Scene-based push-broom hyperspectral image adaptive non-uniformity correction method

Abstract

The invention discloses a scene-based push-broom hyperspectral image adaptive non-uniformity correction method, which comprises the following steps of: automatically sorting each scanning line of a wave band of a hyperspectral image from small to large according to pixel DN values of different spatial dimensions, and adaptively selecting at least one dark uniform region and one bright uniform region of the spatial dimensions; a correction coefficient of each spatial dimension pixel is obtained, and non-uniform correction of each spatial dimension is completed; on the basis of spectral dimension characteristics, spectral dimension non-uniform correction is completed according to the spatial dimension non-uniform correction result, spatial dimension non-uniform correction is repeated according to the spectral dimension non-uniform correction result, and hyperspectral image non-uniform correction is completed. The method is applied in a full-automatic mode, limitation of scenes can be broken through, space and spectrum combined information is used for non-uniform correction, space structure information and spectrum information of the hyperspectral images are reserved to the maximum extent, the quality of the hyperspectral images can be effectively improved, and the method is suitable for self-adaptive non-uniform correction of mass hyperspectral images of various complex scenes. And the method has good universality.

Description

Technical Field

[0001] The invention relates to the technical field of hyperspectral remote sensing imaging, and in particular to a scene-based push-broom hyperspectral image adaptive non-uniformity correction method. Background Art

[0002] Hyperspectral imagers can simultaneously collect the geometric, radiation and spectral information of the target to form an image cube. While imaging the target, hundreds or thousands of continuous and detailed radiation intensity data of different wavelengths of the target are obtained within a wide spectrum range to form the spectral characteristic curve "fingerprint" of the target, which can effectively identify the target. Therefore, it has been widely used in energy resource exploration, environmental disaster monitoring, agricultural and forestry remote sensing, greenhouse gas monitoring, urban investigation, etc., with great scientific significance and application value. In the existing technology, the imaging methods of hyperspectral imagers are mainly divided into three categories: swing-scan imaging, push-scan imaging and stare imaging. At present, most satellite-borne hyperspectral imagers use push-scan scanning imaging, such as Hyperion, DESIS, AHSI, EnMAP, etc. Take the imaging process of the AHSI payload of the Gaofen-5 (GF-5) hyperspectral satellite as an example: the push-broom hyperspectral camera uses an array detector to acquire images. One dimension of the array detector is the spatial dimension (the number of array elements is W), which images the objects in the through-track direction; the other dimension is the spectral dimension (the number of array elements is C), which corresponds to the spectral information of a row of objects in the through-track direction. As the satellite flies H rows along the orbit, the acquired hyperspectral image is a three-dimensional cube of H×W×C.

[0003] However, since the number of detector pixels reaches millions, it is difficult to make the photoelectric response of each pixel completely consistent, especially after the wide spectral range band is subdivided, the response non-uniformity by wavelength will be more significant. This non-uniformity is specifically manifested as a stripe distribution along the push-sweep direction on the push-broom hyperspectral image, which has a great impact on the hyperspectral feature extraction and quantitative application of ground objects. Non-uniformity correction has become a major challenging task for remote sensing image data, especially three-dimensional cube hyperspectral image data.

[0004] The commonly used non-uniform correction methods in the prior art can be roughly divided into two categories: one is the calibration-based method, which can achieve very high accuracy, but because the state of the satellite changes during continuous operation, it is necessary to regularly correct the calibration coefficient on the satellite; the other is the scene-based correction method, which directly updates the non-uniform correction coefficient from the scene. This method often requires certain assumptions about the image. If the scene meets the assumed conditions, the correction effect is good. The scene-based method is easier to implement because it does not require special hardware design, but its correction effect is heavily dependent on the applicability of the method to the scene.

[0005] Traditional scene-based hyperspectral image non-uniformity correction methods mainly use the two-point method. For example, Liu Yinian, Wang Yueming and others published "Research on Two-point Multi-segment Non-uniformity Correction Algorithm of Infrared Focal Plane Devices", which proposed a two-point multi-segment non-uniformity correction method. However, this method relies on manual interpretation to find uniform bright and dark areas in the image, and the area needs to cover the pixels of the entire spatial dimension (or select several small uniform areas to cover the entire spatial dimension). It is not applicable to complex scenes with many ground details and small and scattered uniform areas. In addition, patent ZL201610913590.6 "A scene-based adaptive non-uniform correction method for images" proposes to automatically sort the DN values ​​of the image according to different pixels, determine at least one relatively dark uniform area and bright uniform area respectively, calculate the correction coefficient of each pixel according to the algorithm, and complete the non-uniform correction of the image. It can adapt to the non-uniform correction of a large number of images of various complex ground scenes, and can solve the problem of inconsistent uniform characteristics of each spectral band of the image, and can effectively improve the imaging quality and clarity of the image. However, although this method can adapt to the non-uniform correction of a large number of images of various complex ground scenes during application, it needs to manually debug the preset parameters before application, and it is difficult to achieve true automation and adaptive correction. In addition, this method mainly completes non-uniform correction based on the spatial information of hyperspectral images, and the distortion correction of spectral information of hyperspectral images is not fully considered. Summary of the invention

[0006] In view of the above problems, the present invention proposes a scene-based push-broom hyperspectral image adaptive non-uniformity correction method. The method does not require manual debugging of preset parameters and is fully automatic. It can break through the limitations of the scene and use spatial spectral joint information for non-uniform correction. It can retain the spatial structure information and spectral information of the hyperspectral image to the greatest extent, effectively improve the quality of the hyperspectral image, and is suitable for adaptive non-uniform correction of massive hyperspectral images in complex scenes. It has good versatility.

[0007] For this purpose, the present invention adopts the following technical solutions:

[0008] A scene-based push-broom hyperspectral image adaptive non-uniformity correction method is provided. Each scanning line of a band of a hyperspectral image is automatically sorted from small to large according to the DN values ​​of pixels in different spatial dimensions, and at least one dark uniform area and a bright uniform area in the spatial dimension are adaptively selected to obtain the correction coefficient of each spatial dimension pixel, so as to complete the non-uniformity correction of each spatial dimension; based on the spectral dimension characteristics, the spectral dimension non-uniformity correction is completed for the non-uniformity correction result of the spatial dimension, and the non-uniformity correction of the spatial dimension is repeated for the non-uniformity correction result of the spectral dimension, so as to complete the non-uniformity correction of the hyperspectral image. The method specifically comprises the following steps:

[0009] 1) Select a hyperspectral image with a size of H×W×C after outlier processing, where H is the row, W is the column, and C is the number of bands. Automatically sort each scan line of the kth band of the hyperspectral image according to the DN values ​​of the pixels in different spatial dimensions from small to large, and calculate the average of each scan line By the mean Derivative curve The derivative curve of the uniform DN value area of ​​the spatial pixel is obtained. The maximum value of the median frequency corresponds to the row number L of the uniform area, where 1≤L≤H;

[0010] 2) Select the area L on the area with uniform DN value of the spatial dimension pixel 上 Behavior dark uniform area (L 上 ×W) k , lower area L 下 Behavior Bright Uniform Area (L 下 ×W) k , where 1≤k≤C, 1≤L 上 ≤H,1≤L 下 ≤H; calculate the dark uniform area (L 上 ×W) k The mean pixel DN value P 暗 (k) and bright uniform area (L 下 ×W) k The mean pixel DN value P 亮 (k); then calculate the dark uniform area (L 上 ×W) k The mean DN value of the pixel in the jth column is Q 暗 (j,k) and bright uniform area (L 下 ×W) k The mean DN value of the pixel in the jth column is Q 亮 (j,k), where 1≤j≤W;

[0011] 3) Construct the following equation to obtain the non-uniformity correction coefficients a(j, k) and b(j, k) of the j-th column pixel in the k-th band:

[0012]

[0013] 4) Perform non-uniform correction on the original response value DN(j,k) of the j-th column pixel in the k-th band of the push-broom hyperspectral image to obtain the corrected response value, and complete the non-uniform correction of the j-th column pixel in the k-th band of the image space dimension:

[0014] DN_spa(j,k)=a(j,k)×DN(j,k)+b(j,k);

[0015] 5) According to the above steps, the non-uniform correction of the original response value DN(i, j, k) of all bands of each pixel in the spatial dimension is completed, and the corrected response value DN of all bands of each pixel in each spatial dimension is obtained. _spa The mean value x of (i,j,k) mean_spa , calculate the original response mean x of all band spatial dimensions of each pixel mean With x mean_spa The ratio between them is used to obtain the non-uniform correction response value DN of the spectral dimension of the hyperspectral image. _spe (i,j,k):

[0016] DN _spe (i,j,k)=DN _spa (i,j,k)×ratio

[0017] Where i is the i-th row of the selected hyperspectral image, 1≤i≤H;

[0018] 6) Response value DN after spectral dimension correction _spe (i, j, k), repeat the above steps 1) to 4) to complete the hyperspectral image non-uniformity correction.

[0019] Furthermore, the mean in step 1) Derivative curve The equation is as follows:

[0020]

[0021] Furthermore, in step 2), the dark uniform area (L 上 ×W) k The mean pixel DN value P 暗 (k):

[0022]

[0023] Bright uniform area (L 下 ×W) k The mean pixel DN value P 亮 (k):

[0024]

[0025] Dark uniform area (L 上 ×W) k The mean DN value of the pixel in the jth column is Q 暗 (j,k):

[0026]

[0027] Bright uniform area (L 下 ×W) kThe mean DN value of the pixel in the jth column is Q 亮 (j,k):

[0028]

[0029] Where m is L 上 The specific row number, 1≤m≤L 上 ; n is L 下 The specific row number, 1≤n≤L 下 .

[0030] Furthermore, the response value DN after correction of all bands of each pixel in each spatial dimension in step 5) is _spa The mean value x of (i,j,k) mean_spa :

[0031]

[0032] The original response mean x of all band spatial dimensions of each pixel mean :

[0033]

[0034] The original response mean x of all band spatial dimensions of each pixel mean The response value DN after correction for all bands of each pixel in each spatial dimension _spa The mean value x of (i,j,k) mean_spa The ratio between:

[0035]

[0036] The present invention has the following advantages:

[0037] 1) No need to manually debug preset parameters, fully automated application, can break through scene limitations, adaptively select uniform areas, and is suitable for adaptive non-uniform correction of massive hyperspectral images in complex scenes, avoiding the instability of processing results and time consumption problems, with good versatility and universality, greatly simplifying the complexity and cost of correction;

[0038] 2) Aiming at the inherent characteristics of the push-broom hyperspectral imager, a joint spatial-spectral non-uniformity correction is performed based on its spatial and spectral characteristics, which can retain the spatial structure and spectral information of the hyperspectral image to the maximum extent, effectively improve the quality of the hyperspectral image, and lay the foundation for subsequent image analysis and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 It is a schematic flow chart of the scene-based push-broom hyperspectral image adaptive non-uniformity correction method in the present invention.

[0040] Figure 2 The present invention provides a schematic diagram of selecting dark uniform areas and bright uniform areas in the scene-based push-broom hyperspectral image adaptive non-uniform correction method.

[0041] Where (a) is the original image; (b) the DN of each column is sorted in ascending order; (c) the average of each row in (b) is calculated; (d) the derivative of the curve in (c) is calculated; (e) a histogram of the curve in (d) is drawn; (f) the uniform area is selected; (g) a dark uniform area and a bright uniform area are selected.

[0042] Figure 3 It is a schematic diagram showing abnormal values ​​of a push-broom hyperspectral image in a specific implementation manner of the present invention.

[0043] Where (a) is the original image; (b) sorts the DN of each column in ascending order; (c) draws the DN curve of each column in Figure (b).

[0044] Figure 4 is the original image of a hyperspectral image.

[0045] Figure 5 To use the method of the present invention Figure 4 The result image obtained after non-uniformity correction is performed on the original image in .

[0046] Figure 6 for Figure 4 and Figure 5 Spectral curve comparison chart. DETAILED DESCRIPTION

[0047] In order to make the purpose, features and advantages of the present invention clearer, a specific implementation of the present invention is described in more detail. In the following description, many specific details are elaborated to facilitate a full understanding of the present invention, but the present invention can be implemented in many other ways different from the description. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0048] The present invention provides a scene-based push-broom hyperspectral image adaptive non-uniform correction method, which automatically sorts each scanning line of a band of a hyperspectral image from small to large according to the DN values ​​of pixels in different spatial dimensions, and adaptively selects at least one dark uniform area and a bright uniform area in the spatial dimension to obtain a correction coefficient for each spatial dimension pixel, thereby completing the non-uniform correction of each spatial dimension; based on the spectral dimension characteristics, the spectral dimension non-uniform correction is completed for the non-uniform correction result of the spatial dimension, and the spatial dimension non-uniform correction is repeated for the non-uniform correction result of the spectral dimension, thereby completing the non-uniform correction of the hyperspectral image.

[0049] Specifically, an image with a size of 2688×2048×180 acquired by the visible shortwave infrared hyperspectral camera of the GF-5 satellite is taken as an example to explain in detail the specific implementation of the scene-based push-broom hyperspectral image adaptive non-uniformity correction method of the present invention.

[0050] Among them, Figure 3 As shown in the figure, the hyperspectral image with the original size of 2688×2048×180 needs to be processed with outliers first. The outlier processing includes the following steps: k ( 1≤k≤180 ) bands, arrange the pixels of the scanned lines from small to large, and display the smallest DN value of the column at the top of each line, and the largest DN value of the column at the bottom. Dozens of lines of high or low DN values ​​correspond to a group of objects with similar reflectivity. If the DN value of a certain pixel hardly changes along the push scan line, verify whether the data is consistent with the actual ground object situation. Even if there may be a constant DN value caused by an airplane runway or other ground objects in the data, such a completely consistent value is still not suitable for effective calibration and correction. For this, you can choose to replace such pixels by interpolating from surrounding pixels.

[0051] The scene-based push-broom hyperspectral image adaptive non-uniformity correction method specifically includes the following implementation steps:

[0052] 1) Select a hyperspectral image with a size of 2688×2048×180 after outlier processing, where 2688 is the number of rows, 2048 is the number of columns, and 180 is the number of bands. Automatically sort each scan line of the kth band of the hyperspectral image according to the DN values ​​of the pixels in different spatial dimensions from small to large, and calculate the average of each scan line By the mean Derivative curve The derivative curve of the uniform DN value area of ​​the spatial pixel is obtained. The maximum value of the median frequency corresponds to the row number L of the uniform area, where 1≤L≤2688;

[0053] Among them, the mean Derivative curve The equation is as follows:

[0054]

[0055] 2) If Figure 2 As shown, select the area L on the area with uniform DN value of the spatial dimension pixel 上 Behavior dark uniform area (L 上 ×2048) k , lower area L 下 Behavior Bright Uniform Area (L 下 ×2048) k , where 1≤k≤180, 1≤L上 ≤2688, 1≤L 下 ≤2688, select L 上 =50,L 下 =50; calculate the dark uniform area of ​​the kth band (50×2048) k The mean pixel DN value P 暗 (k) and bright uniform area (50×2048) k The mean pixel DN value P 亮 (k); then calculate the dark uniform area (50×2048) k The mean DN value of the pixel in the jth column is Q 暗 (j,k) and bright uniform area (50×2048) k The mean DN value of the pixel in the jth column is Q 亮 (j,k), where 1≤j≤2048;

[0056] Among them, the dark uniform area of ​​the kth band (50×2048) k The mean pixel DN value P 暗 (k):

[0057]

[0058] Bright uniform area (50×2048) k The mean pixel DN value P 亮 (k):

[0059]

[0060] Dark uniform area (50×2048) k The mean DN value of the pixel in the jth column is Q 暗 (j,k):

[0061]

[0062] Bright uniform area (50×2048) k The mean DN value of the pixel in the jth column is Q 亮 (j,k):

[0063]

[0064] Where m is L 上 The specific row number, 1≤m≤L 上 ; n is L 下 The specific row number, 1≤n≤L 下 .

[0065] 3) Construct the following equation to obtain the non-uniformity correction coefficients a(j, k) and b(j, k) of the j-th column pixel in the k-th band:

[0066]

[0067] 4) Perform non-uniform correction on the original response value DN(j,k) of the j-th column pixel in the k-th band of the push-broom hyperspectral image to obtain the corrected response value, and complete the non-uniform correction of the j-th column pixel in the k-th band of the image space dimension:

[0068] DN_spa(j,k)=a(j,k)×DN(j,k)+b(j,k);

[0069] 5) According to the above steps, the non-uniform correction of the original response value DN(i, j, k) of all bands of each pixel in the spatial dimension is completed, and the corrected response value DN of all bands of each pixel in each spatial dimension is obtained. _spa The mean value x of (i,j,k) mean_spa , calculate the original response mean x of all band spatial dimensions of each pixel mean With x mean_spa The ratio between them is used to obtain the non-uniform correction response value DN of the spectral dimension of the hyperspectral image. _spe (i,j,k):

[0070] DN _spe (i,j,k)=DN _spa (i,j,k)×ratio

[0071] Where i is the i-th row of the selected hyperspectral image, 1≤i≤2688;

[0072] Among them, the response value DN after correction of all bands of each pixel in each spatial dimension is _spa The mean value x of (i,j,k) mean_spa :

[0073]

[0074] The original response mean x of all band spatial dimensions of each pixel mean :

[0075]

[0076] The original response mean x of all band spatial dimensions of each pixel mean The response value DN after correction for all bands of each pixel in each spatial dimension _spa The mean value x of (i,j,k) mean_spa The ratio between:

[0077]

[0078] 6) Response value DN after spectral dimension correction _spe(i, j, k), repeat the above steps 1) to 4) to complete the hyperspectral image non-uniformity correction.

[0079] Figure 1 It is a schematic flow chart of the scene-based push-broom hyperspectral image adaptive non-uniformity correction method in the present invention.

[0080] In addition, we selected another uncorrected hyperspectral image (i.e. Figure 4 ), the original image is corrected using the scene-based push-broom hyperspectral image adaptive non-uniform correction method of the present invention, and the corrected result is shown in FIG. Figure 5 At the same time, Figure 4 and Figure 5 The spectral curve comparison is shown in the figure Figure 6 By comparison, it can be seen that the non-uniformity of the image has been significantly improved, the stripe noise phenomenon has been eliminated, the spatial structure information and spectral information of the hyperspectral image have been retained to the greatest extent, the quality of the hyperspectral image has been greatly improved, and the correction effect is very significant.

[0081] The above description is only a 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 scope of the present invention.

Claims

1. A scene-based push-broom hyperspectral image adaptive non-uniformity correction method, characterized in that: Each scanning line of a band of the hyperspectral image is automatically sorted from small to large according to the DN values ​​of pixels in different spatial dimensions, and at least one dark uniform area and bright uniform area in the spatial dimension are adaptively selected to obtain the correction coefficient of each spatial dimension pixel, and complete the non-uniform correction of each spatial dimension; based on the spectral dimension characteristics, the spectral dimension non-uniform correction is completed for the non-uniform correction result of the spatial dimension, and the non-uniform correction of the spatial dimension is repeated for the non-uniform correction result of the spectral dimension, so as to complete the non-uniform correction of the hyperspectral image, which specifically includes the following steps: 1) Select a hyperspectral image of size H×W×C where H is the row, W is the column, and C is the number of bands. Automatically sort each scan line of the kth band of the hyperspectral image according to the DN values ​​of the pixels in different spatial dimensions from small to large, and calculate the average of each scan line. By the mean Derivative curve The derivative curve of the uniform DN value area of ​​the spatial pixel is obtained. The maximum value of the median frequency corresponds to the row number L of the uniform area, where 1≤L≤H; 2) Select the area L on the area with uniform DN value of the spatial dimension pixels 上 Behavior dark uniform area (L 上 ×W) k , lower area L 下 Behavior Bright Uniform Area (L 下 ×W) k , where 1≤k≤C, 1≤L 上 ≤H,1≤L 下 ≤H; calculate the dark uniform area (L 上 ×W) k The mean pixel DN value P 暗 (k) and bright uniform area (L 下 ×W) k The mean pixel DN value P 亮 (k); then calculate the dark uniform area (L 上 ×W) k The mean DN value of the pixel in the jth column is Q 暗 (j,k) and bright uniform area (L 下 ×W) k The mean DN value of the pixel in the jth column is Q 亮 (j,k), where 1≤j≤W; 3) Construct the following equation to obtain the non-uniformity correction coefficients a(j, k) and b(j, k) of the j-th column pixel in the k-th band: 4) Perform non-uniform correction on the original response value DN(j,k) of the j-th column pixel in the k-th band of the push-broom hyperspectral image to obtain the corrected response value, and complete the non-uniform correction of the j-th column pixel in the k-th band of the image space dimension: DN_spa(j,k)=a(j,k)×DN(j,k)+b(j,k); 5) According to the above steps, the non-uniform correction of the original response value DN(i, j, k) of all bands of each pixel in the spatial dimension is completed, and the corrected response value DN of all bands of each pixel in each spatial dimension is obtained. _spa The mean value x of (i,j,k) mean_spa , and then calculate the original response mean x of all band spatial dimensions of each pixel mean With x mean_spa The ratio between them is used to obtain the non-uniform correction response value DN of the spectral dimension of the hyperspectral image. _spe (i,j,k): DN _spe (i,j,k)=DN _spa (i,j,k)×ratio Where i is the i-th row of the selected hyperspectral image, 1≤i≤H; 6) Response value DN after spectral dimension correction _spe (i, j, k), repeat the above steps 1) to 4) to complete the hyperspectral image non-uniformity correction.

2. The method for adaptive non-uniform correction of push-broom hyperspectral images based on a scene according to claim 1, characterized in that: The mean value in step 1) Derivative curve The equation is as follows:

3. The scene-based push-broom hyperspectral image adaptive non-uniformity correction method according to claim 1, characterized in that: The dark uniform area (L 上 ×W) k The mean pixel DN value P 暗 (k): Bright uniform area (L 下 ×W) k The mean pixel DN value P 亮 (k): Dark uniform area (L 上 ×W) k The mean DN value of the pixel in the jth column is Q 暗 (j,k): Bright uniform area (L 下 ×W) k The mean DN value of the pixel in the jth column is Q 亮 (j,k): Where m is L 上 The specific row number, 1≤m≤L 上 ; n is L 下 The specific row number, 1≤n≤L 下 .

4. The scene-based push-broom hyperspectral image adaptive non-uniformity correction method according to claim 1, characterized in that: The corrected response value DN of all bands of each pixel in each spatial dimension in step 5) _spa The mean value x of (i,j,k) mean_spa : The original response mean x of all band spatial dimensions of each pixel mean : The original response mean x of all band spatial dimensions of each pixel mean The response value DN after correction for all bands of each pixel in each spatial dimension _spa The mean value x of (i,j,k) mean_spa The ratio between:

Images