A hyperspectral image radiation bit depth residual quantization method

Abstract

The application provides a hyperspectral image radiation bit depth residual quantization method, which mainly comprises the following steps: image outlier removal, radiation bit depth residual quantization coefficient determination, bit depth feature image and residual image generation, etc. Image outlier removal: import the hyperspectral image to be processed, detect and remove outliers. Radiation bit depth residual quantization coefficient determination: calculate the maximum radiation value of the hyperspectral image, determine the radiation bit depth residual quantization level of the hyperspectral image, and then determine the quantization coefficient. Bit depth feature image and residual image generation: obtain the hyperspectral bit depth feature image and its residual image with different radiation bit depths. The application is suitable for quantization of radiation bit depth residuals of various hyperspectral images.

Description

Technical Field

[0001] This invention relates to a method for quantizing the radiometric depth residual of hyperspectral images, which belongs to the field of remote sensing and is applicable to any hyperspectral image. Background Technology

[0002] Improving image classification accuracy and highlighting weak information are goals pursued by researchers. Radiative potential depth residual quantization (RBD) segments the high-resolution radiometric information of hyperspectral images to construct potential depth feature images and residual images at different radiometric potential depths, thereby deeply mining radiometric information and improving land cover classification accuracy. Although a few researchers have studied the impact of hyperspectral image radiometric resolution on land cover classification, these studies only superficially analyze the influence of different radiometric resolutions on land cover classification, failing to fully explore the high-resolution radiometric information and neglecting RBD residual quantization of hyperspectral images. Therefore, existing techniques have shortcomings and limitations, failing to fully utilize the radiometric information of hyperspectral images and failing to fully explore the impact of different radiometric potential depths on land cover classification. Summary of the Invention

[0003] The technical problem to be solved by this invention is to overcome the shortcomings of the prior art and provide a method for quantizing the radiometric potential depth residual of hyperspectral images. Based on the radiometric resolution of the image, the radiometric potential depth quantization level is determined, and potential depth feature images and residual images with different radiometric potential depths are constructed. The potential depth feature images reduce data redundancy while ensuring classification accuracy; the residual images increase image details and highlight the characteristics of certain land features without reducing classification accuracy. This method is suitable for quantifying the radiometric potential depth residual of various types of hyperspectral images.

[0004] The technical solution adopted in this invention is a method for quantizing the radiometric depth residual of hyperspectral images, comprising the following steps:

[0005] Step (1): Detect outliers in each band of the hyperspectral image and replace the outliers with the neighborhood mean to obtain the replaced hyperspectral image;

[0006] Step (2): Calculate the maximum radiometric value of the replaced hyperspectral image, determine the radiometric depth residual quantization level of the hyperspectral image, and finally obtain the radiometric depth residual quantization coefficient;

[0007] Step (3): Based on the radiation depth residual quantization coefficient obtained in step (2), perform residual quantization on the radiation value of each pixel in the hyperspectral image to generate depth feature images and residual images with different radiation depths.

[0008] Furthermore, in step (1), outlier detection is performed on each band, and outliers are replaced with the neighborhood mean, which includes:

[0009] Step (11): Perform band filtering for pixels with radiation values ​​less than or equal to 0, using rows and columns as units;

[0010] Step (12): Replace the pixel radiation value of the selected band in step (11) with the pixel radiation mean value of all points in the neighborhood N to obtain the replaced hyperspectral image.

[0011] Furthermore, the steps for determining the quantization coefficient of the firing depth residual in step (2) are as follows:

[0012] Step (21): Calculate the maximum radiance of the replaced hyperspectral image;

[0013] Step (22): Determine the range to which the maximum radiation value belongs, and then determine the quantization level of the radiation potential depth residual;

[0014] Step (23): Finally determine the quantization coefficient of radiation depth residual.

[0015] Furthermore, in step (3), residual quantization is performed on the radiometric value of each pixel in the hyperspectral image to generate depth feature images and residual images with different radiometric depths, including:

[0016] Step (31): Determine the equation for calculating the pixel radiation value of the bit depth feature image by using the calculated radiation bit depth residual quantization coefficient, and generate the radiation bit depth feature image;

[0017] Step (32): Using the calculated radiation depth residual quantization coefficient and the depth feature image obtained in step (31), determine the equation for calculating the pixel radiation value of the residual image and generate the residual image.

[0018] The advantages of this invention compared to the prior art are:

[0019] (1) In terms of the application of radiometric dimension information in hyperspectral images, traditional methods are mostly limited to multispectral images and rarely utilize the radiometric information of hyperspectral images. This invention successfully applies the radiometric depth residual quantization method to hyperspectral images.

[0020] (2) Existing technologies only analyze the impact of different radiometric resolutions on land cover classification on the surface, without fully exploring the radiometric information of high resolution. This invention generates depth feature images and residual images of different radiometric depths through the hyperspectral image radiometric depth residual quantization method, and deeply explores the radiometric dimension information of hyperspectral images.

[0021] (3) The depth feature image obtained by this invention not only solves the problem of large data volume and high redundancy of hyperspectral images, but also ensures the classification accuracy of the images; the obtained residual image highlights the detailed information of the hyperspectral image, and at the same time highlights the classification accuracy of a certain land feature.

[0022] (4) The algorithm proposed in this invention has low algorithm complexity, and the low algorithm complexity also requires less system hardware level. Attached Figure Description

[0023] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0024] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the protection scope of the present invention.

[0025] like Figure 1 As shown, the method for quantizing radiation potential depth residuals provided by this invention includes the following main steps:

[0026] Step 1: Replace outliers using neighborhood mean

[0027] In a hyperspectral image, each pixel is p(i,j,k). The radiation value f(i,j,k) corresponding to pixel p(i,j,k) is replaced with the mean radiation value of all pixels in the neighborhood N, thereby removing pixels in each band of the image whose radiation value is less than or equal to 0.

[0028]

[0029] In the formula, i, j, k are the row number, column number, and band number of the hyperspectral image, respectively.

[0030] Step 2: Determine the quantization coefficient of radiation depth residual based on the maximum radiation value in the image.

[0031] This invention begins by determining the maximum value of pixel radiance in a hyperspectral image, then determines the range to which the maximum radiance value belongs, and finally determines the radiance bit depth residual quantization level.

[0032] The maximum radiance value corresponding to a pixel in a hyperspectral image is represented as f(i,j,k). max Determine the quantization level of radiation depth features:

[0033] 2 m ≤f(i,j,k) max ≤2 m+1 (2)

[0034] In the formula, m is the quantization level of the radiation potential depth residual. The quantization coefficient of the radiation potential depth residual can be expressed as:

[0035]

[0036] In the formula, m is the pixel X i The radiation depth, n represents the specific residual quantization series, β n This represents the quantization coefficient of the radiation potential depth residual.

[0037] Step 3: Using the radiation depth residual quantization coefficients, obtain the depth feature images and residual images for different radiation depths. Assume the input image X = {X1, X2, X3, ... X...} i ,…X N}(where X1, X2, X3, ... X i ,…X N The radiance value of each pixel in image X is a non-negative integer value represented in mbits, i.e., 0 ≤ X. i ≤2 m -1, firstly, then each pixel X can be... i Quantified as

[0038] X i =β n H i +R i (X) i=1,2,3,…,N (4)

[0039] Where N represents the number of pixels in the hyperspectral image, H i Represents pixel X i Bit depth features, R i (X) represents pixel X i The residual can then be used to quantize pixel X using formula (5). i To the corresponding radiation depth H i Then, round it to the nearest integer and generate the radiation bit depth feature image H = {H1, H2, H3, ... H} for i bits. i …,H N}

[0040]

[0041] H was calculated i Next, the pixel X is quantized using formula (6). i residual R i (X), and generate an n-bit residual image R = {R1, R2, R3, ... R i …,R N}

[0042] R i (X)=X i -β n H i i = 1, 2, 3, ..., N (6)

[0043] Table 1: Advantages of the present invention compared with the prior art

[0044]

[0045] The above description is merely an embodiment of a radiation potential depth residual quantization method according to the present invention. The present invention is not limited to the above embodiments. This specification is for illustrative purposes only and does not limit the scope of the claims. It will be apparent to those skilled in the art that many substitutions, improvements, and variations are possible. All technical solutions formed by equivalent substitutions or equivalent transformations fall within the protection scope claimed by the present invention.

Claims

1. A method for quantizing the radiometric depth residual of a hyperspectral image, comprising the following steps: Step (1): Detect outliers in each band of the hyperspectral image and replace the outliers with the neighborhood mean to obtain the replaced hyperspectral image; Step (2): Calculate the maximum radiometric value of the replaced hyperspectral image, determine the radiometric depth residual quantization level of the hyperspectral image, and finally obtain the radiometric depth residual quantization coefficient; Step (3): Based on the radiation depth residual quantization coefficients obtained in step (2), perform residual quantization on the radiation value of each pixel in the hyperspectral image to generate depth feature images and residual images with different radiation depths. The steps for determining the quantization coefficients of the radiation potential depth residual in step (2) are as follows: Step (21): Calculate the maximum radiance of the replaced hyperspectral image; Step (22): Determine the range to which the maximum radiation value belongs, and then determine the quantization level of the radiation potential depth residual; Step (23): Finalize the quantization coefficient of the radiation potential depth residual; In step (3), residual quantization is performed on the radiance value of each pixel in the hyperspectral image to generate depth feature images and residual images with different radiance depths, including: Step (31): Determine the equation for calculating the pixel radiation value of the bit depth feature image by using the calculated radiation bit depth residual quantization coefficient, and generate the bit depth feature image; Step (32): Using the calculated radiation depth residual quantization coefficient and the depth feature image obtained in step (31), determine the equation for calculating the pixel radiation value of the residual image and generate the residual image.

2. The hyperspectral image radiation depth residual quantization method according to claim 1, characterized in that: The outlier detection and replacement of outliers with neighborhood mean in step (1) includes: Step (11): Perform band filtering on pixels with radiation values ​​less than or equal to 0, using rows and columns as units; Step (12): Replace the pixel radiation value of the selected band in step (11) with the pixel radiation mean value of all points in the neighborhood N to obtain the replaced hyperspectral image.

Images