WO2019047248A1

WO2019047248A1 - Feature extraction method and device for hyperspectral remotely sensed image

Info

Publication number: WO2019047248A1
Application number: PCT/CN2017/101353
Authority: WO
Inventors: 贾森; 吴奎霖; 朱家松; 邓琳
Original assignee: 深圳大学
Priority date: 2017-09-07
Filing date: 2017-09-12
Publication date: 2019-03-14
Also published as: CN107633216A; CN107633216B

Abstract

A feature extraction method for a hyperspectral remotely sensed image, comprising: performing normalization processing on an original image H, to obtain a normalized image R (S101); calculating, according to a preset gradient template, the gradient of the normalized image R (S102); encoding according to the gradient of the normalized image R, to obtain an encoded feature graph (S103); and performing a histogram feature extraction on the encoded feature graph, to obtain a three dimensional surface feature (3DSF) (S104). Further provided is a feature extraction device for a hyperspectral remotely sensed image. The above-mentioned manner can make full use of the three dimensional space-spectral structure of a hyperspectral remotely sensed image, improves classification performance, and reduces time complexity and space complexity.

Description

Feature extraction method and device for hyperspectral remote sensing image

This application claims the priority of the domestic patent application filed on September 7, 2017, the National Intellectual Property Office, the application number is 201710800249.4, and the invention name is "Feature Extraction Method and Apparatus for Hyperspectral Remote Sensing Images", the entire contents of which are incorporated by reference. In this application.

Technical field

The present application relates to the field of image processing, and in particular to a method and apparatus for classifying hyperspectral remote sensing images.

Background technique

Hyperspectral remote sensing image refers to the hyperspectral image data obtained by imaging the sensor in different wavelengths in the visible, near-infrared, mid-infrared and thermal infrared bands of the electromagnetic spectrum. Therefore, hyperspectral remote sensing images contain a wealth of spatial, radiation and spectral triple information, which provides a possibility for the fine classification and identification of surface materials.

The hyperspectral remote sensing image formed on hundreds of bands contains the triple information of radiation, space and spectrum, which makes the identification and classification of features more effective. In order to classify features through hyperspectral remote sensing images, feature extraction is first required. The existing feature extraction methods mainly include two categories: spatial feature extraction and spatial spectral feature extraction.

Spatial feature extraction mainly uses spatial information of different bands to express hyperspectral remote sensing images. It first extracts the spatial features of each band and then overlays the spatial features of the different bands. Gabor and Local Binary Pattern (LBP) are two typical spatial feature extraction methods. Gabor features are robust to illumination in images, and LBP can make full use of local spatial dependencies in images. In recent years, researchers have been inspired by Gabor features and LBP feature extraction methods, and proposed Gabor Surface Feature (GSF) extraction methods. Specifically, the Gabor amplitude features are first extracted from the two-dimensional image, and then the images are characterized by the first and second derivatives of these Gabor amplitude features. The advantage of GSF is that the features obtained by the Gabor filter are not affected by changes in image illumination conditions. However, for hyperspectral remote sensing images, there are two main disadvantages in the GSF extraction method. First, it only characterizes the spatial relationship of two-dimensional images, resulting in the three-dimensional spatial spectrum structure of hyperspectral remote sensing images not fully exploited. Second, hyperspectral The remote sensing image contains hundreds of bands. The multi-scale analysis and histogram feature representation of GSF greatly expands the feature dimension, which reduces the classification performance and increases the complexity of time and space.

Summary of the invention

The embodiment of the present application provides a feature extraction method and device for hyperspectral remote sensing images, which can fully utilize the three-dimensional spatial spectrum structure of hyperspectral remote sensing images, improve classification performance, and reduce time and space complexity.

In a first aspect, an embodiment of the present application provides a feature extraction method for a hyperspectral remote sensing image, where the method includes:

Normalizing the original image H to obtain a normalized image R;

Calculating a gradient of the normalized image R according to a preset gradient template;

Encoding according to the gradient of the normalized image R to obtain a coded feature map;

The histogram feature extraction is performed on the encoded feature map to obtain 3D surface surface features (3DSF).

In a second aspect, the embodiment of the present application provides a feature extraction device for a hyperspectral remote sensing image, where the device includes:

a normalization processing unit for normalizing the original image H to obtain a normalized image R;

a calculating unit, configured to calculate a gradient of the normalized image R according to a preset gradient template;

a coding unit, configured to perform coding according to the gradient of the normalized image R to obtain a coded feature map;

And a feature extraction unit, configured to perform histogram feature extraction on the encoded feature map to obtain a three-dimensional surface feature 3DSF.

It can be seen that, in the solution of the embodiment of the present application, first, the feature extraction device normalizes the original image to obtain a normalized image; secondly, the feature extraction device calculates a normalization according to the preset gradient template. The gradient value of the image; again, the feature extraction device binarizes the normalized image and the gradient thereof to obtain binarized data; finally, the feature extraction device encodes the binarized data and encodes the data according to the encoding The result is a 3DSF feature. By this method, the three-dimensional spatial spectrum structure of the hyperspectral remote sensing image can be fully utilized, the classification performance is improved, and the time and space are reduced. Miscellaneous.

DRAWINGS

In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings used in the embodiments or the prior art description will be briefly described below. Obviously, the drawings in the following description are only It is a certain embodiment of the present application, and other drawings can be obtained according to the drawings without any creative work for those skilled in the art.

FIG. 1 is a schematic flowchart of a method for extracting features of a hyperspectral remote sensing image according to an embodiment of the present application;

2 is a schematic flow chart of another feature extraction method for hyperspectral remote sensing images according to an embodiment of the present application;

Figure 3 is a schematic diagram of a three-dimensional coordinate system;

4a is a schematic diagram of pixel points of a band image;

Figure 4b is a schematic diagram of pixel points of the processed band image;

4c is a schematic diagram of pixel point gradient values of a band image;

FIG. 5 is a schematic structural diagram of a feature extraction apparatus for a hyperspectral remote sensing image according to an embodiment of the present disclosure;

FIG. 6 is a schematic structural diagram of another feature extraction apparatus for hyperspectral remote sensing images according to an embodiment of the present application.

Detailed ways

The embodiment of the present application provides a method and a device for extracting features of a hyperspectral remote sensing image, which can fully utilize the three-dimensional spatial spectrum structure of the hyperspectral remote sensing image, improve the classification performance, and reduce the complexity of time and space.

The technical solutions in the embodiments of the present application are clearly and completely described in the following with reference to the accompanying drawings in the embodiments of the present application. It is an embodiment of the present application, and not all of the embodiments. Based on the embodiments in the present application, those obtained by those of ordinary skill in the art without creative efforts Other embodiments are intended to fall within the scope of the present disclosure.

The terms "first", "second" and "third" and the like in the specification and claims of the present application and the above-mentioned drawings are used to distinguish different objects, and are not intended to describe a specific order. Moreover, the term "comprise" and any variants thereof are intended to cover a non-exclusive inclusion. For example, a process, method, system, product, or device that comprises a series of steps or units is not limited to the listed steps or units, but optionally also includes steps or units not listed, or, optionally, Other steps or units inherent to these processes, methods, products or equipment.

A method for extracting features of a hyperspectral remote sensing image provided by an embodiment of the present application includes:

Normalizing the original image H to obtain a normalized image R; calculating a gradient of the normalized image R according to a preset gradient template; encoding according to the gradient of the normalized image R, Coding a feature map; performing histogram feature extraction on the coded feature map to obtain a 3DSF.

Referring to FIG. 1, FIG. 1 is a schematic flowchart diagram of a first embodiment of a method for extracting features of a hyperspectral remote sensing image according to an embodiment of the present application. As shown in FIG. 1 , the method for extracting features of a hyperspectral remote sensing image provided by an embodiment of the present application includes the following steps:

S101: Normalize the original image H to obtain a normalized image R.

The normalizing the original image H to obtain the normalized image R includes:

Obtaining the mean and variance of the pixels of the original image H;

The original image H is normalized according to the mean and variance of the pixels of the original image H to obtain a normalized image R.

Specifically, the feature extraction device divides the original image H into M band images according to different bands in the spectral dimension. The feature extraction device calculates the mean and variance of the pixels of each of the M band images. Then, the feature extraction device normalizes each band image according to a preset formula to obtain a processed image. The above preset formula is:

Wherein H _b is any one of the M band images, the mean (H _b ) is a mean value of H _b , the std (H _b ) is a variance of H _b , and the R _b is a normalization of the H _b Process the resulting image.

Further, after normalizing each band image of the original image H, a normalized image R is obtained, and

Optionally, the normalization process may also be a maximum-minimum normalization process and a median normalization process.

S102. Calculate a gradient of the normalized image R according to a preset gradient template.

The gradient of the normalized image R includes gradients R ^x and R ^{y of} the normalized image R in the spatial dimension and a gradient R ^{b in the} spectral dimension.

Optionally, the preset gradient template may be [-1, 0, 1].

The calculating, by the feature extraction device, the gradient value of the normalized image R according to the preset gradient template refers to calculating a gradient value of the normalized image R in the spatial dimension and the spectral dimension. As shown in 3, the above three directions can be regarded as three axes of the space coordinate system. The two directions of the spatial dimension can be represented by X and Y, and the direction of the spectral dimension is represented by B.

Specifically, the process of calculating, by the feature extraction device, the gradient values in the three directions of X, Y, and B according to the preset gradient template [-1, 0, 1] to the image R is as follows:

Among them, the above

Is the gradient value of the pixel point of the coordinate (x, y, b) in the X direction in the normalized image R described above, and the coordinate of R _(x+1)yb in the normalized image R is (x) The pixel points of +1, y, b), R _{(x-1) yb} are the pixel points of the above normalized image R whose coordinates are (x-1, y, b).

Above

Is the gradient value of the pixel point of the coordinate (x, y, b) in the Y direction in the normalized image R described above, and the coordinates of R _x(y+1)b in the normalized image R described above are ( The pixel points of x, y+1, b), R _x(y-1)b are the pixel points of the above normalized image R whose coordinates are (x, y-1, b).

Above

Is the gradient value of the pixel point of the coordinate (x, y, b) in the B direction in the normalized image R described above, and the coordinate of R _xy(b+1) in the normalized image R is (x) The pixel of y, b+1) and R _xy(b-1) are the pixel points of the normalized image R whose coordinates are (x, y, b-1).

According to the above method, the feature extracting means calculates the normalized image R in three directions (gradient in the spatial dimension and gradient in the spectral dimension). [R ^x , R ^y , R ^b ] can be used to represent the gradient of the normalized image R in three directions,

Wherein R ^x and R ^y represent a gradient of the normalized image R in the spatial dimension, and R ^b is a gradient of the normalized image R in the spectral dimension.

It should be noted that, before the gradient of the pixel is calculated, the pixel value of the pixel of the image R that is not normalized is set to zero.

S103. Perform encoding according to the gradient of the normalized image R to obtain a coded feature map.

The encoding is performed according to the gradient of the normalized image R to obtain a coding feature map, including:

Calculating corresponding binarized data S, S ^x , S ^y and S ^b according to the normalized image R and the gradients R ^x and R ^{y in the} spatial dimension and the gradient R ^{b in the} spectral dimension;

Calculating the coded feature map according to S, S ^x , S ^y , S ^b and a preset formula, the preset formula is: coded feature map = 2 ³ S+2 ² S ^x + 2 ¹ S ^y + 2 ⁰ S ^b , where 2 ³ , 2 ² , 2 ¹ , and 2 ⁰ are weights of S, S ^x , S ^y , and S ^b , respectively.

Specifically, the above S is an image obtained by binarizing the normalized image R, and the S ^x and S ^y are the gradient R ^y of the normalized image R in the spatial dimension, and R ^x is binary. In the processed image, the above S ^b is an image obtained by binarizing the gradient R ^b of the normalized image in the spectral dimension.

Illustratively, the feature extraction device described above performs binarization processing on the normalized image R by the following formula. The formula is as follows:

Wherein the mean(R) is the mean of the normalized image R pixel values, and R _xyb is the normalized image R on the spatial coordinate axis as shown in FIG. 3 (x, y, z The pixel value of the pixel, the above S _xyb is the _binarized value of the above R _xyb . When R _{xyb is} greater than mean(R), S _xyb is equal to 1; when R _{xyb is} less than or equal to mean(R), S _xyb is equal to 0.

Similarly, the above feature extraction device performs binarization processing on the above R ^x by the following formula. The formula is as follows:

Wherein the mean (R ^x ) is the mean value of the R ^x pixel values,

Is the pixel value of the pixel of the above R ^x whose coordinates are (x, y, z) on the spatial coordinate axis as shown in FIG. 3,

For the above

The value of the binary. When above

When it is greater than mean(R ^x ),

Equal to 1; when

When less than or equal to mean(R ^x ),

Equal to 0.

Similarly, the above feature extraction device performs binarization processing on the above R ^y by the following formula. The formula is as follows:

Wherein the mean(R ^y ) is a mean value of the R pixel values, and R _xyb is a pixel value of the pixel of the R ^y at a coordinate of (x, y, z) on a spatial coordinate axis as shown in FIG. 3 . Above

For the above

The value of the binarization. When above

When it is greater than mean(R ^y ),

Equal to 1; when

When less than or equal to mean(R ^y ),

Equal to 0.

Similarly, the above feature extraction device performs binarization processing on the normalized image R ^b by the following formula. The formula is as follows:

Wherein the mean (R ^b ) is the mean value of the R ^b pixel values,

The pixel value of the pixel of the above R ^b having the coordinate of (x, y, z) on the spatial coordinate axis as shown in FIG. 3,

For the above

The value of the binarization. When above

When it is greater than mean(R ^b ),

Equal to 1; when

When less than or equal to mean(R ^b ),

Equal to 0.

The coded value of the above-mentioned coded feature map Z represents the spatial spectrum structure characteristic around the pixel. The above fusion rule is that the feature extraction means assigns a weight to the above S, S ^x , S ^y and S ^b according to the saliency of the feature.

Specifically, the weight of the S is the first weight A1, the weight of the S ^x is the second weight A2, the weight of the S ^y is the third weight A3, and the weight of the S ^b is the fourth value. The weight A4, because the original image (ie, the original hyperspectral remote sensing image) contains rich distribution information of the feature structure, and the spatial characteristics of the feature distribution are more obvious than the spectral features, therefore, A1>A2>A3>A4. The above joint feature map Z=A1S+A2S ^x +A3S ^y +A4S ^b

For example, assume that the weights of S, S ^x , S ^y and S ^b are set to 2 ³ , 2 ² , 2 ¹ and 2 ^{0 respectively} , and the specific coding method is expressed as:

Z=2 ³ S+2 ² S ^x +2 ¹ S ^y +2 ⁰ S ^b

Wherein, the above-mentioned coding feature map Z fuses the original hyperspectral remote sensing image (ie, the above-mentioned original image R) with a stepwise amplitude (ie, S ^x , S ^y and S ^b described above).

S104. Perform histogram feature extraction on the coding feature map to obtain a three-dimensional surface feature 3DSF.

Wherein, the above histogram feature is statistically obtained from a cubic neighborhood V around each pixel. According to the calculation formula of Z in the above step S204, it can be seen that the values of S, S ^x , S ^y and S ^b are both 0 or 1, and it can be seen that the coding feature map Z can have 16 different coding values, respectively 0, 1. 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, so the histogram feature around each pixel can be 16 dimensions. assumed

To represent the histogram features of pixels with coordinates (x, y, b), then the statistical formula is:

Wherein, the above V _x , V _y , V _b respectively represent the spatial dimension and the spectral dimension of the cubic neighborhood,

The operator indicates that the data is rounded down, and h(i) represents the number of times the 3DSF code i appears in the cube neighborhood. When Z(x+j, y+k, b+l)=i, the above h(i) is equal to 1; when Z(x+j, y+k, b+l)≠i, the above h(i) ) is equal to 0.

Further, the feature extraction device acquires the number of times each of the above codes appears in the cube field, and obtains 16 numbers (ie, h(0), h(1), h(2), ..., h(14), h. (15)). The feature extraction device combines the above-mentioned 16 numbers into an array, and the number of the groups can be regarded as the above-mentioned 3DSF feature.

In other words, the histogram feature around each pixel is 16 dimensions, and the original image H is X*Y*B, which can be seen as 16 (X*Y*B) cubes stacked together, so it can be expressed as X* Y*(16*B). For the original image H, the resulting 3DSF feature is

This feature F can be used directly for subsequent pixel level classification.

It can be seen that, in the solution of the embodiment of the present application, first, the feature extraction device normalizes the original image to obtain a normalized image; secondly, the feature extraction device calculates a normalization according to the preset gradient template. The gradient value of the image; again, the feature extraction device binarizes the normalized image and the gradient thereof to obtain binarized data; finally, the feature extraction device encodes the binarized data and encodes the data according to the encoding The result is a 3DSF feature. Through this method, it can be fully profitable The three-dimensional spatial spectrum structure of hyperspectral remote sensing images improves classification performance and reduces time and space complexity.

Referring to FIG. 2, FIG. 2 is a schematic flowchart diagram of another feature extraction method for hyperspectral remote sensing images according to an embodiment of the present application. As shown in Figure 2, the method includes:

S201. The feature extraction device normalizes the image of each band of the original image H to obtain a normalized image R.

Wherein, the above original image

Represents real numbers, X, Y, and B represent the number of spatial and spectral dimensions of the original image, respectively.

It should be noted that the above original image H is a hyperspectral remote sensing image.

S202. The feature extraction device calculates a gradient value of the normalized image R according to the preset gradient template.

Optionally, the preset gradient template may be [-1, 0, 1].

Specifically, the feature extraction device calculates the image R according to the preset gradient template [-1, 0, 1]. The process of gradient values in the three directions of X, Y, and B is as follows:

Among them, the above

The gradient value of the pixel point of the coordinate (x, y, b) in the above normalized image R in the X direction, R _{(x+1) yb} is the coordinate in the normalized image R (x) The pixel points of +1, y, b) and R _{(x-1) yb} are the pixel points of the above normalized image R whose coordinates are (x-1, y, b).

Above

The gradient value in the Y direction of the pixel whose coordinates are (x, y, b) in the normalized image R, and the coordinates of R _x(y+1)b in the normalized image R are ( The pixel points of x, y+1, b) and R _x(y-1)b are the pixel points of the above normalized image R whose coordinates are (x, y-1, b).

Above

The gradient value in the B direction of the pixel whose coordinates are (x, y, b ₎ in the normalized image R, R _{xy (b+1)} is the coordinate in the normalized image R (x) The pixel of y, b+1) and R _xy(b-1) are the pixel points of the normalized image R whose coordinates are (x, y, b-1).

For example, assume that a band image in the normalized image R is as shown in FIG. 4a, and the resolution of the band image is 4*4, and each pixel value is as shown in FIG. 4a. Before calculating the gradient value of the pixel point of the band image, the feature extraction means sets the pixel value of the pixel point of the non-band image to 0, as shown in FIG. 4b. Then, based on the gradient template [-1, 0, 1], the gradient value of each pixel is calculated. For example, if the pixel value P ₁₁ =1 of the first pixel of the first row, the gradient value of the pixel point=P ₁₂ —0=2; the pixel value of the third pixel of the second navigation P ₂₃ =0, The gradient value of the pixel = P _{24 -} P ₂₂ = 2-1 = 1. And so on, the above feature extraction device can calculate the pixel gradient of the band image, as shown in FIG. 4c. 4a is a schematic diagram of pixel points of a band image; FIG. 4b is a schematic diagram of pixel points of the processed band image; FIG. 4c is a schematic diagram of pixel point gradient values of the band image.

S203. The feature extraction device performs binarization processing on the normalized image and the gradient to obtain binarized data.

Specifically, the feature extraction device binarizes the normalized image R and the gradients R ^x , R ^y and R ^{b of} the image in three directions to obtain a binarized image S, S ^x , S ^y and S ^b .

Wherein the S image after binarizing processing is above the normalized image R, the above-described S ^X and S ^Y to above the normalized image R gradient R ^Y in the spatial dimension, R ^X is binarized In the processed image, the above S ^b is an image obtained by binarizing the gradient R ^b of the normalized image in the spectral dimension.

Wherein the mean (R ^x ) is the mean value of the R ^x pixel values,

For the above

The value of the binarization. When above

When it is greater than mean(R ^x ),

Equal to 1; when

When less than or equal to mean(R ^x ),

Equal to 0.

Wherein the mean(R _y ) is a mean value of the R pixel values, and R _xyb is a pixel value of the pixel of the R _y at a coordinate of (x, y, z) on a spatial coordinate axis as shown in FIG. 3 . Above

For the above

The value of the binarization. When above

When it is greater than mean(R ^y ),

Equal to 1; when

When less than or equal to mean(R ^y ),

Equal to 0.

Wherein the mean (R ^b ) is the mean value of the R ^b pixel values,

For the above

The value of the binarization. When above

When it is greater than mean(R ^b ),

Equal to 1; when

When less than or equal to mean(R ^b ),

Equal to 0.

S204. The feature extraction device jointly encodes S, S ^x , S ^{y ,} and S ^b according to a preset fusion rule to obtain a coded feature map.

Z=2 ³ S+2 ² S ^x +2 ¹ S ^y +2 ⁰ S ^b

S205. The feature extraction device performs histogram feature extraction on the encoded feature map to obtain a 3DSF feature.

In other words, the histogram feature around each pixel is 16 dimensions, and the original image H is X*Y*B, which can be seen as 16 (X*Y*B) cubes stacked together, so it can be expressed as X* Y*(16*B). For the above original image H, the resulting 3DSF feature is

This feature F can be used directly for subsequent pixel level classification.

It can be seen that, in the solution of the embodiment of the present application, first, the feature extraction device normalizes the original image to obtain a normalized image; secondly, the feature extraction device calculates a normalization according to the preset gradient template. The gradient value of the image; again, the feature extraction device binarizes the normalized image and the gradient thereof to obtain binarized data; finally, the feature extraction device pairs the binary value The data is encoded and the 3DSF features are acquired based on the encoded results. By this method, the three-dimensional spatial spectrum structure of the hyperspectral remote sensing image can be fully utilized, the classification performance is improved, and the complexity of time and space is reduced.

Referring to FIG. 5, FIG. 5 is a schematic structural diagram of a device for extracting features of a hyperspectral remote sensing image according to an embodiment of the present application. As shown in FIG. 5, the apparatus 500 includes:

The normalization processing unit 501 is configured to perform normalization processing on the original image H to obtain a normalized image R.

The normalization processing unit 501 includes

The obtaining subunit 5011 is configured to obtain the mean and variance of the pixels of the original image H;

The processing sub-unit 5012 is configured to perform normalization processing on the original image H according to the mean and variance of the pixels of the original image H to obtain a normalized image R.

The calculating unit 502 is configured to calculate a gradient of the normalized image R according to the preset gradient template.

The coding unit 503 is configured to perform coding according to the gradient of the normalized image R to obtain a coded feature map.

The coding unit 503 includes:

a first calculating subunit 5031, configured to calculate corresponding binarized data S, S according to the normalized image R and the gradients R ^x and R ^{y in the} spatial dimension and the gradient R ^{b in the} spectral dimension ^x , S ^y and S ^b ;

a second calculating subunit 5032, configured to calculate the encoded feature map according to S, S ^x , S ^y , S ^b and a preset formula, where the preset formula is:

Coding feature map = 2 ³ S+2 ² S ^x +2 ¹ S ^y +2 ⁰ S ^b ,

Among them, 2 ³ , 2 ² , 2 ¹ , and 2 ⁰ are the weights of S, S ^x , S ^y , and S ^b , respectively.

The feature extraction unit 504 is configured to perform histogram feature extraction on the encoded feature map to obtain a three-dimensional surface feature 3DSF.

The feature extraction unit 504 is configured to:

Obtaining, according to a statistical formula, the number of occurrences of the encoded value i composed of the S, S ^x , S ^y , and S ^b in a cubic neighborhood around each pixel;

The statistical formula is:

Wherein said

a histogram feature representing pixels of coordinates (x, y, b), the V _x , V _y , V _b representing the spatial and spectral dimensions of the cube neighborhood, respectively,

The operator indicates that the data is rounded down, h(i) represents the number of times the encoded value i appears in the cube neighborhood, and the array consisting of the number of times is the 3DSF.

In the present embodiment, the hyperspectral remote sensing image feature extraction device 500 is presented in the form of a unit (normalization processing unit 501, calculation unit 502, encoding unit 503, and feature extraction unit 504). A "unit" herein may refer to an application-specific integrated circuit (ASIC), a processor and memory that executes one or more software or firmware programs, integrated logic circuits, and/or other devices that provide the functionality described above. .

It can be understood that the functions of the functional units of the hyperspectral remote sensing image feature extraction device 500 of the present embodiment can be specifically implemented according to the method in the foregoing method embodiments. For the specific implementation process, reference may be made to the related description of the foregoing method embodiments. I won't go into details here.

Referring to FIG. 6, FIG. 6 is a schematic structural diagram of a hyperspectral remote sensing image feature extraction apparatus according to an embodiment of the present application, which is used to implement the feature extraction method of hyperspectral remote sensing image disclosed in the embodiment of the present application. The hyperspectral remote sensing image feature extraction device 600 may include at least one bus 601, at least one processor 602 connected to the bus 601, and at least one memory 603 connected to the bus 601.

The processor 602 calls the code stored in the memory through the bus 601 for normalizing the original image H to obtain a normalized image R; and calculating the normalized image R according to the preset gradient template. a gradient; encoding according to the gradient of the normalized image R to obtain a coded feature map; performing histogram feature extraction on the coded feature map to obtain a three-dimensional surface feature 3DSF.

In the present embodiment, the hyperspectral remote sensing image feature extraction device 600 is presented in the form of a unit. The "unit" herein may refer to an application-specific integrated application (application-specific integrated Circuit, ASIC), a processor and memory that executes one or more software or firmware programs, integrated logic circuits, and/or other devices that provide the functions described above.

It can be understood that the functions of the functional units of the hyperspectral remote sensing image feature extraction device 600 of the present embodiment can be specifically implemented according to the method in the foregoing method embodiments. For the specific implementation process, reference may be made to the related description of the foregoing method embodiments. I won't go into details here.

The embodiment of the present application further provides a computer storage medium, wherein the computer storage medium may store a program, where the program includes some or all of the steps of the hyperspectral remote sensing image feature extraction method described in the foregoing method embodiments. .

It should be noted that, for the foregoing method embodiments, for the sake of simple description, they are all expressed as a series of action combinations, but those skilled in the art should understand that the present application is not limited by the described action sequence. Because certain steps may be performed in other sequences or concurrently in accordance with the present application. In the following, those skilled in the art should also understand that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present application.

In the above embodiments, the descriptions of the various embodiments are different, and the details that are not detailed in a certain embodiment can be referred to the related descriptions of other embodiments.

In the several embodiments provided herein, it should be understood that the disclosed apparatus may be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the unit is only a logical function division. In actual implementation, there may be another division manner, for example, multiple units or components may be combined or may be Integrate into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed may be an indirect coupling or communication connection through some interface, device or unit, and may be electrical or otherwise.

The units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, may be located in one place, or may be distributed to multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of the embodiment.

In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above integrated unit can be implemented in the form of hardware or a software function list. The realization of the form of the yuan.

The integrated unit, if implemented in the form of a software functional unit and sold or used as a standalone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present application, in essence or the contribution to the prior art, or all or part of the technical solution may be embodied in the form of a software product stored in a storage medium. A number of instructions are included to cause a computer device (which may be a personal computer, server or network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present application. The foregoing storage medium includes: a U disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a removable hard disk, a magnetic disk, or an optical disk, and the like. .

The above embodiments are only used to explain the technical solutions of the present application, and are not limited thereto; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still The technical solutions described in the embodiments are modified, or equivalent to some of the technical features are replaced; and the modifications or substitutions do not deviate from the scope of the technical solutions of the embodiments of the present application.

Claims

A feature extraction method for a hyperspectral remote sensing image, characterized in that the method comprises:

Normalizing the original image H to obtain a normalized image R;

Calculating a gradient of the normalized image R according to a preset gradient template;

Encoding according to the gradient of the normalized image R to obtain a coded feature map;

Histogram feature extraction is performed on the coded feature map to obtain a three-dimensional surface feature 3DSF.
The method according to claim 1, wherein the normalizing the original image H to obtain a normalized image R comprises:

Obtaining the mean and variance of the pixels of the original image H;

The original image H is normalized according to the mean and variance of the pixels of the original image H to obtain a normalized image R.
The method according to claim 1, characterized in that the gradient of the normalized image R comprises gradients R x and R y of the normalized image R in the spatial dimension and a gradient R in the spectral dimension b .
The method according to claim 3, wherein the encoding according to the gradient of the normalized image R to obtain a coded feature map comprises:

Calculating corresponding binarized data S, S x , S y and S b according to the normalized image R and the gradients R x and R y in the spatial dimension and the gradient R b in the spectral dimension;

Calculating the coded feature map according to S, S x , S y , S b and a preset formula, the preset formula is:

Coding feature map = 2 3 S+2 2 S x +2 1 S y +2 0 S b ,

Among them, 2 3 , 2 2 , 2 1 , and 2 0 are the weights of S, S x , S y , and S b , respectively.
The method according to claim 4, wherein the performing a histogram feature extraction on the encoded feature map to obtain a three-dimensional surface feature 3DSF comprises:

Obtaining, according to a statistical formula, the number of occurrences of the encoded value i composed of the S, S x , S y , and S b in a cubic neighborhood around each pixel;

The statistical formula is:

Wherein said
a histogram feature representing pixels of coordinates (x, y, b), the V x , V y , V b representing the spatial and spectral dimensions of the cube neighborhood, respectively,
The operator indicates that the data is rounded down, h(i) represents the number of times the encoded value i appears in the cube neighborhood, and the array consisting of the number of times is the 3DSF.
A device for extracting features of a hyperspectral remote sensing image, comprising:

a normalization processing unit for normalizing the original image H to obtain a normalized image R;

a calculating unit, configured to calculate a gradient of the normalized image R according to a preset gradient template;

a coding unit, configured to perform coding according to the gradient of the normalized image R to obtain a coded feature map;

And a feature extraction unit, configured to perform histogram feature extraction on the encoded feature map to obtain a three-dimensional surface feature 3DSF.
The apparatus of claim 6 wherein said normalization processing unit comprises

Obtaining a subunit for obtaining a mean and a variance of pixels of the original image H;

Processing subunits for normalizing the original image H according to the mean and variance of the pixels of the original image H to obtain a normalized image R.
The apparatus according to claim 7, wherein the gradient of the normalized image R comprises gradients R x and R y of the normalized image R in the spatial dimension and a gradient R in the spectral dimension b .
The apparatus according to claim 8, wherein said coding unit comprises:

a first calculating subunit, configured to calculate corresponding binarized data S, S x according to the normalized image R and the gradients R x and R y in the spatial dimension and the gradient R b in the spectral dimension , S y and S b ;

a second calculating subunit, configured to calculate the encoded feature map according to S, S x , S y , S b and a preset formula, where the preset formula is:

Coding feature map = 2 3 S+2 2 S x +2 1 S y +2 0 S b ,

Among them, 2 3 , 2 2 , 2 1 , and 2 0 are the weights of S, S x , S y , and S b , respectively.
The apparatus according to claim 9, wherein said feature extraction unit is configured to:

Obtaining, according to a statistical formula, the number of occurrences of the encoded value i composed of the S, S x , S y , and S b in a cubic neighborhood around each pixel;

The statistical formula is:

Wherein said
a histogram feature representing pixels of coordinates (x, y, b), the V x , V y , V b representing the spatial and spectral dimensions of the cube neighborhood, respectively,
The operator indicates that the data is rounded down, h(i) represents the number of times the encoded value i appears in the cube neighborhood, and the array consisting of the number of times is the 3DSF.