the peak in the directional statistical histogram.

In this step, the acquiring device establishes a direction angle statistical histogram according to the direction angle of each pixel in the preset neighborhood, where the horizontal axis of the histogram is the angle of the direction angle, and the vertical axis is the direction angle of the pixel. Amplitude cumulative value. After establishing the direction angle statistical histogram, the acquiring device will acquire the angle of the direction angle corresponding to the peak in the direction angle statistical histogram.

S103: Calculate an angle corresponding to a peak in the histogram in the direction direction as a main direction, and rotate the hyperspectral image according to the main direction to obtain a rotated hyperspectral image.

In this step, the acquisition means rotates the hyperspectral image in the main direction by the angle of the direction angle corresponding to the peak in the histogram in the direction angle of the step S102, and obtains the rotated hyperspectral image. Specifically, the acquisition device rotates the hyperspectral image clockwise according to the main direction, centering on the point of interest.

S104. Acquire a feature vector of the rotated hyperspectral image, and save the feature vector as a descriptor of the hyperspectral image.

In this step, after obtaining the rotated hyperspectral image, the acquiring device acquires the feature vector of the rotated hyperspectral image, and uses the feature vector as a descriptor of the hyperspectral image.

The embodiment provides a method for acquiring a descriptor of a hyperspectral image spatial domain interest point, and the obtaining method first calculates a direction angle of each pixel in a preset neighborhood of the hyperspectral image spatial domain interest point, and A main direction is determined by using the direction angle of each pixel of the preset neighborhood of the point of interest, and then the hyperspectral image is rotated by one main direction and the descriptor is created by using the direction information of the neighborhood pixel of the point of interest. The descriptor obtained by the obtaining method of the descriptor provided in this embodiment has certain rotation invariance and robustness.

The present invention further provides a second embodiment as shown in FIG. 2, a method for obtaining a hyperspectral image feature descriptor, comprising:

S201: The circular neighborhood of the interest point is obtained by taking the interest point of the hyperspectral image in the hyperspectral image as a circle and using a preset distance as a radius.

S202, using each pixel in the circular neighborhood as a sampling point, determining a square neighborhood of the sampling point according to a preset side length.

S203. Calculate a spectral angle of each pixel in the square neighborhood and the sampling point, and determine a pixel with the largest spectral angle in the square neighborhood.

S204. Calculate an angle between a pixel having the largest spectral angle in the square neighborhood and the sampling point, and use the included angle as a direction angle of the sampling point.

S205, the direction angle of the sampling point is the horizontal axis, and the amplitude accumulated value corresponding to the direction angle of the sampling point after the weight processing is the vertical axis, and the direction angle statistical histogram is established, and the direction angle is obtained. Count the angles corresponding to the peaks in the histogram.

S206: Calculate an angle corresponding to the peak in the histogram in the direction direction as a main direction, and rotate the hyperspectral image according to the main direction to obtain a rotated hyperspectral image.

S207. The rotated hyperspectral image is equally divided into a plurality of sub-regions;

S208. Calculate a direction angle of each pixel in the sub-area, and establish a direction angle statistical histogram of the sub-area according to a direction angle of each pixel in the sub-area;

S209, the amplitude value of each column in the histogram is calculated by using the direction angle of the sub-region, and the component of the feature vector of the sub-region is determined, and the feature vector of the sub-region is determined;

S210: splicing feature vectors of the sub-regions, acquiring feature vectors of the rotated hyperspectral image, and saving the feature vectors as descriptors of the hyperspectral images.

The following describes the embodiment in detail:

In this embodiment, the acquisition method of the descriptor is mainly divided into calculating a direction angle of each pixel in a preset neighborhood of the interest point, establishing a statistical histogram of the direction angle, obtaining a peak of the statistical histogram of the direction angle, and rotating The post-hyperspectral image is counted on the direction angle and the descriptor is generated in four steps. The specific implementation process is as follows:

(1) Determine the size of the neighborhood

The acquisition method in this embodiment describes the hyper-spectral image in the circular neighborhood where the interest points of the optical spectrum domain are located, and the selection of the radius of the circular neighborhood needs to be set in a specific application to achieve an optimal algorithm. Efficiency and effect, as shown in FIG. 3, before describing the point of interest I, first determine the size of the circular neighborhood where the point of interest I is located, that is, determine the radius R of the circular neighborhood.

After the circular neighborhood of the point of interest I is determined, the direction angle of each pixel in the circular neighborhood needs to be calculated, and the calculation of the direction angle of each pixel is performed in a square neighborhood, as shown in FIG. Assuming that the direction angle of the pixel I _i needs to be calculated, the size of the square neighborhood of the pixel I _i , that is, the side length W of the square neighborhood must first be determined.

The radius R of the circular neighborhood and the length W of the square neighborhood in this embodiment affect the calculation performance of the descriptor acquisition method and the judgment accuracy of the main direction. If the radius R of the circular neighborhood is too small, the main direction will be inaccurate. If the radius R is too large, the calculation amount will be increased and the efficiency of the algorithm will be reduced. The side length W of the square neighborhood will determine the resolution of the direction angle, for example, when calculating the image. when the angular direction of the element I _i, square neighborhood like element I _i is the size of 3 * 3, the image element I _i direction angular resolution of approximately 45 °, like the square neighborhood if the element I _i is 5 * 5 At a large hour, the angular resolution of the pixel I _i is about 22.5°.

(2) Calculation of direction angle

After the circular neighborhood and the square neighborhood are determined, the direction angle of each pixel in the circular neighborhood of the point of interest is calculated. 3, the present embodiment to an angle of maximum spectral image element I _i, for example, first calculates the square neighborhood pixel image I _j spectral angle α with the central element of I _i (I _i, I _j), determination of embodiments Pixel I _k , ie

Then calculate the angle between the pixel I _k and the pixel I _i

Angle

As the direction angle of the pixel I _i .

In the following, the square neighborhood size of the pixel I _i is 3*3 as an example for specific description:

Assuming that the direction angle is calculated for I _i , the central pixel I _i and the other eight pixels I ₁ , I ₂ , I ₃ , I ₄ , I ₅ , I ₆ , I ₇ are respectively calculated in the square neighborhood of 3*3. , I ₈ spectral angle, the calculated results are θ ₁ , θ ₂ , θ ₃ , θ ₄ , θ ₅ , θ ₆ , θ ₇ , θ ₈ , and the largest θ value is selected from the calculation results, that is, θ=max ( θ ₁ , θ ₂ , θ ₃ , θ ₄ , θ ₅ , θ ₆ , θ ₇ , θ ₈ ), let the pixel corresponding to the largest θ value be I _k , then the pixel I _k and the central pixel I _i The angle is the direction angle of the pixel I _i .

(3) Establish a direction angle statistical histogram

After the calculation of the direction angles of all the cells in the circular neighborhood of the point of interest is completed, it is necessary to use the repeated graph to count the direction angles and amplitudes of all the cells in the circular neighborhood. In the present embodiment, the horizontal axis of the direction angle statistical histogram is the direction angle of the pixel, and the vertical axis is the amplitude accumulated value corresponding to the pixel direction angle. The number of columns of the direction angle statistical histogram is determined by the size of the square neighborhood, as shown in Figure 4a, the arrows in the figure indicate the direction angle of the pixel, and Figure 4b shows the square neighborhood when the size is 3*3. The histogram of the direction angle of the circular neighborhood of the interest point. The direction angle statistical histogram has a total of 8 columns, representing the direction angle of other pixels except the central pixel in the square neighborhood, and 8 columns from the left. The angles to the right are 0°, 45°, 90°, 135°, 180°, 225°, 270°, and 315°, respectively. Each sampling point that adds a histogram of the direction angle statistics (the sampling point represents all the cells in the circular neighborhood of the point of interest that need to calculate the direction angle, and all the pixels include the prototype boundary that falls on the circular neighborhood of the point of interest. The magnitude of the pixel needs to be weighted, and the weighting uses a circular Gaussian weighting function, so that the magnitude near the point of interest has a larger weight.

(4) Determine the main direction and generate a descriptor

After the direction angle statistical histogram is established, the angle corresponding to the peak in the direction angle statistical histogram is determined, and the angle is used as the main direction of the point of interest. As shown in FIG. 4b, the main direction of the point of interest is 180°. After rotating the hyperspectral image in one main direction, a rotated hyperspectral image is obtained. The rotated hyperspectral image is shown in Figure 5a.

In this embodiment, after acquiring the rotated spectral image as shown in FIG. 5a, the acquiring device divides the circular neighborhood of the interest point into n*n sub-regions according to a preset dividing rule (FIG. 5a). Medium, n=2), and calculating the direction angle of each pixel in the divided sub-region, and establishing a statistical histogram of the direction angle of the sub-region according to the calculated direction angle of each pixel of the sub-region, The histogram of the direction angle of the sub-area is shown in Figure 5b. The amplitude accumulated value (ie, d ₁ , d ₂ , . . . , d ₈ ) corresponding to each column of the histogram is counted as the component of the sub-region 501 in FIG. 5a as the component of the feature vector D1 of the sub-region 501, and D is generated. ₁ , that is, D ₁ = (d ₁ , d ₂ , ... d ₈ ). There are 2*2 sub-regions in Fig. 5a, so there are 2*2 feature vectors, D ₁ , D ₂ , D ₃ , D ₄ , and the feature vectors of each sub-region are spliced to form the final feature vector of interest points, ie interest. The feature vector of the point D = (D ₁ , D ₂ , D ₃ , D ₄ ). Because there are 2*2 sub-regions in FIG. 5a shown in this embodiment, each sub-region has information of 8 directions, so the feature vector D of the points of interest has 2*2*8 data, and finally forms 32-dimensional. Feature vector.

The present invention also provides a third embodiment as shown in FIG. 6, a device for acquiring a hyperspectral image feature descriptor, comprising:

a direction angle obtaining unit 601, configured to acquire a direction angle of each pixel in a preset neighborhood of the high-spectral image spatial domain domain interest point;

a main direction angle obtaining unit 602, configured to establish a direction angle statistical histogram according to a direction angle of each of the pixels, and acquire an angle corresponding to a peak in the direction angle statistical histogram;

The image rotation unit 603 is configured to calculate, according to the direction angle, an angle corresponding to a peak in the histogram as a main direction, and rotate the hyperspectral image according to the main direction to obtain a rotated hyperspectral image;

The descriptor obtaining unit 604 is configured to acquire a feature vector of the rotated hyperspectral image, and save the feature vector as a descriptor of the hyperspectral image.

Further, as shown in FIG. 7, the direction angle acquiring unit 601 includes:

The circular neighborhood determining module 6011 is configured to obtain a circular neighborhood of the point of interest by using the hyperspectral image spatial domain point of interest as a dot and a preset distance as a radius;

The square neighborhood determining module 6012 is configured to determine, by using each pixel in the circular neighborhood as a sampling point, a square neighborhood of the sampling point according to a preset side length;

The spectral angle calculation module 6013 is configured to calculate a spectral angle of each pixel in the square neighborhood and the sampling point, and determine a pixel with the largest spectral angle in the square neighborhood;

The angle calculation module 6014 is configured to calculate an angle between the pixel with the largest spectral angle in the square neighborhood and the sampling point, and use the included angle as the direction angle of the sampling point.

Further, as shown in FIG. 8, the description sub-acquisition unit 604 includes:

The image dividing module 6041 is configured to divide the rotated hyperspectral image into equal intervals into a plurality of sub-regions;

The vector obtaining module 6042 is configured to acquire feature vectors of each of the sub-regions, and perform splicing of feature vectors of the sub-regions to acquire feature vectors of the rotated hyperspectral image.

Further, the vector obtaining module 6042 includes:

The above is only the preferred embodiment of the present invention, and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection of the present invention. Within the scope.

Claims

A method for obtaining a hyperspectral image feature descriptor, wherein the obtaining method comprises:

Obtaining a direction angle of each pixel in a preset neighborhood of the hyperspectral image null spectrum domain of interest;

Generating a direction angle statistical histogram according to a direction angle of each of the pixels, and acquiring an angle corresponding to a peak in the direction angle statistical histogram;

Calculating, according to the direction angle, an angle corresponding to a peak in the histogram as a main direction, and rotating the hyperspectral image according to the main direction to obtain a rotated hyperspectral image;

Obtaining a feature vector of the rotated hyperspectral image, and saving the feature vector as a descriptor of the hyperspectral image.
The acquisition method according to claim 1, wherein the obtaining the orientation angle of each pixel in the preset neighborhood of the hyperspectral image spatial domain of interest points comprises:

Obtaining a circular neighborhood of the point of interest by using the hyperspectral image spatial spectral domain interest point as a dot and a preset distance as a radius;

Determining a square neighborhood of the sampling point according to a preset side length by using each pixel in the circular neighborhood as a sampling point;

Calculating a spectral angle of each pixel in the square neighborhood and the sampling point, and determining a pixel having the largest spectral angle in the square neighborhood;

Calculating an angle between the pixel having the largest spectral angle in the square neighborhood and the sampling point, and using the included angle as a direction angle of the sampling point.
The acquisition method according to claim 2, wherein the establishing a direction angle statistical histogram according to the direction angle of each of the pixels comprises:

The direction angle of the sampling point is the horizontal axis, and the amplitude accumulated value corresponding to the direction angle of the sampling point after the weight processing is the vertical axis, and the direction angle statistical histogram is established.
The acquiring method according to claim 1, wherein the acquiring the feature vector of the rotated hyperspectral image comprises:

Dividing the rotated hyperspectral image into equal intervals into a plurality of sub-regions;

Obtaining feature vectors of each of the sub-regions;

The feature vectors of the sub-regions are spliced to obtain feature vectors of the rotated hyperspectral image.
The acquiring method according to claim 4, wherein the acquiring feature vectors of each of the sub-regions comprises:

Calculating a direction angle of each pixel in the sub-area, and establishing a direction angle statistical histogram of the sub-area according to a direction angle of each pixel in the sub-area;

The component of the feature vector of the sub-region is determined by the direction angle of the sub-region, and the component of the feature vector of the sub-region is determined.
A device for acquiring a hyperspectral image feature descriptor, wherein the obtaining device comprises:

a direction angle acquiring unit, configured to acquire a direction angle of each pixel in a preset neighborhood of the high-spectrum image null spectrum domain interest point;

a main direction angle acquiring unit, configured to establish a direction angle statistical histogram according to a direction angle of each of the pixels, and obtain an angle corresponding to a peak in the direction angle statistical histogram;

An image rotation unit configured to calculate an angle corresponding to a peak in the histogram in the direction direction as a main direction, and rotate the hyperspectral image according to the main direction to obtain a rotated hyperspectral image;

a descriptor acquisition unit is configured to acquire a feature vector of the rotated hyperspectral image, and save the feature vector as a descriptor of the hyperspectral image.
The acquiring device according to claim 6, wherein the direction angle acquiring unit comprises:

a circular neighborhood determining module, configured to obtain a circular neighborhood of the point of interest by using the hyperspectral image spatial domain point of interest as a circle and using a preset distance as a radius;

a square neighborhood determining module, configured to determine, by using each pixel in the circular neighborhood as a sampling point, a square neighborhood of the sampling point according to a preset side length;

a spectral angle calculation module, configured to calculate a spectral angle of each pixel in the square neighborhood and the sampling point, and determine a pixel with the largest spectral angle in the square neighborhood;

An angle calculation module is configured to calculate an angle between a pixel having the largest spectral angle in the square neighborhood and the sampling point, and the angle is used as a direction angle of the sampling point.
The acquiring device according to claim 7, wherein the main direction angle obtaining unit is specifically configured to:

The direction angle of the sampling point is the horizontal axis, and the amplitude accumulated value corresponding to the direction angle of the sampling point after the weight processing is the vertical axis, and the direction angle statistical histogram is established.
The obtaining device according to claim 6, wherein the descriptor obtaining unit comprises:

An image dividing module, configured to divide the rotated hyperspectral image into several sub-regions at equal intervals;

And a vector acquiring module, configured to acquire a feature vector of each of the sub-regions, and splicing the feature vectors of the sub-regions to obtain a feature vector of the rotated hyperspectral image.
The obtaining device according to claim 9, wherein the vector obtaining module comprises:

a direction angle calculation sub-module, configured to calculate a direction angle of each picture element in the sub-area, and establish a direction angle statistical histogram of the sub-area according to a direction angle of each picture element in the sub-area;

The vector determining sub-module is configured to calculate, according to a direction angle of the sub-region, a component of a feature vector corresponding to each column in the histogram, and a feature vector of the sub-region.