A hyperspace resolution reconstruction method for multi-slit hyperspectral image data

Abstract

The application discloses a hyperspace resolution reconstruction method for multi-slit hyperspectral image data, and solves the problems of low spatial resolution and poor quality of multi-slit hyperspectral data.The application establishes a unified spectral endmember matrix of each slit hyperspectral data and a difference-abundant matrix of each slit data by hyperspectral unmixing according to the correlation and difference between the multi-slit hyperspectral data.Then, the abundance coefficients of each slit data are unified under a reference image coordinate system (i.e., a first coordinate system) according to the image shift difference values between the multi-slit data, and the multi-slit image shift abundance matrix is obtained by interpolation according to the abundance matrix set.Finally, the high spatial resolution hyperspectral data are obtained by multiplying and superimposing the endmember spectral matrix and the endmember abundance matrix.

Description

Technical Field

[0001] This invention relates to multi-slit hyperspectral data, and more specifically to a method for superspatial resolution reconstruction of multi-slit hyperspectral image data. Background Technology

[0002] Multi-slit hyperspectral imaging systems acquire overall data of the entire surface area through push-broom scanning. Due to the use of multi-slit spectral dispersion, hyperspectral data from different columns of objects can be acquired at a single moment. Then, push-broom scanning can acquire multiple sets of hyperspectral image data of the same object at different times after spectral dispersion through different slits. There is correlation between multi-slit hyperspectral data. However, due to factors such as optical distortion and imaging conditions, there are slight image shift differences between the hyperspectral image data from each slit, resulting in variability among the multi-slit hyperspectral data. As an emerging hyperspectral imaging technology, multi-slit hyperspectral imaging currently lacks specific applications that address its correlation and variability characteristics; the correlation and variability of its multiple sampling are not fully realized or utilized.

[0003] The differences between multi-slit hyperspectral data are caused by different scanned targets, and the image shift between data points represents the movement of the imaging region. In general, these differences reflect the target variations in the spatial resolution of the original hyperspectral data and can be used for hyperspatial resolution reconstruction based on the original data. Hyperspatial resolution reconstruction is significant for improving the quality of original hyperspectral data and for identifying small targets and classifying ground features with high accuracy. However, currently, there are no known methods for hyperresolution reconstruction of multi-slit hyperspectral data. Summary of the Invention

[0004] To address the technical problems of low spatial resolution and poor quality in multi-slit hyperspectral data, this invention provides a super-spatial resolution reconstruction method for multi-slit hyperspectral image data.

[0005] The inventive concept of this invention:

[0006] This invention addresses the correlations and differences among multi-slit hyperspectral data. Through hyperspectral unmixing, a unified endmember spectral matrix for each slit's hyperspectral data and distinct abundance matrices for each slit are established. Then, based on the image shift differences among the multi-slit data, the abundance coefficients of each slit are unified in a reference image coordinate system (i.e., the first coordinate system). Finally, the image shift abundance matrix of the multi-slit data is obtained through interpolation using the abundance matrix set. The endmember spectral matrix and the endmember abundance matrix are multiplied and superimposed to obtain high spatial resolution hyperspectral data.

[0007] To achieve the above objectives and complete the above inventive concept, the present invention adopts the following technical solution:

[0008] A method for superspatial resolution reconstruction of multi-slit hyperspectral image data, characterized by the following steps:

[0009] Step 1: Acquire multiple hyperspectral image data of the same target from the multi-slit hyperspectral imaging system. Set the hyperspectral image data acquired by one slit as the reference and record it as the first hyperspectral image data. The rest are the second hyperspectral image data.

[0010] Step 2: Obtain the image shift of each pixel in the second hyperspectral image data relative to each pixel in the first hyperspectral image data;

[0011] Step 3: Obtain the unified endmember spectral matrix of the first hyperspectral image data and the second hyperspectral image data, and obtain the abundance matrix of each first hyperspectral image data and the second hyperspectral image data respectively;

[0012] Step 4: Based on the first hyperspectral image data, and according to the image shift, transform the abundance matrix of the second hyperspectral image data from its original second coordinate system to the first coordinate system of the first hyperspectral image data;

[0013] Step 5: Combine the abundance matrix of the first hyperspectral image data with the abundance matrix of the first hyperspectral image data in the first coordinate system to form a multi-slit image-shifting abundance matrix;

[0014] Step 6: Based on the preset hyperspatial resolution factor and the first hyperspectral image data, obtain the position coordinates of each pixel under high spatial resolution;

[0015] Step 7: Based on the multi-slit image shift abundance matrix formed in Step 5, interpolate to obtain the endmember abundance matrix of each endmember at each pixel position coordinate obtained in Step 6.

[0016] Step 8: Multiply and superimpose the endmember spectral matrix obtained in Step 3 with the endmember abundance matrix obtained in Step 7 to obtain high spatial resolution hyperspectral data, thus completing the superspatial resolution reconstruction of the multi-slit hyperspectral image data.

[0017] Furthermore, step 2 specifically involves:

[0018] The image shift of each pixel in the second hyperspectral image data relative to each pixel in the first hyperspectral image data is obtained based on the detection error value of the multi-slit hyperspectral imaging system.

[0019] Furthermore, step 2 specifically involves:

[0020] The image shift of each pixel in the second hyperspectral image data relative to each pixel in the first hyperspectral image data is calculated using the HS dense optical flow method.

[0021] Furthermore, step 1 specifically includes:

[0022] Multiple hyperspectral image data of the same target are acquired by a multi-slit hyperspectral imaging system. The hyperspectral image data acquired by the slit with the best imaging quality is set as the benchmark and denoted as the first hyperspectral image data. The rest are the second hyperspectral image data.

[0023] The beneficial effects of this invention are:

[0024] This invention provides a hyperspatial resolution reconstruction method for multi-slit hyperspectral image data. Targeting the characteristics of multi-slit hyperspectral image data, it enriches the application scope of slit hyperspectral image data, obtains hyperspectral data with higher spatial resolution than the original data, and can further increase the number of pixels occupied by small sub-pixel targets, thereby improving recognition accuracy. The acquisition of different endmember spectral matrices and endmember abundance matrices can further remove the influence of smoke and noise on slit hyperspectral image data, while also improving target classification accuracy. This method is of great significance for the application of multi-slit hyperspectral imaging systems. Attached Figure Description

[0025] Figure 1 This is a schematic diagram of push-broom imaging on a dual-slit hyperspectral imaging system platform;

[0026] Figure 2 This is a flowchart of a method for superspatial resolution reconstruction of multi-slit hyperspectral image data according to the present invention. Detailed Implementation

[0027] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0028] First, it's important to clarify that due to spatial resolution limitations, the spectrum obtained from a single pixel in hyperspectral data is not necessarily the spectrum of a single substance; it may be a combination of the spectra of several different substances. Such pixels are called mixed pixels. The spectral characteristics of a single substance are called an endmember, and the abundance of this substance within a hyperspectral data pixel is called its abundance. Therefore, hyperspectral data can be viewed as a combination of spectral endmembers of multiple substances and the abundance of each endmember in each pixel. Hyperspectral demixing is the process of decomposing hyperspectral data into multiple spectral endmembers and the abundance of data for each pixel.

[0029] This embodiment provides a method for superspatial resolution reconstruction of multi-slit hyperspectral image data, such as... Figure 1As shown, this embodiment uses a dual-slit hyperspectral imaging system with a spatial dimension of 1×3 (length × width) and a spectral dimension of one band. Other embodiments may use a three-slit hyperspectral imaging system, etc.

[0030] like Figure 2 As shown, the method includes the following steps:

[0031] Step 1: Determine two sets of low spatial resolution hyperspectral image data acquired by the dual-slit hyperspectral imaging system for the same target. Set the hyperspectral image data acquired by slit 1 as the benchmark and denot it as the first hyperspectral image data. The slit hyperspectral image data acquired by slit 2 is denoted as the second hyperspectral image data.

[0032] Step 2: Based on the detection error value of the dual-slit hyperspectral imaging system or by obtaining the image shift of each pixel in the second hyperspectral image data relative to each pixel in the first hyperspectral image data through the HS dense optical flow method, obtain the image shift matrix M = [0.2, 0.4, 0.3].

[0033] Step 3: Combine the first and second hyperspectral image data to obtain a unified endmember spectral matrix D(D1,D2,…D…). t ),D1,D2,…D t Given the spectral matrices of endmembers 1, 2, ..., t, and obtain the abundance matrix F1 = [F...] of the first hyperspectral image data. 11 ,F 12 ,…F 1t ], where F 1t =[(1,1,f 1t1 ),(1,2,f 1t2 ),(1,3,f 1t3 The abundance matrix F2 of the second hyperspectral image data is F2 = [F 21 ,F 22 ,…F 2t ], where F 2t =[(1,1,f 2t1 ),(1,2,f 2t2 ),(1,3,f 2t3 )], where f 1t1 f1 t 2. f1 t 3 represents the abundance coefficients of endmember t corresponding to the reference pixel coordinates (1,1), (1,2), and (1,3), respectively; f 2t1 f 2t2 f 2t3 These are the abundance coefficients of endmember 1, endmember 2, ..., endmember t corresponding to the coordinate positions of the second hyperspectral image data at its own pixel coordinates (1,1), (1,2), (1,3), where t is the number of endmembers.

[0034] Step 4: Based on the first hyperspectral image data, and according to the image shift obtained in Step 2, transform the image pixel coordinates corresponding to the abundance matrix F2 of the second hyperspectral image data from its original image coordinate system (second coordinate system) to the reference image coordinate system (first coordinate system) of the first hyperspectral image data to obtain F2' = [F 21 ',F 22 ',…F 2t '], where F 2t =[(1,1.2,f 2t1 ),(1,2.4,f 2t2 ),(1,3.3,f 2t3 )).

[0035] Step 5: Combine the abundance matrix of the first hyperspectral image data and the abundance matrix of the first hyperspectral image data in the reference image coordinate system (first coordinate system) into a single multi-slit image-shifting abundance matrix F. all =[F 1all ,F 2all ,…F tall ],in

[0036] F tall =[(1,1,f 1t1 ),(1,1.2,f 2t1 ),(1,2,f 1t2 ),(1,2.4,f 2t2 ),(1,3,f 13 ),(1,3.3,f 2t3 )).

[0037] Step 6: Based on the hyperspectral data with a spatial resolution of 1 times that of the original data and the position coordinates of each pixel in the first hyperspectral image data, obtain the position coordinates P of each pixel in the first hyperspectral image data. The size of P is 1×5. The corresponding coordinates of each pixel are: (1,1), (1,1.5), (1,2), (1,2.5), (1,3).

[0038] Step 7: Based on the multi-slit image shift abundance matrix F formed in Step 5 all Interpolation yields the endmember abundance matrix F of each endmember under the high spatial resolution pixel location coordinates P obtained in step 6. allc =[F 1allc ,F 2allc ,…F tallc ]; Specifically as follows:

[0039] F tallc =[((1,1,f 1t1 ),(1,1.5,f tc1),(1,2,f 1t2 ),(1,2.5,f tc2 ),(1,3,f 1t3 )]

[0040] Among them, f c1 f c2 f 13 The abundance coefficients are obtained from super-resolution interpolation.

[0041] Step 8: Based on the endmember spectral matrix D(D1,D2,…D) obtained in Step 3 t The coefficients F of the hyperspatial resolution abundance matrix obtained in step 7 allc =[F 1allc ,F 2allc ,…F tallc The corresponding products are multiplied and superimposed to obtain high spatial resolution hyperspectral data S. c .

[0042]

[0043] S c This refers to the hyperspectral data obtained after super spatial resolution reconstruction, which has a spatial resolution that is twice that of the original hyperspectral data.

[0044] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions within the technical scope disclosed in the present invention should be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for superspatial resolution reconstruction of multi-slit hyperspectral image data, characterized in that, Includes the following steps: Step 1: Acquire multiple hyperspectral image data of the same target from the multi-slit hyperspectral imaging system. Set the hyperspectral image data acquired by one slit as the reference and record it as the first hyperspectral image data. The rest are the second hyperspectral image data. Step 2: Obtain the image shift of each pixel in the second hyperspectral image data relative to each pixel in the first hyperspectral image data; Step 3: Obtain the unified endmember spectral matrix of the first hyperspectral image data and the second hyperspectral image data, and obtain the abundance matrix of each first hyperspectral image data and the second hyperspectral image data respectively; Step 4: Based on the first hyperspectral image data, and according to the image shift, transform the abundance matrix of the second hyperspectral image data from its original second coordinate system to the first coordinate system of the first hyperspectral image data; Step 5: Combine the abundance matrix of the first hyperspectral image data with the abundance matrix of the first hyperspectral image data in the first coordinate system to form a multi-slit image-shifting abundance matrix; Step 6: Based on the preset hyperspatial resolution factor and the first hyperspectral image data, obtain the position coordinates of each pixel under high spatial resolution; Step 7: Based on the multi-slit image shift abundance matrix formed in Step 5, interpolate to obtain the endmember abundance matrix of each endmember at each pixel position coordinate obtained in Step 6. Step 8: Multiply and superimpose the endmember spectral matrix obtained in Step 3 with the endmember abundance matrix obtained in Step 7 to obtain high spatial resolution hyperspectral data, thus completing the superspatial resolution reconstruction of the multi-slit hyperspectral image data.

2. The method for superspatial resolution reconstruction of multi-slit hyperspectral image data according to claim 1, characterized in that, Step 2 is as follows: The image shift of each pixel in the second hyperspectral image data relative to each pixel in the first hyperspectral image data is obtained based on the detection error value of the multi-slit hyperspectral imaging system.

3. The method for superspatial resolution reconstruction of multi-slit hyperspectral image data according to claim 1, characterized in that: Step 2 is as follows: The image shift of each pixel in the second hyperspectral image data relative to each pixel in the first hyperspectral image data is calculated using the HS dense optical flow method.

4. The method for superspatial resolution reconstruction of multi-slit hyperspectral image data according to claim 1, 2, or 3, characterized in that, Step 1 is as follows: Multiple hyperspectral image data of the same target are acquired by a multi-slit hyperspectral imaging system. The hyperspectral image data acquired by the slit with the best imaging quality is set as the benchmark and denoted as the first hyperspectral image data. The rest are the second hyperspectral image data.

Images