A method for atlas processing based on binocular stereo interference imaging spectrometer

Abstract

The present application relates to the technical field of spectral instrument, especially to a kind of atlas processing method based on binocular stereo interferometric imaging spectrometer, binocular stereo interferometric imaging spectrometer acquires multidimensional interference data cube, the atlas processing method includes decoupling process and reconstruction process;The decoupling process includes multidimensional data decoupling process;The reconstruction process includes three-dimensional space information reconstruction process and spectral information reconstruction process;The atlas processing method provided by the present application can realize high-precision information decoupling and reconstruction processing of multidimensional interference data cube, realize high-precision detection and analysis to target three-dimensional space and spectral multidimensional information.

Description

Technical Field

[0001] This invention relates to the field of spectroscopic instrument technology, and in particular to a spectral processing method, device, and computer-readable storage medium based on a binocular stereo interferometric imaging spectrometer. Background Technology

[0002] In recent years, with the continuous improvement of social and technological development, the demand for multi-dimensional, high-performance imaging detection instruments in important fields such as military reconnaissance, agricultural remote sensing, environmental monitoring, resource exploration, and space exploration has become increasingly urgent. However, traditional imaging detection technologies often only acquire two-dimensional spatial information of a target in a specific wavelength band, resulting in a relatively singular information acquisition dimension, which is insufficient to meet the application needs of many fields for comprehensive multi-dimensional and multi-angle descriptions of targets. For example, in the field of agricultural remote sensing, imaging technology typically only meets the needs of simple agricultural monitoring, and is insufficient to meet the application needs of estimating the sugar content, temperature, and humidity of agricultural products, fine crop classification and identification, and yield estimation. In the field of military reconnaissance, due to the development of camouflage technology towards multi-functional and multi-spectral stealth, targets and backgrounds can be nearly "of the same color and spectrum" within a certain wavelength range, making traditional thermal infrared imaging reconnaissance methods difficult to effectively identify. In the field of environmental monitoring, commonly used telemetry methods can usually only estimate the two-dimensional spatial distribution of pollutant column concentrations, and cannot obtain the overall morphology of the air mass, thus failing to calculate the total amount of pollutants.

[0003] As a novel multidimensional imaging detection technology, the binocular stereo interferometric imaging spectrometer organically integrates binocular stereo vision imaging technology and static interferometric imaging spectroscopy technology. It can simultaneously acquire four-dimensional information—three-dimensional spatial information and spectral information—of the target, enabling steady-state collaborative analysis of the target's three-dimensional morphology and composition. Currently, conventional interferometric data processing methods mainly include image preprocessing, interferometric information demodulation, and spectral reconstruction, which can meet basic spectral reconstruction requirements. However, the binocular stereo interferometric imaging spectroscopic detection system obtains a multidimensional interferometric data cube that couples three-dimensional spatial information and spectral information. The data structure is complex, and there is data complementarity between different dimensions, making it difficult for general data processing methods to meet the system's application requirements.

[0004] The binocular stereo interferometric imaging spectral detection system organically integrates binocular stereo vision imaging technology and static interferometric imaging spectral technology. It can simultaneously acquire four-dimensional information—three-dimensional spatial information and spectral information—of the target, enabling steady-state collaborative analysis of the target's three-dimensional morphology and composition. Unlike traditional imaging and spectral detection instruments, this system obtains a multi-dimensional interferometric data cube containing both three-dimensional spatial and spectral information of the target. The data structure and processing are more complex, and there is data complementarity between different information dimensions. Currently, conventional interferometric data processing methods can meet basic image reconstruction needs, but their application to this system has certain limitations and drawbacks. For example, infrared images have low signal-to-noise ratios; conventional edge detection operators in image processing rely on prior knowledge for threshold settings, which may lead to the loss of some image edge information and unsatisfactory detection results; conventional feature point detection algorithms suffer from mismatch problems under environmental influences; semi-global matching algorithms have problems such as mismatch, view occlusion, and susceptibility to noise interference; and the Delaunay triangular mesh reconstruction algorithm is prone to hole defects during three-dimensional reconstruction and has difficulty handling large point cloud datasets. Summary of the Invention

[0005] To address the aforementioned problems, this invention provides a spectral processing method suitable for binocular stereo interferometric imaging spectral detection systems, enabling high-precision detection and analysis of multi-dimensional spatial and spectral information of targets.

[0006] This invention provides a spectral processing method based on a binocular stereo interferometric imaging spectrometer, wherein the binocular stereo interferometric imaging spectrometer acquires a multidimensional interferometric data cube, and the spectral processing method includes a decoupling process and a reconstruction process; The decoupling process includes a multi-dimensional data decoupling process; the reconstruction process includes a three-dimensional spatial information reconstruction process and a spectral information reconstruction process. The multidimensional data decoupling process includes the following steps: S11. Perform interference image unit segmentation and extraction on the multidimensional interference data cube; S12. Perform optical path difference matching based on the target contrast characteristics in the interference image unit; S13. Perform feature registration on the interferometric image units after optical path difference matching to obtain the decoupled binocular stereo image sequence and two-dimensional interferometric primitive sequence. S14. Perform dimensionality reduction processing on the two-dimensional interference primitive sequence to obtain a one-dimensional interference intensity sequence; The three-dimensional spatial information reconstruction process includes the following steps: S31. Establish a nonlinear transformation model between the pixel coordinate system of the system image and the world coordinate system of the real scene where the target image is located, and correct the radial and tangential distortion of the system. S32. By analyzing the grayscale features of the target image, a precise matching algorithm for binocular channel images is designed. Combined with the enhancement of the texture details of the target image, high-precision stereo matching of the target image is achieved. S33. Combine the nonlinear transformation model to calculate the matching point features, reconstruct the dense three-dimensional point cloud coordinates of the target, and realize high-precision measurement of the three-dimensional shape of the target. The spectral information reconstruction process includes the following steps: S41. Establish the correspondence between the one-dimensional interference intensity sequence and the optical path difference distribution in each channel; S42. Perform interference data processing on the one-dimensional interference intensity sequence; S43. Perform spectral analysis on the one-dimensional interference intensity sequence to obtain the one-dimensional spectral information of each parameter in the frequency domain, thereby realizing the spectral composition analysis of the target.

[0007] Preferably, the binocular stereo interferometric imaging spectrometer includes a scanning mirror, a collimating mirror, a binocular mirror, an aperture array, a multiple imaging lens array, a static interferometric system, a relay imaging system, and an area array detector. The scanning mirror is used to scan the target under test to obtain the initial light field information of the dual field of view of the target under test; the collimating mirror is used to collimate the dual field of view parallel beam; The binocular reflector divides the dual-field parallel beam into a first imaging channel and a second imaging channel to form a binocular field imaging channel. The aperture array and the multiple imaging lens array are used to perform aperture segmentation on the light field of the binocular field imaging channel to achieve binocular multiple stereo imaging. The static interference system is used to spatially segment and phase-modulate the target light field of each imaging channel to form a binocular multi-dimensional interference image field. The relay imaging system is used to couple the target light field in each imaging channel to the area array detector to form a binocular multi-dimensional interferometric image array with different states.

[0008] Preferably, the segmentation and extraction of interferometric image units in S11 of the multidimensional interferometric data cube includes: based on the grayscale distribution differences between image units of different interferometric levels, using an image unit edge detection method to segment and extract interferometric image units from the multidimensional interferometric data cube.

[0009] Preferably, the nonlinear transformation model between the pixel coordinate system of the system image and the world coordinate system of the real scene where the target image is located in step S31 includes: through system imaging calibration, analyzing the binocular parallax imaging parameters of the system, and combining the known parameters of the target, establishing a nonlinear transformation model between the pixel coordinate system of the system image and the world coordinate system of the real scene where the target image is located; the known parameters include the shape of the target and the position of the target.

[0010] Preferably, establishing the correspondence between the one-dimensional interference intensity sequence and the optical path difference distribution in each channel in step S41 includes: establishing the correspondence between the one-dimensional interference intensity sequence and the optical path difference distribution in each channel by combining the system interference modulation parameters.

[0011] Preferably, the interference data processing of the one-dimensional interference intensity sequence in S42 includes: baseline correction, apodization, and phase error correction of the one-dimensional interference intensity sequence.

[0012] Preferably, the spectral calculation of the one-dimensional interference intensity sequence in step S43 includes: performing spectral calculation of the one-dimensional interference intensity sequence through discrete Fourier transform.

[0013] The present invention also provides a computer device, comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the spectral processing method based on a binocular stereo interferometric imaging spectrometer of the present invention.

[0014] The present invention also provides a non-transient computer-readable storage medium storing computer instructions for causing the computer to execute the spectral processing method based on a binocular stereo interferometric imaging spectrometer of the present invention.

[0015] Compared with the prior art, the present invention can achieve the following beneficial effects: The spectral processing method provided by this invention can achieve high-precision information decoupling and reconstruction processing of multidimensional interferometric data cubes, enabling high-precision detection and analysis of multidimensional information of the target's three-dimensional space and spectrum. Attached Figure Description

[0016] Figure 1 This is a schematic flowchart of a spectral processing method based on a binocular stereo interferometric imaging spectrometer according to a specific embodiment of the present invention; Figure 2This is a schematic diagram of the decoupling process in the spectral processing method based on a binocular stereo interferometric imaging spectrometer according to a specific embodiment of the present invention; Figure 3 This is a schematic diagram of the three-dimensional spatial information reconstruction process in the spectral processing method based on a binocular stereo interferometric imaging spectrometer according to a specific embodiment of the present invention; Figure 4 This is a schematic diagram of the spectral information reconstruction process in the spectral processing method based on a binocular stereo interferometric imaging spectrometer according to a specific embodiment of the present invention; Figure 5 This is a schematic diagram of the optical path structure of a binocular stereo interferometric imaging spectrometer according to a specific embodiment of the present invention; Figure 6 This is a schematic diagram of a multidimensional interference data cube acquired by a binocular stereo interferometric imaging spectrometer according to a specific embodiment of the present invention; Figure 7 This is a schematic diagram of the process of edge detection of image units according to a specific embodiment of the present invention; Figure 8 This is a schematic diagram of an image unit obtained by segmenting and extracting an interferometric image unit according to a specific embodiment of the present invention; Figure 9 This is a schematic diagram of the process for obtaining a binocular image sequence according to a specific embodiment of the present invention; Figure 10 This is a schematic diagram of the decoupled interference image sequence obtained according to a specific embodiment of the present invention; Figure 11 This is a schematic diagram of the target left field of view panoramic image obtained according to a specific embodiment of the present invention; Figure 12 This is a schematic diagram of the target right field of view panoramic image obtained according to a specific embodiment of the present invention; Figure 13 This is a schematic diagram of the process of obtaining a disparity map by binocular stereo matching according to a specific embodiment of the present invention; Figure 14 This is a flowchart illustrating the process of achieving high-precision measurement of the three-dimensional shape of a target according to a specific embodiment of the present invention; Figure 15 This is a schematic diagram illustrating the high-precision measurement of the three-dimensional shape of a target according to a specific embodiment of the present invention; Figure 16 This is a schematic diagram of obtaining one-dimensional spectral information of each parameter in the frequency domain according to a specific embodiment of the present invention; Figure 17 This is a structural block diagram of a computer device according to a specific embodiment of the present invention.

[0017] Figure label: 1 Scanning mirror, 2 Collimating mirror, 3 Binocular mirror, 4 Aperture array, 5 Multiple imaging lens array, 6 Beam splitter, 7 Symmetrical multi-stage micromirror, 8 Step-by-step micromirror, 9 Relay imaging system, 10 Area array detector. Detailed Implementation

[0018] In the following description, embodiments of the invention will be described with reference to the accompanying drawings. In the description below, the same modules are denoted by the same reference numerals. Where the same reference numerals are used, their names and functions are also the same. Therefore, their detailed description will not be repeated.

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and do not constitute a limitation thereof.

[0020] This invention provides a spectral processing method based on a binocular stereo interferometric imaging spectrometer, which can achieve high-precision information decoupling and reconstruction processing of multidimensional interferometric data cubes. The specific flow of the spectral processing method is as follows: Figure 1 As shown in the figure, the spectral processing method includes a decoupling process and a reconstruction process; the decoupling process includes a multidimensional data decoupling process; the reconstruction process includes a three-dimensional spatial information reconstruction process and a spectral information reconstruction process; and a multidimensional interference data cube is acquired through a binocular stereo interferometric imaging spectrometer.

[0021] In a specific implementation, the multidimensional data decoupling process includes the following steps: S11. Perform interference image unit segmentation and extraction on the multidimensional interference data cube; S12. Perform optical path difference matching based on the target contrast characteristics in the interference image unit; S13. Perform feature registration on the interferometric image units after optical path difference matching to obtain the decoupled binocular stereo image sequence and two-dimensional interferometric primitive sequence. S14. The two-dimensional interference primitive sequence is reduced in dimensionality to obtain a one-dimensional interference intensity sequence.

[0022] In specific implementation methods, such as Figure 2The diagram illustrates the multidimensional data decoupling process. As can be seen, the process includes: First, based on the grayscale distribution differences between image units of different interference levels, a suitable image unit edge detection method is used to segment and extract interference image units from the multidimensional interference data cube. Then, optical path difference matching is performed based on the target contrast characteristics within each unit. Finally, feature registration is performed on the matched image units to obtain the decoupled binocular image sequence and interference primitive sequence. Further dimensionality reduction of the two-dimensional interference primitive sequence yields a one-dimensional interference intensity sequence for subsequent spectral reconstruction.

[0023] In the process of decoupling the multidimensional interferometric data cube acquired by a binocular stereo interferometric imaging spectral detection system, this invention addresses the problems of high background, low contrast, and low signal-to-noise ratio in infrared images. It integrates the Canny algorithm based on mathematical morphology and the adaptive thresholding method based on energy optimization to perform edge detection on infrared interferometric images. This method can clearly and completely extract edge contours, adaptively smooth redundant information, highlight key edge information, and accurately detect the contour information of interferometric image units. Furthermore, to address the mismatch problem that occurs in conventional feature point detection algorithms under environmental influences, a neural network optimization detection algorithm is applied to statistically identify mismatched points. Then, the RANSAC algorithm is applied to fit accurate matching points from these mismatched points, improving matching accuracy.

[0024] In a specific implementation, the three-dimensional spatial information reconstruction process includes the following steps: S31. Establish a nonlinear transformation model between the pixel coordinate system of the system image and the world coordinate system of the real scene where the target image is located, and correct the radial and tangential distortion of the system. S32. By analyzing the grayscale features of the target image, a precise matching algorithm for binocular channel images is designed. Combined with the enhancement of the texture details of the target image, high-precision stereo matching of the target image is achieved. S33. Combine the nonlinear transformation model to calculate the matching point features, reconstruct the dense three-dimensional point cloud coordinates of the target, and realize high-precision measurement of the three-dimensional shape of the target.

[0025] In specific implementation methods, such as Figure 3The diagram illustrates the 3D spatial information reconstruction process. As shown, the process includes: First, system imaging calibration is performed. The binocular parallax imaging parameters are analyzed, and combined with known parameters such as the shape and position of the calibration object, a nonlinear transformation model is established between the pixel coordinate system of the system image and the world coordinate system of the real scene where the target is located, correcting radial and tangential distortions. Then, by analyzing the grayscale features of the target image, a precise matching algorithm for the binocular channel images is designed, combined with image texture feature enhancement, to achieve high-precision stereo matching of the target binocular images. Finally, the matching point features are calculated using the coordinate system transformation model to reconstruct the dense 3D point cloud coordinates of the target, achieving high-precision measurement of the target's 3D shape.

[0026] In the process of reconstructing the three-dimensional spatial information of a multidimensional interferometric data cube acquired by a binocular stereo interferometric imaging spectral detection system, this invention addresses the problems of mismatch, view occlusion, and susceptibility to noise interference inherent in semi-global matching algorithms. During cost calculation, it integrates Gensus transform and filtering windows to reduce the impact of noise on the grayscale value of the center pixel, making the cost calculation results more accurate. During cost aggregation, it combines a mean shift algorithm and an energy function to accurately segment different regions of the image and applies different penalty coefficients to different regions, effectively reducing the mismatch rate in weak texture and occluded areas during aggregation. In the disparity calculation process, disparity is optimized: first, a left-right consistency detection method is applied to identify mismatches; then, an adaptive window method is applied to fill holes and distortion points on the disparity map; finally, median filtering is used to obtain a smooth disparity image. To address the issues of holes and poor handling of large point cloud datasets in the Deloitte triangulation reconstruction algorithm during 3D reconstruction, this paper first applies a point cloud preprocessing algorithm to simplify the point cloud data and improve the accuracy of the 3D reconstruction algorithm: denoising is performed using the Laplacian algorithm and bilateral filtering algorithm, and improved adaptive downsampling based on the feature points near the centroid of voxels is used to simplify the point cloud data; then, the point cloud space is partitioned into octrees based on the density of the point cloud data; finally, the region growing method is applied to reconstruct the mesh and perform smoothing processing to achieve high-precision measurement of the target's 3D shape.

[0027] In a specific implementation, the spectral information reconstruction process includes the following steps: S41. Establish the correspondence between the one-dimensional interference intensity sequence and the optical path difference distribution in each channel; S42. Perform interference data processing on the one-dimensional interference intensity sequence; S43. Perform spectral analysis on the one-dimensional interference intensity sequence to obtain the one-dimensional spectral information of each parameter in the frequency domain, thereby realizing the spectral composition analysis of the target.

[0028] In specific implementation methods, such as Figure 4The diagram shows a schematic of the spectral information reconstruction process. As can be seen from the diagram, the spectral information reconstruction process includes: First, establishing the correspondence between the interference intensity sequence of each channel and the optical path difference distribution by combining the system interference modulation parameters; then, performing interference data processing such as baseline correction, apodization, and phase error correction on the interference intensity sequence to improve the interference sampling accuracy; finally, performing spectral calculation on the interference intensity sequence through discrete Fourier transform (DFT) to obtain one-dimensional spectral information of each parameter in the frequency domain, thereby realizing the spectral composition analysis of the target.

[0029] In specific implementation methods, such as Figure 5 The diagram shows the optical path structure of a binocular stereo interferometric imaging spectrometer. As can be seen from the diagram, the binocular stereo interferometric imaging spectrometer includes a scanning mirror 1, a collimating mirror 2, a binocular mirror 3, an aperture array 4, a multiple imaging lens array 5, a beam splitter 6, a symmetrical multi-stage tiered mirror (2N sub-tiers arranged laterally) 7, a step-type micro-mirror 8 (4 sub-tiers arranged vertically), a relay imaging mirror 9, and an area array detector 10. The scanning mirror 1 scans the target to obtain the initial light field information of the two fields of view of the target. The collimating mirror 2 collimates the beams to obtain two parallel beams of different axes. The binocular mirror 3 then obtains a coaxial parallel beam. Finally, the aperture array 4 and the multiple imaging lens array 5 divide the light field into M independent imaging channels, achieving multiple array imaging of the target on its image-side focal plane. Then, beam splitter 6 divides the target image field intensity equally and incident it onto the symmetrical multi-stage micromirror 7 and the step-index micromirror 8 respectively, and performs multi-dimensional phase modulation on the target image field to form M There are N interference channels, each of which corresponds to an imaging channel. Finally, the target light field in each channel is coupled to the area array detector 10 via the relay imaging system 9 to form an interferometric image array of the target.

[0030] In a specific implementation, the decoupling process of the multidimensional interferometric data cube based on the binocular stereo interferometric imaging spectral detection system of the present invention includes the following steps: 1. Obtain a multidimensional interferometric data cube based on a binocular stereo interferometric imaging spectral detection system, specifically as follows: Figure 6 As shown; 2. Based on the grayscale distribution differences between image units of different interference orders, edge detection is performed on the interferometric image of the multidimensional interferometric data cube. The edge detection process is as follows: Figure 7 As shown.

[0031] Edge detection is performed on the interferometric image of the multidimensional interferometric data cube to obtain interferometric image units for subsequent image unit segmentation. The grayscale value of the target region in the infrared image is greater than the background grayscale value, characterized by high background and low contrast, and the signal-to-noise ratio is also lower than that of the visible light image, making edge detection in infrared images quite difficult. Although traditional edge detection algorithms are simple in image processing and have low computational complexity, the detection results have obvious defects, such as poor image smoothing, insufficient denoising, and the influence of redundant information during detection. To clearly and completely extract edge contours, adaptively smooth redundant information, highlight key edge information, and accurately detect the contour information of interferometric image units, this invention proposes an infrared image edge detection method that integrates the Canny algorithm based on mathematical morphology and the adaptive thresholding method based on energy optimization. Its main steps include morphological filtering, initial edge extraction of the infrared image target, construction of the initial edge expectation gradient field, target edge detection based on energy optimization, and target edge enhancement processing.

[0032] Morphological filtering of the interferometric image: (1) Apply two structural elements and Perform filtering operations on the interference image. and The size is and , To smooth the noise reduction results, The scale size is greater than , Its noise reduction capability is greater than ,but The ability to preserve edge details is greater than .

[0033] Applying two morphological structuring elements to perform opening and closing filtering operations on the image not only effectively removes noise but also better preserves detail information.

[0034] The gradient magnitude and direction of the interferometric image are calculated using the first-order finite difference method. The convolution model used in the calculation by the first-order finite difference method is as follows: (2) (3) The first-order finite difference method was used to calculate the result at... , The partial derivative matrices in both directions are: (4) (5) Represents pixels in an interferometric image The gradient magnitude, the magnitude of which characterizes the edge strength of a pixel. Represents the pixels of the interference image The gradient direction, and the magnitude of this value characterizes the normal vector of the pixel. Their expressions are: (6) (7) According to the Canny algorithm, non-maximum suppression is applied to the gradient magnitude of the interference image. By comparing adjacent pixels on the interference image, the local maxima of the interference image are found. Then, the gray values ​​of all points that do not belong to the maxima are set to zero, so as to filter out most of the non-edge points.

[0035] The Otsu algorithm is used to obtain the adaptive threshold for the interferometric image. Let the segmentation threshold for the foreground and background of the interferometric image be... The proportion of foreground and background points in the interferometric image is: and The average gray level of the foreground and background images is and The total average gray level of the interference image is: (8) The variances of the foreground and background images in the interference image are: (9) When the variance When the grayscale level reaches its maximum, that is, when the difference between the foreground and background images of the interference image is maximized, the grayscale value at this point is... The optimal threshold is [value].

[0036] (10) The edges of the interference image are thinned by non-maximum suppression gradient image, then thresholded, and finally the pixels are connected to obtain the initial edge of the interference image.

[0037] Construct the desired gradient field of the interferometric image, and set... , The gradient in the direction is , The desired gradient fields in the horizontal and vertical directions are: and Then we have: (11) (12) in , Indicates the edge length and direction of a pixel. As a monotonically non-decreasing function, this process adaptively filters out detailed information from the initial edges of the interference image, and can be expressed as: (13) (14) The level at which details in the interference image are smoothed and the level at which details around the edges are enhanced are affected by variance. and mean Limitations, subtle details will follow The increase was smoothed out when When the value is large, larger details will receive greater exposure.

[0038] To minimize the energy function and avoid excessive differences between the final edge detection result of the interferometric image and the input interferometric image, the energy function is: (15) For the output value, , The cost function represents the cost of color values ​​and gradient values, when At that time, it can be used to balance the two sub-energy terms of the cost function for color value and gradient value. and Defined as: (16) (17) , Indicates the output image , gradient of direction, Let this be a constraint condition for color values. , , This represents the desired gradient field of the acquired interferometric image. The weighting function for weighted processing of color values ​​and gradient constraints is: (18) (19) If it appears , reduce , When the value is too large, the influence of the weight reduction on the output image can be reduced. By solving the energy function and the weighting function, the target edge detection result that is closest to the original interferometric image can be obtained.

[0039] Based on the smoothed and enhanced edge detection image obtained by solving the energy function of the interferometric image, enhancement processing is performed to make the target edge contour features more prominent. The enhancement amplitude is: (20) (twenty one) , This is represented by continuous long edges and shorter edges. Represents a 2-bit Gaussian kernel, pixels Pointed The maximum value in the domain is .

[0040] The final output interferometric image shows the target contour edges as follows: (twenty two) 3. After edge detection, the edge detection length of the interferometric image is controlled by the imaging channel length. Image segmentation is then performed on each interferometric image to obtain interferometric image units, specifically as follows: Figure 8 As shown.

[0041] 4. In each interferometric image of the multidimensional interferometric data cube, interferometric image units of the same order are feature-matched according to the scanning direction, and then the images are stitched together to obtain a binocular image sequence. The process is as follows: Figure 9 As shown; Interferometric image units of the same order in each image of the interferometric data cube are used as image units to be registered. The ORB fast feature point detection and extraction algorithm based on grid statistics is combined. The ORB algorithm includes two algorithms: O-FAST corner detection algorithm and R-BRIEF algorithm. Then, the RANSAC algorithm is applied for post-processing to remove mismatches.

[0042] First, corner detection is performed. If a pixel's attributes differ from a sufficient number of its surrounding pixels, then this pixel is called a FAST corner. The main processing steps are: based on pixel... With the center as the center, Draw a circle with a radius of 10 ... Then use the points on the center and the edge of the circle. The grayscale value represents the dissimilarity of pixels. A threshold is defined. , Point and The number of points that are dissimilar is : (twenty three) in, This represents the grayscale value of a pixel.

[0043] when At that time, The point is a corner point, when At that time, generally take The value is 9 or 12. In application, comparing all pixels on the edge of a circle is extremely inefficient. Therefore, the FAST algorithm is used to quickly eliminate all non-corner points, comparing only the four symmetrical pixels on the top, bottom, left, and right sides of the circle. If at least one of the two symmetrical points on the top and bottom of the circle is not a disjoint point, then the two symmetrical points on the left and right sides are analyzed. If at least two of these four points are disjoint points, then the point... For non-corner points, this method can eliminate most non-corner points and obtain preliminary corner point information. Non-maximum constraints are used to remove redundant corner points, but the resulting corner points do not possess multi-scale invariance. Therefore, a pyramid scaling model is applied, using Harris filtering at each level of the pyramid to generate keypoints, resulting in scale-invariant FAST corner points. Then, centroid calculation is used to obtain the principal direction of the feature points, making the corner points direction-invariant. The expression for the principal direction is: (twenty four) in, Grayscale value and The formula for the centroid of the interference image, where 0 or 1, is: , (25) in, This indicates the centroid coordinates, which are the coordinate axes of the image.

[0044] At feature points Neighborhood selected according to Gaussian distribution Pairs of dots form binary bits, describing the operator. The length can be selected as 128, 256, 512, etc., based on feature points. Mark the points with the center of the circle as the reference point: , , wait.

[0045] (26) (27) in express The grayscale of the point. Sequentially... , , The operation yields a string of 0s and 1s describing the operator. A rotation factor is added to ensure rotational consistency. Using the centroid direction of corner detection as the feature point direction, then... within the area The set of points is: (28) in Individual Dimensional data.

[0046] Let the direction angle of the center be . Let the rotation matrix be... Then, a feature point with direction can be represented as: (29) The smoothness of the original feature points in the interference image leads to a large number of matching points in the neighborhood of the feature points during the matching process. The authenticity of the matching points can be determined by simply counting the matching points in the neighborhood. Therefore, the optimization algorithm of ensemble network statistics removes the incorrect matching points.

[0047] In the interferogram to be matched, after ORB detection, the surrounding areas of the correct matching points are the same, so there are many similar characteristics in the area. Conversely, the surrounding areas of the incorrect matching points are not the same, so there are few similar features in the area.

[0048] Determining similar features within a surrounding area requires an expression and a critical threshold. By statistically analyzing the distribution of matching points within the neighborhood, true and false matches can be distinguished. An approximate match can be represented using a pairwise binomial distribution. The number of points with neighborhood features The distribution is as follows: (30) in, This represents the probability of a correct feature point in the neighborhood region. This represents the probability of an erroneous feature point in the neighborhood region. It is calculated by taking the average number of feature points in the neighborhood of the matching point. By using a threshold The expression used to distinguish the accuracy of a match and to differentiate between true and false matches is: (31) The neural network optimization algorithm identifies erroneous matching points in the interferometric images to be matched. Next, the RANSAC algorithm is used to fit the most accurate matching feature points from a large dataset of erroneous matches. Then, randomly selected... Among the points, the point that fits best is found through continuous iteration.

[0049] Feature detection, filtering of mismatched points, and finding the optimal homography matrix. The RANSAC algorithm satisfies the condition of having the largest number of data points in a homography matrix. Using Normalizing a matrix requires solving eight linear equations. Each pair of points can be represented by two equations, requiring at least four pairs of matching points. The system of linear equations is expressed as follows: (32) in, Indicates the corner positions in the target interferometric image. Indicates the corner positions in the interferometric image to be matched. This is the scale parameter.

[0050] Feature point matching is performed on multidimensional interferometric images acquired from the same viewpoint. After matching, the interferometric image units of the same order under the same field of view are feature registered and stitched together according to the scanning direction to obtain a decoupled binocular image sequence.

[0051] 5. Interferometric image units of different orders containing the same target from each image of the multidimensional interferometric data cube under the same field of view are stitched together according to optical path difference order to obtain a decoupled interferometric image sequence, as shown below. Figure 10 As shown.

[0052] 6. The obtained two-dimensional interference primitive sequence is reduced in dimensionality to obtain a one-dimensional interference intensity sequence, which is then used for subsequent spectral reconstruction to complete the information decoupling process.

[0053] In a specific implementation, the spectral processing method based on a binocular stereo interferometric imaging spectrometer of the present invention includes the following three-dimensional spatial information reconstruction process: 1. Perform system imaging calibration, establish a nonlinear transformation model between the pixel coordinate system of the system image and the world coordinate system of the real scene where the target is located, correct the radial and tangential distortion of the system, and at the same time correct and constrain the image and establish a binocular geometric model. When using a spectrometer for imaging, four major coordinate systems are involved: the spectrometer coordinate system, the image coordinate system, the pixel coordinate system, and the world coordinate system. Different coordinates are used to represent different scenes and locations, and the coordinates need to be transformed into each other. At the same time, the transformation of coordinates also lays a good foundation for subsequent feature matching and reconstruction.

[0054] Coordinate systems can be equivalently transformed using matrices, and their relationships can be expressed mathematically using matrix theory. When a point is captured by an interferometer to obtain an interferometric image, it is equivalent to an angular transformation and a positional translation occurring in three dimensions. The spectrometer coordinate system is projected onto a two-dimensional plane, which is the imaging plane, and the coordinate system in which it lies is the image coordinate system.

[0055] Let the center of the spectrometer coordinate system be... coordinate system A straight line parallel to the optical axis, at any point in the real world. The specific coordinates of this point are , The intersection point formed by the line it lies on and the imaging plane is... Positional relationships after imaging The actual position can be expressed by matrix transformation as follows: (33) in, It is a square matrix with dimension 3. .

[0056] Based on the knowledge of matrix rotation, assume that by Shaft selection Obtain the matrix , Axis rotation Obtain the matrix , Axis rotation Obtain the matrix The change of angle in three dimensions can be represented by the product of three matrices: (34) in:

[0057] (35)

[0058] Point corresponding to projection point The coordinates of this point are The relationship between the two is as follows: (36) in Indicates the scaling factor. To detect the focal length of the system, The angle of inclination of the optical axis of the spectrometer relative to the horizontal plane.

[0059] The unit of an interference image is a pixel, so the pixel coordinate system represents the position of a projection point in three-dimensional space onto a two-dimensional plane. The unit of the image coordinate system is a length unit. Let the size of each pixel in the horizontal axis be . The size along the vertical axis is Let the position of the image be... The position of the pixel is The relationship between the two is as follows: (37) By rearranging the pixel position and image position according to equations (34), (36), and (37), we can obtain: (38) The first two terms on the right side of the equation are related to the spectrometer and represent its internal parameters. and Representing the extrinsic parameter matrix, we can obtain the following by reduction: (39) Therefore, given the intrinsic and extrinsic parameter matrices of the spectrometer, the conversion between pixel coordinates and actual coordinates can be performed.

[0060] The imaging process is limited by manufacturing technology and optical principles. Spectrometer imaging does not follow the ideal pinhole imaging principle, therefore, distortion exists. Ignoring distortion will introduce significant errors in the 3D point calculation during binocular positioning. Distortion is divided into radial and tangential distortion. Because distortion exists during imaging, both radial and tangential distortion should be considered when calibrating the spectrometer. Therefore, a nonlinear model is constructed, assuming... For the spectrometer coordinate system, The coordinate axis where the image is located. Let the coordinates of the imaging point be the coordinates of the spectrometer coordinate system. For the projection point without distortion, For the projection point under distortion conditions, the formulas for radial distortion and tangential distortion are: (40) (41) (42) (43) (44) in, and It is along direction and Radial deviation of direction, and It is along direction and Tangential deviation in direction. , , It is the radial deformation factor. , It is the tangential distortion coefficient, and its expression, after transformation, is: (45) (46) and For the projection point at direction and The deviation caused by direction.

[0061] When a spectrometer acquires images, factors such as manufacturing process and placement can cause distortion in the acquired images. Therefore, the acquired images must be corrected first to eliminate the effects of distortion.

[0062] In a binocular stereo interferometric imaging spectrometer system, the images acquired from the two fields of view have an epipolar geometric constraint relationship. Let any point... The imaging points corresponding to the left and right fields of view of the binocular system are and The three points together form a plane, and the line of intersection between the plane and the imaging position of the left field of view is... The straight line passes through the point It is called the point corresponding to The epipolar line, the line intersecting the plane and the imaging position of the right field of view is... The line passes through the point , for corresponding The polar line.

[0063] After the above calibration and image correction steps, the next step is to match the dual-field panoramic images of the target. The positional offset corresponding to the matching point needs to be calculated, which is the parallax. The distance can be calculated by applying the parallax, thus obtaining the third-dimensional information, and the true appearance of the target can be restored.

[0064] by The center of the spectrometer is shown as the left and right centers, and the distance between the left and right centers is the baseline. focal length is any point Imaging via dual fields of view and The two image points have the same ordinate and a distance of 0 between their x-coordinates. ,set up Click The perpendicular distance between the points is By the principle of similar triangles, we can obtain: (47) Therefore, we can conclude that: (48) Parallax is .

[0065] 2. Analyze the grayscale features of the target image and design a reasonable precise matching algorithm for binocular channel images.

[0066] The decoupled binocular image sequence is stitched together according to the scanning direction to obtain a dual-field-of-view panoramic image of the target's left and right fields of view, as shown below. Figure 11 and Figure 12 As shown, stereo matching of target dual-field panoramic images is performed by optimizing the SGM algorithm. Traditional SGM algorithms suffer from susceptibility to noise interference, high mismatch rates in low-texture regions, and view occlusion. Therefore, optimization is needed for stereo matching based on the SGM algorithm. Census transform fusion filtering window is applied during cost calculation, a drift algorithm is applied during cost aggregation, and a left-right consistency detection method is used to identify and remove mismatches. Then, an adaptive window is applied to fill and repair invalid points in the disparity map, resulting in higher accuracy of the disparity map. The flowchart is shown below. Figure 13 As shown.

[0067] Stereo matching based on the SGM algorithm begins with cost calculation. During the calculation, a non-parametric transformation—the Gensus transform—is applied. This transformation is not only relatively simple to implement and highly efficient compared to mutual information, but it also solves the problem of incorrect matching caused by illumination and is robust to noise to a certain extent.

[0068] When calculating the cost, the relative magnitudes of the grayscale values ​​of the pixel to be matched and its surrounding pixels are compared, as shown below: (49) Represents the coordinates of the point to be matched. This represents other points within a local area centered on the point to be matched. Point The image grayscale value, The grayscale value representing the surrounding points is determined by the relationship between the two grayscale values, taking values ​​of 0 or 1, resulting in a binary bit string, represented as: (50) in, Represented as a transform code formed after pixel comparison. Indicates a connector symbol. This indicates the horizontal range of the search, centered on the point to be matched, searching to the left and right. 1 pixel length This indicates the vertical range of the search, centered on the point to be matched, searching upwards and downwards. The cost is calculated using Hamming distance for a length of 1 pixel, expressed as follows: (51) While the above methods have certain advantages, their heavy reliance on the grayscale value of the center pixel has limitations. Once the center point is affected by noise, an incorrect transform code will be obtained, which will affect the cost calculation. Therefore, a filtering window is fused. The center of the window is used as the center pixel of the cost calculation region. A region smaller than the cost calculation region is calculated and selected as the median filtering window. The pixels within the filtering window are filtered, and then the center pixel of the region is filled with the median. Other pixels outside the filtering window are selected for transformation to obtain the transform code. This fusion of Gensus transform and filtering window reduces the impact of noise on the grayscale value of the center pixel, making the cost calculation result more accurate.

[0069] After the cost is calculated, cost aggregation is performed. In order to reduce the mismatch rate of repeated and weak texture regions during the aggregation process, the mean offset algorithm is applied.

[0070] Traditional SGM algorithms use a fixed penalty coefficient during cost aggregation. However, applying the mean-shift algorithm to accurately segment different regions of the image and assigning different penalty coefficients to different regions can effectively reduce the mismatch rate in weak texture and occluded areas during aggregation. Combining the mean-shift algorithm with the energy function yields the following energy function: (52) (53) For disparity maps, Let be the energy function. , Represents a pixel, and yes The surrounding pixels, Represents the cost function. represent disparity value of a point represent disparity value of a point , As a penalty factor, This is the decision function.

[0071] Next, we perform multi-path aggregation. Because the results differ significantly across different paths, multi-path aggregation is clearly superior to single-path aggregation. The path aggregation is as follows: (54) in, Represents the path, Represents the calculated cost value. Represents aggregated value. represent The value of the adjacent pixel on the left or right, The disparity of pixels within the path is The aggregate value of time The disparity of pixels within the path is The aggregate value of time The pixel difference within the path is The aggregate value of time The minimum cost value of all pixels within the path, representing the total cost value of the path. Represented as: (55) After the above cost calculation and cost fusion, the next step is to calculate disparity. The minimum cost is selected as the final cost, and the disparity corresponding to the final cost is taken as the optimal disparity. However, due to the influence of factors such as noise, lighting and weak texture, the results of disparity calculation may be biased. Therefore, disparity optimization should be performed, which mainly consists of three steps: eliminating incorrect matching points, filling occluded points and incorrect matching points, and refining the disparity map.

[0072] To eliminate mismatches, the left-right consistency detection method is first applied to identify mismatches. Assuming that the interferogram obtained from the left field of view in the binocular field of view is used as the reference to complete a stereo matching, the interferograms of the left and right fields of view are swapped, and another stereo matching is performed to obtain two disparity maps. If the disparity value of the same point in the left and right disparity maps is within a certain range, then the disparity value is unique and is the correct disparity; otherwise, it is an incorrect disparity and will be eliminated.

[0073] Because the imaging angles of the left and right fields of view are different, there will be occlusion points, which are also called holes. Therefore, these holes and the mismatched points eliminated mentioned above need to be filled. Thus, an adaptive window method is proposed to fill the holes and distorted points on the disparity map. The processing methods for these two types of invalid points are different, so the first step is to distinguish between these two types of points: (56) in, This represents the disparity value of a point on the left disparity map. This represents the disparity value of a point on the right disparity map. If this formula does not hold true and the disparity is not zero, it is a distorted point; otherwise, it is considered a hole point. As the distortion point, with Centered on a point, compare its disparity value with the surrounding 8 pixels, find the maximum and minimum disparity values, and fill the gap with the average of the two values: (57) (58) in, The set of disparity values ​​between adjacent points of distortion. The minimum disparity value is [value], and the maximum disparity value is [value]. The average disparity value. Fill the disparity at the distortion points with the minimum value. Distortion points are filled with the average value. Distortion points are filled with the maximum value.

[0074] if For the void point, from , Traverse in two directions to find points with non-zero disparity values. and , constructing a point Taking the coordinates of the origin as the starting point, For width, High matrix This forms a matrix that creates an adaptive window. : (59) The voids are filled as follows: (60) in, Representation matrix Sum of corresponding elements This is the average value. This represents the number of non-zero elements in the matrix. The process continues until all pixels have been traversed, and disparity optimization is performed. Finally, median filtering is used to obtain a smooth image.

[0075] In a specific implementation, a coordinate system transformation model is used to calculate matching point features, reconstruct the coordinates of the dense 3D point cloud of the target, and achieve high-precision measurement of the target's 3D shape. The process is as follows: Figure 14 As shown.

[0076] The initial point cloud is noisy and redundant, which can easily lead to low efficiency and poor accuracy of 3D reconstruction algorithms. Therefore, it is very important to simplify the point cloud data through point cloud preprocessing algorithms. First, the Laplacian algorithm and bilateral filtering algorithm are used for noise reduction. Then, the improved adaptive downsampling based on the feature points near the centroid of the voxel is used to simplify the point cloud data. Next, the octree method is applied to divide the space. Then, triangular mesh reconstruction is performed. Finally, the Laplacian algorithm is used to process the mesh to make the point cloud reconstruction model smooth.

[0077] First, the point cloud data is denoised, and then a Laplacian operation is performed on each 3D data point: (61) Let the coordinates of a certain point be... When applying the Laplace algorithm to solve the problem, it depends on... Neighboring points When determining the coordinates of moving data points, the following method should be used: (62) The Laplacian denoising method is applicable not only to uniformly distributed point cloud data but also to non-uniformly distributed point cloud data. The algorithm requires multiple iterations during operation, which can cause the coordinates of the original data points to shift. Therefore, it is combined with a bilateral filtering algorithm for further correction. By solving for the gray values ​​of neighboring points and weighting the gray values, the processing result is replaced to further correct the position of the sampling points, thereby achieving the effect of noise removal.

[0078] Next, the point cloud data is simplified. To achieve simplification while preserving fine features of the point cloud, a downsampling algorithm based on the feature points near the centroid of the voxels is adopted. However, to obtain the desired number of point clouds, the side lengths of the voxel grid need to be continuously adjusted, which not only complicates the operation but also consumes more time. Therefore, an adaptive downsampling method is applied. The specific simplification process is as follows: First, determine the side length of the voxel mesh and divide the point cloud into... A cube element: (63) in, Let be the side length of the voxel mesh. This indicates rounding up to the nearest integer.

[0079] Encode the voxel mesh and determine the spatial index of the mesh. : (64) Calculate the point cloud normal vector and establish the topological relationship of the point cloud data for any point in the point cloud. The k-dimensional tree algorithm is used to search for the number of neighboring points, and the covariance matrix is ​​obtained: (65) in, The number of point clouds in the neighborhood. Let the coordinates be the centroid coordinates of the neighboring points. The eigenvector corresponding to the smallest eigenvalue is The normal vector of a point.

[0080] Next, calculate any point in the point cloud. The arithmetic mean of the angle between the vector and the normal vector of its neighboring points: (66) The numerical value represents the degree of change in point cloud features; setting a reasonable threshold is important. ,when hour, For feature points, hour, For non-feature points, the final step is to traverse all spatial indices to find the feature point closest to the voxel centroid. This feature point is then used to replace all points in the voxel mesh, thus implementing the downsampling process. However, to avoid manually adjusting the voxel mesh side lengths, the downsampling method is optimized by applying an adaptive downsampling method to obtain the desired number of point clouds. (67) in, For voxel filtering functions, For the desired number of point clouds, Let the initial side length be , For the voxel grid growth step size, Adaptive growth factor: , (68) This represents the total number of points in the current point cloud. The total number of target point clouds.

[0081] After simplifying the point cloud data, the 3D space is partitioned into eight subspaces based on the point cloud density using an octree. Each subspace is then further divided into eight new cubes. By traversing the point cloud, the maximum and minimum values ​​in the 3D directions are obtained to determine the size of the partitioned space. Each cube corresponds to an octree node and is represented using octal encoding. Before performing octree partitioning, all point cloud data are first traversed to find the spatial coordinate system. The maximum and minimum values ​​of the coordinates determine the length, width, and height of the smallest space containing the original scene. : (69) Point cloud density is: (70) This represents the total number of point clouds. Surface area of ​​the cube The side length of the octree partitioned based on area density is: (71) After partitioning using an octree, the data is encoded. For a partitioning operation with a number of partitions... of In a three-dimensional space, the encoding of a child node is: (72) For any point in the cloud data in the octree, the octal value of its corresponding three-dimensional coordinate system is... and encoding for: (73) Binary encoding: (74) After dividing the point cloud data space, the next step is to apply the region growing method for mesh reconstruction. The process is as follows: First, construct a seed triangle, and then randomly select a point. Searching for points Distance less point set ,radius The average point distance of the point cloud is calculated to obtain: (75) From point set Select two points and , must meet , and Not on the same straight line, passing through , and The three points are not included in the sphere. any other point in the array.

[0082] After determining the seed triangle, new triangles are constructed based on one side of the seed triangle. Typically, a fixed radius is used for the search. However, if the point cloud data is uneven, holes can easily appear in areas of low point cloud density. Therefore, an adaptive radius method is applied for the search. (The last sentence appears to be incomplete and requires further context.) Construct new triangles for the edges, and search for all containing points within the triangle mesh. and points Calculate the length of all edges, then remove the longest and shortest edges. Take the average of all edge lengths as the search radius. midpoint With the center of the ball, Draw a sphere with radius . Points inside the sphere that are not gridded or points on the boundary are candidates for new triangles. Finally, select the best candidate point. The maximum empty circle principle can be applied to constrain the candidate points. The circumcircle of any triangle does not contain any points other than the vertices of the triangle, thus excluding the principle of multiple points sharing the same circle.

[0083] After mesh reconstruction, the Laplacian algorithm is applied for smoothing to achieve high-precision measurement of the target's 3D shape. Specifically, as follows... Figure 15 As shown.

[0084] In a specific implementation, the spectral information reconstruction process in the spectral processing method based on a binocular stereo interferometric imaging spectrometer of the present invention includes the following steps: 1. Based on the system's interference modulation parameters, establish the correspondence between the interference intensity sequence of each channel and the optical path difference distribution; Let the sub-step height of the symmetrical multi-stage micromirror be... The longitudinal first stage of symmetrical multi-stage micromirrors and step-index micromirrors Rank The optical path difference of the interference channel corresponding to the sub-step is: .

[0085] 2. Baseline correction, apodization, and phase error correction are performed on the one-dimensional interference intensity sequence to improve the interference sampling accuracy; Because spectrometers are affected by noise and background, baseline drift can occur. Therefore, an important preprocessing step for spectrometer spectral data is baseline correction. Polynomial iterative fitting can be used to continuously compare and adjust the original spectral data during the fitting process. The adjusted spectral data is then directly compared with the points on the fitted curve. The coefficients of the polynomial are gradually adjusted to gradually approach the actual baseline shape, making the calculated limit function form closer to the actual baseline.

[0086] Next, the baseline-corrected one-dimensional interference intensity sequence is apodized to eliminate possible sidelobe oscillations in the reconstructed spectrum: Phase correction is performed on the one-dimensional interference intensity sequence after apodization.

[0087] Phase correction was performed using the optimized Mertz algorithm, and the intensity distribution of the interferogram was as follows: (76) in, Indicates wave number, Indicates the corresponding position.

[0088] Suppose that a phase error occurs at a certain wavelength position as The phase error corresponding to different wavelengths is expressed as , … Equation (76) can be expressed as: (77) And there are: (78) The light intensity distribution after Fourier transform is as follows: (79) The spectral distribution including phase error is as follows: (80) The interference fringes are converted into a real sequence by sampling, and then the spectral data is obtained by FFT calculation. In the spectral reconstruction, the parameters of different phase errors are respectively substituted into the spectral function.

[0089] To optimize the Mertz algorithm described above, we first iteratively detect and find extreme points in the interference fringe data across all data points. The number of extreme points is determined by analyzing the number of characteristic wavelengths. Numerical non-linear calculations (NUFFTs) are then performed using these extreme points as midpoints to obtain the corresponding real and imaginary parts. The phase error is then calculated as follows: (81) The obtained interferometric data is interpolated and zero-padded to obtain the corrected phase error. Calculate its spectrum: (82) The one-dimensional interference intensity sequence is spectrally calculated using Discrete Fourier Transform (DFT) to obtain the one-dimensional spectral information of each parameter in the frequency domain, specifically as follows: Figure 16 As shown.

[0090] The spectral processing method provided by this invention can achieve high-precision information decoupling and reconstruction processing of multidimensional interferometric data cubes, enabling high-precision detection and analysis of multidimensional information of the target's three-dimensional space and spectrum.

[0091] Accordingly, according to embodiments of the present invention, the present invention also provides a computer device, a readable storage medium, and a computer program product.

[0092] Figure 17 This is a schematic diagram of the structure of a computer device 12 provided in an embodiment of the present invention. Figure 17 A block diagram of an exemplary computer device 12 suitable for implementing embodiments of the present invention is shown. Figure 17 The computer device 12 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.

[0093] like Figure 17As shown, computer device 12 is represented in the form of a general-purpose computing device. Computer device 12 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices can also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the invention described and / or claimed herein.

[0094] The components of the computer device 12 may include, but are not limited to: one or more processors or processing units 16, system memory 28, and bus 18 connecting different system components (including system memory 28 and processing unit 16).

[0095] Bus 18 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0096] Computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by computer device 12, including volatile and non-volatile media, removable and non-removable media.

[0097] System memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory. Computer device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 34 may be used to read and write non-removable, non-volatile magnetic media (…). Figure 17 Not shown; usually referred to as a "hard drive"). Although Figure 17 Not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk") and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 18 via one or more data media interfaces. Memory 28 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present invention.

[0098] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in memory 28. Such program modules 42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 42 typically perform the functions and / or methods described in the embodiments of the present invention.

[0099] Computer device 12 can also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), and with one or more devices that enable a user to interact with the computer device 12, and / or with any device that enables the computer device 12 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed through input / output (I / O) interface 22. Furthermore, computer device 12 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) through network adapter 20. Figure 17 As shown, network adapter 20 communicates with other modules of computer device 12 via bus 18. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with computer device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0100] The processing unit 16 executes various functional applications and data processing by running programs stored in the system memory 28, such as implementing the spectral processing method based on a binocular stereo interferometric imaging spectrometer provided in the embodiments of the present invention.

[0101] This invention also provides a non-transitory computer-readable storage medium storing computer instructions, on which a computer program is stored, wherein the program, when executed by a processor, is the spectral processing method based on a binocular stereo interferometric imaging spectrometer provided in all embodiments of this application.

[0102] The computer storage medium of this invention can be any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. More specific examples (a non-exhaustive list) of computer-readable storage media include: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0103] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0104] The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, optical fiber, RF, etc., or any suitable combination thereof. The computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0105] This invention also provides a computer program product, including a computer program that, when executed by a processor, implements the above-described spectral processing method based on a binocular stereo interferometric imaging spectrometer.

[0106] It should be understood that the various forms of processes shown above can be used to reorder, add, or delete steps. For example, the steps described in this invention disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this invention can be achieved, and this is not limited herein.

[0107] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A spectral processing method based on a binocular stereo interferometric imaging spectrometer, characterized in that, The binocular stereo interferometric imaging spectrometer acquires multidimensional interferometric data cubes, and the spectral processing method includes a decoupling process and a reconstruction process. The decoupling process includes a multi-dimensional data decoupling process; the reconstruction process includes a three-dimensional spatial information reconstruction process and a spectral information reconstruction process. The multidimensional data decoupling process includes the following steps: S11. Perform interference image unit segmentation and extraction on the multidimensional interference data cube; S12. Perform optical path difference matching based on the target contrast characteristics in the interference image unit; S13. Perform feature registration on the interferometric image units after optical path difference matching to obtain the decoupled binocular stereo image sequence and two-dimensional interferometric primitive sequence; S14. Perform dimensionality reduction processing on the two-dimensional interference primitive sequence to obtain a one-dimensional interference intensity sequence; The three-dimensional spatial information reconstruction process includes the following steps: S31. Establish a nonlinear transformation model between the pixel coordinate system of the system image and the world coordinate system of the real scene where the target image is located, and correct the radial and tangential distortion of the system; S32. By analyzing the grayscale features of the target image, a precise matching algorithm for binocular channel images is designed. Combined with the enhancement of the texture details of the target image, high-precision stereo matching of the target image is achieved. S33. Combine the nonlinear transformation model to calculate the matching point features, reconstruct the dense three-dimensional point cloud coordinates of the target, and realize high-precision measurement of the three-dimensional shape of the target; The spectral information reconstruction process includes the following steps: S41. Establish the correspondence between the one-dimensional interference intensity sequence and the optical path difference distribution in each channel; S42. Perform interference data processing on the one-dimensional interference intensity sequence; S43. Perform spectral analysis on the one-dimensional interference intensity sequence to obtain the one-dimensional spectral information of each parameter in the frequency domain, thereby realizing the spectral composition analysis of the target.

2. The spectral processing method based on a binocular stereo interferometric imaging spectrometer as described in claim 1, characterized in that, The binocular stereo interferometric imaging spectrometer includes a scanning mirror, a collimating mirror, a binocular mirror, an aperture array, a multiple imaging lens array, a static interferometric system, a relay imaging system, and an area array detector. The scanning mirror is used to scan the target under test to obtain the initial light field information of the dual field of view of the target under test; the collimating mirror is used to collimate the dual field of view parallel beam; The binocular reflector divides the dual-field parallel beam into a first imaging channel and a second imaging channel to form a binocular field imaging channel. The aperture array and the multiple imaging lens array are used to perform aperture segmentation on the light field of the binocular field imaging channel to achieve binocular multiple stereo imaging. The static interference system is used to spatially segment and phase-modulate the target light field of each imaging channel to form a binocular multi-dimensional interference image field. The relay imaging system is used to couple the target light field in each imaging channel to the area array detector to form a binocular multi-dimensional interferometric image array with different states.

3. The spectral processing method based on a binocular stereo interferometric imaging spectrometer as described in claim 1, characterized in that, The S11 step of segmenting and extracting interferometric image units from the multidimensional interferometric data cube includes: based on the grayscale distribution differences between image units of different interference levels, using an image unit edge detection method to segment and extract interferometric image units from the multidimensional interferometric data cube.

4. The spectral processing method based on a binocular stereo interferometric imaging spectrometer as described in claim 1, characterized in that, The nonlinear transformation model established in S31 between the pixel coordinate system of the system image and the world coordinate system of the real scene where the target image is located includes: through system imaging calibration, analyzing the binocular parallax imaging parameters of the system, and combining the known parameters of the target, establishing a nonlinear transformation model between the pixel coordinate system of the system image and the world coordinate system of the real scene where the target image is located; the known parameters include the shape of the target and the position of the target.

5. The spectral processing method based on a binocular stereo interferometric imaging spectrometer as described in claim 1, characterized in that, The establishment of the correspondence between the one-dimensional interference intensity sequence and the optical path difference distribution in each channel in S41 includes: combining the system interference modulation parameters to establish the correspondence between the one-dimensional interference intensity sequence and the optical path difference distribution in each channel.

6. The spectral processing method based on a binocular stereo interferometric imaging spectrometer as described in claim 1, characterized in that, The interference data processing in S42 for the one-dimensional interference intensity sequence includes: baseline correction, apodization, and phase error correction of the one-dimensional interference intensity sequence.

7. The spectral processing method based on a binocular stereo interferometric imaging spectrometer as described in claim 1, characterized in that, The spectral calculation of the one-dimensional interference intensity sequence in S43 includes: performing spectral calculation of the one-dimensional interference intensity sequence through discrete Fourier transform.

8. A computer device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, enables the at least one processor to perform the spectral processing method based on a binocular stereo interferometric imaging spectrometer as described in any one of claims 1 to 7.

9. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to execute the spectral processing method based on a binocular stereo interferometric imaging spectrometer as described in any one of claims 1 to 7.

Images