A pixel-level calibration and correction method based on a diamond-arranged DMD

Abstract

The present invention discloses a pixel-level calibration and correction method based on a diamond arrangement DMD, comprising the following steps; Step (1): spectral calibration; Step (2): obtaining a multi-spectral image and an RGB image; Step (3): jointly filling the RGB image with the multi-spectral image: Step (4): pixel-level image correction: Step (5): deriving the image pixel ratio relationship of each spectral image in the three-dimensional spectral data cube according to the image pixel ratio relationship in the original RGB image coordinate information, and updating the three-dimensional spectral data cube to obtain the three-dimensional spectral data cube. The present invention fully considers the uniqueness of each sampling point in the spectral calibration process, fills the spatial sampling points with prior channel losses caused by DMD deflection, solves the block effect problem in the pushbroom imaging process, thereby completing the pixel-level calibration and correction of the multi-spectral image and improving the quality of the reconstructed multi-spectral image.

Description

Technical Field

[0001] The present invention relates to the technical field of spectral image processing, and particularly relates to a pixel-level calibration and correction method based on a diamond-arranged DMD. Background Art

[0002] Currently, for spectral calibration of an imaging system based on a Digital Micromirror Device (DMD), a monochromator is mainly used to traverse all pixel points and record the relative positions, and the point with the maximum pixel value is selected as the response position of the current sampling point. However, under actual application conditions, the point spread functions formed by the sampling points corresponding to different spatial positions and different wavelength bands are not the same. Directly selecting the point with the maximum pixel value cannot accurately measure the energy distribution of the sampling point, resulting in errors in the spectral calibration results.

[0003] Existing DMD-based multispectral imaging systems have not considered the influence of the DMD arrangement on the hardware system. Specifically, although the square-arranged DMD can achieve one-to-one pixel registration with the detector, its characteristic of reversing according to the diagonal will cause a height difference in the system, which is not conducive to alignment debugging. For the more widely used diamond-arranged DMD, there is a lack of research on pixel-level registration with a general detector.

[0004] Currently, a DMD-based dual-channel multispectral video imaging system uses DMD reflection to generate two complementary sets of spatially encoded images. Therefore, regardless of the spatial encoding method used, the prior channel cannot obtain a complete scene space point at a certain moment. The incomplete spatial sampling points prevent the system from directly applying traditional image restoration algorithms for reconstruction.

[0005] The diamond-arranged DMD dual-channel multispectral imaging system inevitably has a matching problem between diamond encoding and square detection, which is particularly obvious in pushbroom imaging. The block effect problem generated by pushbroom will seriously affect the quality of the reconstructed multispectral image and hinder the development of subsequent related research.

[0006] The existing patent "A Method and Storage Medium for Correcting Asymmetric Deformation of DMD Micro-Mirror Element Imaging" (Publication No.: CN107845077A, Application No.: 201711057691.9, Application Date: November 1, 2017) discloses a method for accurately calibrating the position relationship between DMD mirror elements and detector pixels by using the DMD to load standard square-loop encoded stripes. By calculating the slope, offset number, and correction accuracy of the square-loop encoded stripes in the loaded encoded image, this method solves the problem that the imaging system cannot achieve clear imaging in the full field of view when solving this problem by adjusting the detector angle in the traditional way. The deficiency of this patent is that the processing object is the DMD arranged in a square shape, and for the DMD arranged in a rhombus shape, there will be a problem that the pixels of the rhombus and the square cannot be accurately registered, and the reconstructed image has obvious block artifacts.

[0007] The existing patent "A Method for Measuring and Correcting Light Intensity Non-Uniformity in a Digital Lithography System" (Publication No.: CN105652607A, Application No.: 201610216641.X, Application Date: April 8, 2016) discloses a method for calibrating a CCD camera using an integrating sphere, dividing the object plane of the DMD into a rectangular area array of N×N, and then obtaining the light intensity distribution of each rectangular area, obtaining the overall light intensity distribution map of the image plane and the overall light intensity distribution on the DMD plane, and completing the measurement of light intensity non-uniformity. The pre-measurement and correction of light intensity non-uniformity are achieved by combining the control of the exposure times of each pixel point in the scanning exposure to ensure that the final illumination uniformity reaches 95%. The deficiency of this patent is that it mainly considers the correction method of light intensity non-uniformity and does not involve the correction of the reconstructed image in terms of deformation.

[0008] In summary, although the existing methods can perform some image corrections on the DMD-based spectral imaging system, they do not consider the image deformation problem caused by using the more widely used rhombus-arranged DMD as the encoding device. Summary of the Invention

[0009] In order to overcome the above defects existing in the prior art, the purpose of the present invention is to provide a pixel-level calibration and correction method based on a rhombus-arranged DMD, fully considering the uniqueness of each sampling point in the spectral calibration process, filling the spatial sampling points lost due to the deflection of the DMD to cause prior channel loss, solving the block artifact problem in the pushbroom imaging process, thereby completing the pixel-level calibration and correction of the multi-spectral image and improving the quality of the reconstructed multi-spectral image.

[0010] In order to achieve the above purpose, the technical solution adopted by the present invention is:

[0011] A pixel-level calibration and correction method based on a rhombus-arranged DMD, comprising the following steps;

[0012] Step (1): Spectral calibration;

[0013] In a dual-channel spectral imaging system based on a diamond-arranged DMD, a non-aliasing coding template is used to encode the scene. By changing the coding template pattern, all pixels on the DMD with a resolution of M×N are traversed by pushbroom scanning, where:

[0014] On the spectral channel, monochromatic light is incident band by band. The encoded scene is split by a dispersion prism and then captured by a detector. The coordinates C λ (x,y) and the point spread function P λ (x,y) corresponding to each pixel D(x,y) on the DMD at different wavelengths in the detector are recorded;

[0015] On the prior channel, white light is incident. The encoded scene is reflected by the DMD and then captured by a color detector. The coordinates C RGB (x,y) and the point spread function P RGB (x,y) corresponding to each pixel D(x,y) on the DMD in the color detector are recorded; Step (2): Obtain the multi-spectral image and the RGB image;

[0016] (2a) Use the dual-channel spectral imaging system based on the diamond-arranged DMD to collect the multi-spectral image of the scene, obtaining the multi-spectral image. Load the designed aliasing coding template pattern into the DMD. The sampling points reach the spectral channel after being reflected by the DMD, and then are split by a dispersion prism and captured by the detector. According to the coordinates C λ (x,y) and the point spread function P λ (x,y) corresponding to each pixel D(x,y) on the DMD, the multi-spectral image of the scene corresponding to the sampling points is restored. The template is changed and pushbroom scanning is performed on the static scene to traverse all spatial position points on the DMD, obtaining the three-dimensional spectral data cube I 1 (x,y,λ);

[0017] (2b) Use the dual-channel spectral imaging system based on the diamond-arranged DMD to collect the RGB image of the scene, obtaining the RGB image with missing sampling points. Load the designed aliasing coding template pattern into the DMD. The sampling points complementary to the spectral channel reach the prior channel after being reflected by the DMD and are directly captured by the color detector. According to the coordinates C RGB (x,y) and the point spread function P RGB (x,y) corresponding to each pixel D(x,y) on the DMD, the RGB image of the scene corresponding to the complementary sampling points is restored;

[0018] Step (3): Jointly fill the RGB image with the multi-spectral image:

[0019] (3a) Obtain the pixel value information of each spectral band at the sampling points in the multispectral image in (2a), and jointly fit the pixel values of the corresponding RGB three channels at the sampling points according to the camera response curve;

[0020] (3b) Fill the RGB pixel values corresponding to the sampling points after fitting into the captured scene RGB image in (2b) to complete the filling of the missing points in the RGB image;

[0021] Step (4): Pixel-level image correction:

[0022] (4a) Perform Fourier transform on each spectral image in the three-dimensional spectral data cube I 1 (x, y, λ) to the frequency domain, and remove the blocking effect by suppressing the components of the frequencies in the specified region, obtaining the three-dimensional spectral data cube I 2 (x, y, λ);

[0023] (4b) Perform normalization processing on each spectral image in the three-dimensional spectral data cube I 2 (x, y, λ) separately to obtain the three-dimensional spectral data cube I 3 (x, y, λ);

[0024] Step (5): According to the pixel ratio relationship of the image in the original RGB image coordinate information C RGB (x, y), deduce the pixel ratio relationship of each spectral image in the three-dimensional spectral data cube I 3 (x, y, λ), and update the three-dimensional spectral data cube I 3 (x, y, λ) to obtain the three-dimensional spectral data cube I 4 (x, y, λ).

[0025] In the said step (1), the anti-aliasing encoding template adopts a horizontal stripe encoding template, where the stripe interval T is greater than the dispersion distance S of the sampling points, meeting the requirement of no dispersion aliasing. The size of the encoding template is M×N, the first template pattern is (nT + 1)×N, n = 0, 1,..., M / T - 1. By moving the horizontal stripe encoding template down row by row, the encoding template pattern is changed. The number of template patterns of the anti-aliasing encoding template is equal to the stripe interval distance, and the changing template patterns are successively (nT + p)×N, n = 0, 1,..., M / T, p = 1, 2,..., T, where n is the number of stripes and p is the stripe offset distance.

[0026] The two channels of the rhombus-arranged DMD adopt a two-channel multispectral video imaging system of the rhombus-arranged DMD. The DMD is used for spatial encoding and spectral splitting simultaneously. Two complementary sets of spatially encoded images are generated according to the encoding pattern. One path is dispersed by a dispersion prism to a black and white detector, and the other path is directly acquired by a color detector.

[0027] In the step (2a): for different λ, change the horizontal stripe encoding template one by one, and the changed template patterns are (nT + p)×N in sequence, where n = 0, 1,..., M / T, p = 1, 2,..., T, and record I at each wavelength 1 (x, y, λ) = C λ (x, y)*P λ (x, y); where n is the number of stripes and p is the stripe offset distance.

[0028] In the step (2b), according to each pixel D(x, y) on the DMD and the coordinates C RGB (x, y) and the point spread function P RGB (x, y), recover the scene RGB image I RGB互补 (x, y) corresponding to the complementary sampling points, and at the same time record the RGB background image I RGB背景 (x, y) obtained from the prior channel in the state of the inapplicable template;

[0029] The template pattern of the spectral channel changes to (nT + p)×N, where n = 0, 1,..., M / T, p = 1, 2,..., T, where n is the number of stripes and p is the stripe offset distance, and the template pattern of the prior channel is (nT + q)×N, where n = 0, 1,..., M / T, (q ≠ p) && (q ∈ [1, T]).

[0030] The number of template patterns of the aliasing encoding template is equal to the stripe interval distance, and the changed template patterns are (nT + p)×N in sequence, where n = 0, 1,..., M / T, p = 1, 2,..., T; where n is the number of stripes and p is the stripe offset distance.

[0031] In the step (3a): the pixel value information of each spectral segment of the sampling points in the multispectral image, that is, the value of I 1 (x, y, λ), where x and y are the known sampling points designed by the template;

[0032] The fitting formula is:

[0033]

[0034] where λ 总 is the total number of wavelengths.

[0035] The step (3b) is specifically:

[0036] Fill the RGB pixel values corresponding to the fitted sampling points into the scene RGB image I RGB互补 (x, y) in (2b) to complete the filling of the missing points of the RGB image, and the calculated I R (x, y), IG (x, y), I B (x, y) is filled into I RGB互补 In (x, y, λ), I is obtained RGB (x, y, λ).

[0037] In the said step (4b), the formula for normalization processing is:

[0038]

[0039] The derivation process in the said step (5) is:

[0040] According to the RGB background image I obtained in (2b) RGB背景 (x, y) and the I obtained in (3b) RGB (x, y, λ) to calculate the pixel change rate, and combine with the three-dimensional spectral data cube I 3 The pixel ratio relationship of each spectral image in (x, y, λ) is used to obtain the three-dimensional spectral data cube I 4 (x, y, λ),

[0041] Advantages of the present invention:

[0042] 1. The present invention has completed a spectral calibration scheme based on a diamond arrangement DMD, including determining the point spread and function of each pixel point and the corresponding position, and solving the problem that the diamond and square pixel levels cannot be accurately registered when using a diamond arrangement DMD for imaging currently;

[0043] 2. The present invention has completed the acquisition of two images in the dual-channel spectral imaging based on the diamond arrangement DMD. Specifically, according to the calibration result, a multi-spectral image and an RGB image are obtained, and a static scene is pushed and scanned to obtain a complete spectral data cube;

[0044] 3. The present invention fills the missing sampling points in the RGB image by combining the multi-spectral image and the camera response curve, laying a foundation for realizing dual-channel joint reconstruction;

[0045] 4. The present invention removes the block effect phenomenon by suppressing partial components in the frequency domain and normalizing the spectral data cube, and completes image correction by combining with the RGB image. Description of the Drawings

[0046] Figure 1 It is the multi-spectral imaging flow chart of the diamond arrangement DMD of the present invention.

[0047] Figure 2 It is the schematic diagram of the pixel-level calibration and correction method based on the diamond arrangement DMD of the present invention.

[0048] Figure 3Schematic diagram of field of view loss due to rotation of the DMD.

[0049] Figure 4 Schematic diagram of pixel-level registration of the diamond-arranged DMD and the detector. Among them, (a) is the diamond-arranged DMD, (b) is the detector arrangement, and (c) is the schematic diagram of pixel-level registration error. Detailed implementation manners

[0050] The present invention will be further described in detail below with reference to the accompanying drawings of the specification.

[0051] Figure 1 The flow chart of the diamond-arranged DMD multi-spectral imaging system based on the present invention is shown. Before imaging, the system is spectrally calibrated, that is, the relative positions of the pixel points on the DMD corresponding to the two detectors are determined. Since the point spread functions of the sampling points are different in different bands and spatial positions, point-by-point calibration is adopted, and the energy distribution of each sampling point is recorded and fitted with a Gaussian function. The fusion process requires the RGB image obtained from the prior channel to be in the full sampling mode. Therefore, the missing point calculation and filling in the RGB image are completed by combining the distribution information of the complementary sampling points in the spectral channels. The errors caused by the matching of the diamond DMD and the square detector are jointly corrected to improve the block effect problem. Thus, the pixel-level calibration and correction of the multi-spectral image are completed, and the quality of the reconstructed multi-spectral image is improved.

[0052] As Figure 2 shown, the present invention includes the following steps:

[0053] Spectral calibration:

[0054] (1a) In a dual-channel spectral imaging system based on a diamond-arranged DMD, a non-aliasing coding template is used to encode the scene. By changing the coding template pattern, all pixels on the DMD with a resolution of M×N are traversed by pushbroom. In the spectral channel, monochromatic light is incident band by band using a monochromator. The encoded scene is split by a dispersive prism and captured by the detector, and the coordinates C λ (x,y) and the point spread function P λ (x,y) of each pixel D(x,y) on the DMD at different wavelengths in the detector are recorded;

[0055] (1b) In a dual-channel spectral imaging system based on a diamond-arranged DMD, a non-aliasing coding template is used to encode the scene. By changing the coding template pattern, all pixels on the DMD with a resolution of M×N are traversed by pushbroom. In the prior channel, white light is incident. The encoded scene is reflected by the DMD and captured by the color detector, and the coordinates C RGB (x,y) and the point spread function P RGB (x,y) of each pixel D(x,y) on the DMD in the color detector are recorded;

[0056] Obtain multispectral images and RGB images:

[0057] (2a) Use a dual-channel spectral imaging system based on a diamond-arranged DMD to collect multispectral images of the scene. Load the designed aliasing coding template pattern into the DMD. The sampling points are reflected by the DMD and reach the spectral channel, and then are captured by the detector after being dispersed by the dispersion prism. According to each pixel D(x, y) on the DMD and the coordinates C λ (x, y) and the point spread function P λ (x, y), restore the multispectral image of the scene corresponding to the sampling points. Push and scan by changing the template for the static scene, traverse all spatial position points on the DMD, and obtain the three-dimensional spectral data cube I 1 (x, y, λ);

[0058] (2b) Use a dual-channel spectral imaging system based on a diamond-arranged DMD to collect RGB images of the scene. Load the designed aliasing coding template pattern into the DMD. The sampling points complementary to the spectral channel are reflected by the DMD and reach the prior channel, and are directly captured by the color detector. According to each pixel D(x, y) on the DMD and the coordinates C RGB (x, y) and the point spread function P RGB (x, y), restore the RGB image of the scene corresponding to the sampling points;

[0059] Jointly fill the RGB image with the multispectral image:

[0060] (3a) Obtain the pixel value information of each spectral band of the sampling points in the multispectral image in (2a), and jointly fit the pixel values of the three RGB channels corresponding to the sampling points according to the camera response curve;

[0061] (3b) Fill the pixel values of the RGB corresponding to the fitted sampling points into the RGB image of the scene captured in (2b) to complete the filling of the missing points in the RGB image;

[0062] (4) Pixel-level image correction:

[0063] (4a) Transform each spectral image in the three-dimensional spectral data cube I 1 (x, y, λ) into the frequency domain, and remove the blocking effect by suppressing the components of the specified region frequencies, to obtain the three-dimensional spectral data cube I 2 (x, y, λ);

[0064] (4b) Individually normalize each spectral image in the three-dimensional spectral data cube I 2 (x, y, λ) to obtain the three-dimensional spectral data cube I 3 (x, y, λ);

[0065] (4) Derive the three-dimensional spectral data cube I based on the image pixel ratio relationship in the original RGB image coordinates C RGB (x, y) in the image pixel ratio relationship of each spectral image in (x, y, λ), and update the three-dimensional spectral data cube I 3 (x, y, λ) to obtain the three-dimensional spectral data cube I 3 (x, y, λ). 4 (x, y, λ).

[0066] The present invention is a dual-channel spectral imaging system based on a rhombic arrangement DMD, which performs spectral calibration on the point spread function and corresponding position of each pixel point; obtains multi-spectral images and RGB images according to the corresponding relationship between the DMD and two detectors; fills the missing sampling points in the RGB by combining the multi-spectral images and the camera response curve; realizes pixel-level alignment and matching, uses frequency domain suppression of partial components to remove blocking effects, and combines the RGB image for spectral image correction.

[0067] The principle of the present invention:

[0068] The dual-channel spectral imaging system based on a rhombic arrangement DMD uses the DMD device to simultaneously complete spectral splitting and spatial encoding. The scene is spatially encoded by loading the encoding pattern on the DMD. Part of the information is reflected to the spectral channel, and after being split by a dispersion prism, it reaches the detector. The other part of the complementary encoded information is reflected to the prior channel and captured by a color camera. By fusing the image information of the two channels, the multi-spectral reconstruction task of the scene can be completed.

[0069] During the entire imaging process, it should be noted that:

[0070] System spectral calibration: Determine the relative position of each pixel point on the DMD corresponding to the detector after processes such as reflection and dispersion;

[0071] Obtain multi-spectral images and color images: The sampling points on the spectral channel are split along the dispersion direction and captured by the detector, and the complementary sampling points on the prior channel reflect the real spatial information of the scene. Therefore, regardless of the spatial encoding method used, it is usually necessary that the scene information obtained by the color camera in the prior channel is complete;

[0072] Pixel-level matching: The square DMD will cause the system not to be on a horizontal plane. Therefore, most DMD-based spectral imaging systems use a rhombic arrangement DMD as the encoding device, which inevitably introduces a pixel-level registration problem with the detector.

[0073] The outer square represents the pixel on the detector, the inner square represents the DMD pixel aligned with the detector after being tilted 45°, and the square above the inner square represents the pixel that cannot be captured by the detector.

[0074] Since each micromirror of the DMD can only flip around its own diagonal and the target surface of the DMD is rectangular, in order to align with the pixels of the sensor, the current DMD system must rotate the DMD or the camera by 45° to form an alignment of "square mirror pixels". This may simplify the simulation of the recovery algorithm, but it also brings two disadvantages to the engineering. First, as Figure 3 shown, the sensor cannot capture a large number of micromirrors. Of course, a larger sensor can be used to cover the DMD target surface, but this will result in waste of sensor pixels. Second, excitingly rotating a device makes assembly and debugging more tricky. Therefore, a rhombic DMD is used as a spatial light modulator to ensure that the reflected light path and the incident light path are on the same horizontal plane, improving the stability of the system without rotating the camera. However, the use of a rhombic DMD will inevitably introduce pixel-level registration problems between detectors, as Figure 4 shown.

Claims

1. A pixel-level calibration and correction method based on a diamond-arranged DMD, characterized in that, it includes the following steps; Step (1): Spectral calibration; In a dual-channel spectral imaging system based on a diamond-arranged DMD, a non-aliasing coding template is used to encode the scene. By changing the coding template pattern, all pixels on the DMD with a resolution of M×N are traversed by push-broom, where: On the spectral channel, using a monochromator to incident waveband by waveband, the encoded scene is dispersed by a dispersive prism and then captured by a detector, recording the coordinates C corresponding to each pixel D(x, y) on the DMD at different wavelengths in the detector λ (x, y) and the point spread function P λ (x, y); On the prior channel, with white light incident, the encoded scene is reflected by the DMD and captured by the color detector, recording the coordinates C of each pixel D(x, y) on the DMD in the color detector. RGB (x, y) and the point spread function P RGB (x, y); Step (2): Obtain multi-spectral images and RGB images; (2a) Use a dual-channel spectral imaging system based on a diamond-arranged DMD to collect multi-spectral images of a scene, obtaining multi-spectral images. Load the aliasing coding template pattern into the DMD. The sampling points reach the spectral channels after being reflected by the DMD, and then are captured by the detector after being dispersed by the dispersion prism. According to each pixel D(x, y) on the DMD and the coordinates C λ (x, y) and the point spread function P λ (x, y), restore the multi-spectral image of the scene corresponding to the sampling points. Replace the template and push-scan the static scene, traverse all spatial position points on the DMD, and obtain the three-dimensional spectral data cube I 1 (x, y, λ); (2b) Use a dual-channel spectral imaging system based on a diamond-arranged DMD to collect RGB images of the scene, obtaining an RGB image with missing sampling points. Load the aliasing coding template pattern into the DMD. The sampling points complementary to the spectral channel reach the prior channel after being reflected by the DMD and are directly captured by the color detector. According to each pixel D(x, y) on the DMD and the coordinates C RGB (x, y) and the point spread function P RGB (x, y), recover the RGB image of the scene corresponding to the complementary sampling points; Step (3): Jointly fill the RGB image with the multi-spectral image: (3a) Obtain the pixel value information of each spectral band at the sampling points in the multi-spectral image in (2a), and jointly fit the pixel values of the corresponding RGB three channels of the sampling points with the camera response curve; (3b) Fill the RGB pixel values corresponding to the fitted sampling points into the captured scene RGB image in (2b) to complete the filling of the missing points in the RGB image; Step (4): Pixel-level image correction: (4a) Perform Fourier transform on each spectral image in the three-dimensional spectral data cube I 1 (x, y, λ) to the frequency domain, and remove the blocking effect by suppressing the components of the frequencies in the specified region, obtaining the three-dimensional spectral data cube I 2 (x, y, λ); (4b) Normalize each spectral image in the three-dimensional spectral data cube I 2 (x, y, λ) individually to obtain the three-dimensional spectral data cube I 3 (x, y, λ); Step (5): Derive the three-dimensional spectral data cube I RGB from the image pixel ratio relationship in the original RGB image coordinates C 3 (x, y). Update the three-dimensional spectral data cube I with the image pixel ratio relationship of each spectral image in 3 (x, y, λ) to obtain the three-dimensional spectral data cube I 4 (x, y, λ).

2. A pixel-level calibration and correction method based on a diamond-arranged DMD according to claim 1, characterized in that, in the step (1), the non-aliasing coding template uses a horizontal stripe coding template, where the stripe interval T is greater than the dispersion distance S of the sampling points, meeting the requirement of no dispersion aliasing. The size of the coding template is M×N, and the first template pattern is (nT + 1)×N, n = 0, 1,..., M / T - 1. By moving the horizontal stripe coding template down row by row, the coding template pattern is changed. The number of templates of the non-aliasing coding template pattern is equal to the stripe interval distance, and the changing template patterns are successively (nT + p)×N, n = 0, 1,..., M / T, p = 1, 2,..., T, where n is the number of stripes and p is the stripe offset distance.

3. A pixel-level calibration and correction method based on a diamond-arranged DMD according to claim 1, characterized in that, the dual channels of the diamond-arranged DMD use a dual-channel multi-spectral video imaging system of the diamond-arranged DMD. The DMD is used for spatial encoding and beam splitting simultaneously, and two complementary sets of spatially encoded images are generated according to the coding pattern. One set passes through a dispersion prism and is dispersed onto a black-and-white detector, and the other set is directly obtained by a color detector.

4. A pixel-level calibration and correction method based on a diamond-arranged DMD according to claim 1, characterized in that, In the step (2a): for different λ, change the horizontal stripe encoding template one by one, and the changed template patterns are (nT + p)×N in sequence, where n = 0, 1,..., M / T, p = 1, 2,..., T, and record I 1 (x, y, λ) = C λ (x, y)*P λ (x, y); where n is the number of stripes and p is the stripe offset distance.

5. A pixel-level calibration and correction method based on a diamond-arranged DMD according to claim 1, characterized in that, In the said step (2b), according to each pixel D(x, y) on the DMD and the coordinates C RGB (x, y) in the detector and the point spread function P RGB (x, y), the scene RGB image I RGB互补 (x, y) corresponding to the complementary sampling points is restored, and meanwhile, the RGB background image I RGB背景 (x, y) obtained from the prior channel in the state where the template is not applicable is recorded; the spectral channel template pattern changes to (nT + p)×N, n = 0, 1,..., M / T, p = 1, 2,..., T, where n is the number of stripes and p is the stripe offset distance. The prior channel template pattern is (nT + q)×N, n = 0, 1,..., M / T, (q≠p)&&(q∈[1,T]); the number of templates of the aliasing coding template pattern is equal to the stripe interval distance, and the changing template patterns are successively (nT + p)×N, n = 0, 1,..., M / T, p = 1, 2,..., T; where n is the number of stripes and p is the stripe offset distance.

6. A pixel-level calibration and correction method based on a diamond-arranged DMD according to claim 5, wherein, In the step (3a): the pixel value information of each spectral band of the sampling points in the multispectral image, i.e., the value of I 1 1 (x, y, λ), where x and y are the known sampling points designed by the template; the fitting formula is: where λ 总 is the total number of wavelengths.

7. A pixel-level calibration and correction method based on a diamond-arranged DMD according to claim 5, wherein, the specific step (3b) is: Fill the RGB pixel values corresponding to the sampled points after fitting into the scene RGB image I captured in (2b). RGB互补 In (x, y), complete the filling of the missing points in the RGB image, and the calculated I R (x, y), I G (x, y), I B Fill (x, y) into I RGB互补 (x, y, λ), and obtain I RGB (x, y, λ).

8. A pixel-level calibration and correction method based on a diamond-arranged DMD according to claim 5, wherein, the derivation process in the step (5) is: According to the RGB background image I obtained in (2b) RGB背景 (x, y) and I obtained in (3b) RGB (x, y, λ) calculates the pixel change rate, and combines the three-dimensional spectral data cube I 3 (x, y, λ), the pixel ratio relationship of each spectral image in it, to obtain the three-dimensional spectral data cube I 4 (x, y, λ), 9. A pixel-level calibration and correction method based on a diamond-arranged DMD according to claim 1, wherein, in the step (4b), the formula for normalization is:

Images