Multi-region compressed sensing imaging modulation matrix optimization method

Abstract

The invention relates to a multi-region compressed sensing imaging modulation matrix optimization method. According to the implementation mode, a coding matrix is only arranged at the position with a target to serve as a modulation mask, coefficients of 1 or-1 are given to different coding areas, coefficient sets are linearly independent, the optimized modulation matrix replaces a traditional full-coverage modulation Hadamard matrix to serve as an echo modulation mask, modulated echoes are received by a single-pixel detector, and the modulation efficiency is improved. Two-dimensional information of a target is reconstructed by using a compressed sensing algorithm, depth information is obtained by combining a distance measurement technology, and finally three-dimensional imaging is realized. The problem of spatial modulation redundancy is solved, the coefficient matrix is introduced, one-time imaging of multi-region scenes is supported, high-quality imaging can be completed only by using a low resolution and a small sampling rate, the detection efficiency of complex scenes such as multiple targets and small targets under large-scale terrains is remarkably improved, and the method is suitable for large-scale terrains and large-scale terrains. The method has potential application value in the fields of intelligent driving, high-resolution imaging, multi-target identification and the like.

Description

Technical fields:

[0001] This invention relates to multi-region imaging laser detection technology based on the principle of compressed sensing, which has potential application value in fields such as intelligent driving, high-resolution imaging, and multi-target recognition. Background technology:

[0002] Laser detection, a key technology for capturing single-point information of a target using a laser beam, has been widely applied in various fields such as topographic mapping and autonomous driving. Currently, laser detection methods are mainly divided into two categories: scanning and non-scanning. Scanning laser detection, due to its complex mechanical structure and limited imaging speed, struggles to meet the demands of high-speed, high-resolution imaging. While non-scanning laser detection can quickly acquire target information, its imaging resolution is limited by the size and density of the array detector, and it generates a large amount of data during the detection process. Single-pixel detectors possess advantages such as fast response speed and a wide usable wavelength range. Combining them with compressed sensing (CS) technology can overcome the limitations of the traditional Nyquist sampling theorem, achieving high-quality reconstruction solely by utilizing the sparsity of the signal. Replacing array detectors with single-pixel detectors significantly reduces the amount of data acquired and the storage requirements during detection, improving system efficiency and driving the rapid development of non-scanning lidar imaging, providing a new technological path for achieving high-resolution imaging.

[0003] In single-pixel imaging systems based on compressed sensing, a digital micromirror device (DMD) is typically used to spatially modulate the target echo using a preset modulation pattern to obtain the target's spatial information. Existing systems often employ a Hadamard matrix to generate a set of binary coded patterns, which are then loaded onto the DMD as a modulation mask to sequentially modulate the echoes of the target region, achieving complete and comprehensive encoding of that region. However, this encoding modulation method has a resolution limit, and when facing detection scenarios with a large area and few targets, it suffers from significant regional modulation redundancy. Furthermore, current technologies for multi-region scene detection typically employ a technique of multiple mask loading and block measurement. This method significantly increases the amount of data acquired, raising the system's storage and computational load. Moreover, the reconstructed sub-region images require stitching, which not only increases measurement complexity but also introduces calibration errors, leading to a decrease in depth information recovery accuracy. This fundamentally limits the improvement of imaging accuracy and makes it difficult to apply to large-scale scenes.

[0004] To address the aforementioned issues, this study proposes dividing the target terrain into blocks and setting modulation masks only at target locations. Based on matrices such as the Hadamard matrix that satisfy the finite isometry property (RIP property) to ensure the solvability of the compressed sensing equation, the overall observation matrix is ​​optimized by adding a coefficient matrix. This enables simultaneous scene reconstruction of multiple regions and multiple targets, achieving efficient encoding and sampling of the target scene. The new method supports simultaneous imaging of multi-region scenes, avoiding errors caused by multiple measurements and sub-image stitching, significantly improving the system's imaging resolution and scene adaptability, and solving the problem of modulation redundancy in target-free areas. This technology is expected to promote the application of lidar in high-speed, large-field-of-view imaging and has significant scientific research and technological application value. Summary of the Invention:

[0005] To adapt to detection scenarios with only multiple small targets in a large field of view, and to simultaneously complete scene reconstruction of multiple regions and multiple targets, this invention provides a method for optimizing the modulation matrix of multi-region compressed sensing imaging.

[0006] The technical solution adopted in this invention is as follows: The target location is initially determined using the lowest resolution and fewest samples. The target terrain is divided into blocks according to the target location. Modulation masks are only set at locations where the target is present. Each modulation mask is assigned a coefficient of 1 or -1, and the coefficient groups are linearly independent. This optimized modulation matrix replaces the traditional full-coverage modulation Hadamard matrix as the echo modulation mask. The modulated echo is received by a single-pixel detector. The two-dimensional information of the target is reconstructed using a compressed sensing algorithm. Combined with relevant ranging techniques, three-dimensional imaging can ultimately be achieved. Specific techniques are as follows:

[0007] Compressed sensing theory is an undersampling theory. The measurement process can be represented as y = Φx. After setting the observation matrix Φ, the original signal x can be reconstructed from the detected signal. To achieve multi-area detection, a basic observation matrix Φ0 is first constructed using the lowest resolution and the fewest modulation masks. The target location is then determined using the formula x = Φ'0y. The target terrain is represented as follows:

[0008]

[0009] x ij It is a sub-block, the size of which is determined by the size of the target, with the target location x. ij ≠0 (generally a square matrix), x is at the point where there is no target. ij =0 (can be a square matrix), assuming there are k targets in total, i.e., x ij The number of sub-blocks ≠ 0 is k. The modulation matrix corresponding to the sub-block size is set as follows:

[0010]

[0011] P ijLet x be a basis matrix. ij If the sub-block has a target, then P ij ≠0 (generally a square matrix), if the corresponding x ij If the sub-block has no target, then P ij =0 (can be a non-square matrix), P ij Let k be the number of ≠0, and P be the number of ≠0. ij For x ij Modulation matrices of the same size, different P ij It can accommodate different resolutions. Each sub-block x ij The corresponding baseline measurement value z ij The calculation is as follows:

[0012] z ij =Tr((P ij ) T x ij ⑶

[0013] To reconstruct the target terrain x, the compressed sensing equation y = Φx needs to have a unique solution. Therefore, the coefficient matrix C of the basis matrix P is designed as follows:

[0014]

[0015] c ij The coefficient is either 1 or -1. All valid sub-blocks that are not zero in the above formula are written in vector form: x = [x1 …x2] k ] T P = [P1 … P k ] T The compressed sensing reconstruction formula can be expanded as follows:

[0016]

[0017] The observation matrix Φ can be precisely represented as:

[0018]

[0019] Therefore, as long as the coefficient matrix C is guaranteed h If the regions are linearly independent, then x has a unique solution, and compressed sensing algorithms can be used to complete the one-time reconstruction of multiple regions.

[0020] Compared with the prior art, the beneficial effects of the present invention are:

[0021] (1) This invention solves the problem of spatial modulation redundancy, focuses only on the target area, and matches the size of the modulation matrix with the target size, so that the target area is encoded and modulated in a more detailed manner, which significantly improves the system's ability to reconstruct the target details.

[0022] (2) For small targets in large scenes, the present invention can complete high-quality imaging with only a lower resolution and a smaller sampling rate, which reduces the amount of data collected and the reconstruction time, thereby significantly improving the overall imaging speed of the system while ensuring reconstruction quality.

[0023] (3) By introducing a coefficient matrix, this invention supports one-time imaging of multi-region scenes, avoids the errors caused by multiple measurements and reconstruction of sub-image stitching, supports parallel compressed sensing sampling of multiple resolutions and multiple sizes, and can flexibly match the feature differences of different regions in the scene, which significantly improves the detection efficiency of complex scenes such as multiple targets and small targets under large-scale terrain. Attached image description:

[0024] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0025] Figure 1 With only two targets in a 256×256 pixel field of view, the modulation matrix is ​​set according to the size of the targets. The triangular targets are modulated at a resolution of 32×32 and a sampling rate of 0.2, while the circular targets are modulated at a resolution of 64×64 and a sampling rate of 0.05. The upper figure shows the target terrain, and the lower figure shows the modulation mask corresponding to the optimized modulation matrix under the scheme of this invention.

[0026] Figure 2 With only two targets in a 256×256 pixel field of view, the full-coverage target terrain is modulated using a Hadamard matrix with a resolution of 128×128 and a sampling rate of 0.05. The top image shows the target terrain, and the bottom image shows the modulation mask corresponding to the full-coverage Hadamard matrix modulation matrix. Detailed implementation method:

[0027] This invention provides a method for optimizing the modulation matrix of multi-region compressed sensing imaging, which introduces a coefficient matrix to optimize the modulation mask. The specific technical implementation steps are as follows:

[0028] by Figure 1 Taking a large field of view with small targets as an example, in a 256×256 pixel field of view, there are only two targets, and the number of targets k=2. The target terrain can be represented as:

[0029]

[0030] Where x1 has a size of 32×32, x2 has a size of 64×64, and the rest are all zero matrices (which may not be square matrices). The corresponding basis matrix P is represented as:

[0031]

[0032] The modulation matrix is ​​set according to the size of the target object. The triangular target object x1 is modulated with a resolution of 32×32 and a sampling rate of 0.2, and the size of P1 is 32×32. The circular target object x2 is modulated with a resolution of 64×64 and a sampling rate of 0.05, and the size of P2 is 64×64.

[0033] Represent the region of non-zero values ​​as a vector x = [x1 x2] T P = [P1 P2] T The corresponding baseline measurement value z is calculated as follows:

[0034] z h =Tr((P h ) T x h h = 1, 2 (9)

[0035] Since the target number k = 2, we define a set of linearly independent 2×2 coefficient matrices C as follows:

[0036]

[0037] Then the formula

[0038]

[0039] The target information x can be reconstructed using compressed sensing algorithms.

[0040] This modulation matrix optimization method is compared with... Figure 2 Traditional full-coverage modulation methods reduce the number of measurements and perform more refined encoding only on the target, which can effectively optimize the reconstruction of the target's two-dimensional shape details. Combined with various ranging techniques, it can ultimately achieve three-dimensional imaging.

Claims

1. A method for optimizing the modulation matrix of multi-region compressed sensing imaging, characterized in that, Includes the following steps: The measurement process of compressed sensing theory can be represented as y = Φx. After setting the observation matrix Φ, the original signal x can be reconstructed from the detected signal. The target terrain is represented as follows: x ij It is a sub-block, the size of which is determined by the size of the target, with the target location x. ij ≠0 (generally a square matrix), x is at the point where there is no target. ij =0 (can be a square matrix), assuming there are k targets in total, i.e., x ij The number of sub-blocks ≠ 0 is k, and the modulation matrix corresponding to the sub-block size is set as follows: P ij Let x be a basis matrix. ij If the sub-block has a target, then P ij ≠0 (generally a square matrix), if the corresponding x ij If the sub-block has no target, then P ij =0 (can be a non-square matrix), P ij Let k be the number of ≠0, and P be the number of ≠0. ij For x ij Modulation matrices of the same size, different P ij It can accommodate different resolutions, each sub-block x ij The corresponding baseline measurement value z ij The calculation is as follows: With ij =Tr((P ij ) T x ij ) ⑶ To reconstruct the target terrain x, the compressed sensing equation y = Φx needs to have a unique solution. Therefore, the coefficient matrix C of the basis matrix P is designed as follows: c ij For coefficients that are 1 or -1, write all non-zero valid sub-blocks in the above formula as vectors: x = [x1 … x k ] T P = [P1 … P k ] T The compressed sensing reconstruction formula can be expanded as follows: Guarantee coefficient matrix C h If the regions are linearly independent, then compressed sensing algorithms can be used to complete the reconstruction of multiple regions in one go.

2. The method for optimizing a modulation matrix in multi-region compressed sensing imaging according to claim 1, characterized in that, The division of the target terrain into regions, the size of the regions, and the resolution and sampling rate matched to the regions are arbitrary. They are generally designed according to the size and location of the target. The size and block division of the basis matrix P can correspond to the target terrain x.

3. The method for optimizing a modulation matrix in multi-region compressed sensing imaging according to claim 1, characterized in that, The observation matrix Φ obtained from the design can be precisely represented as: c in the coefficient matrix ij The coefficient matrix C is either "1" or "-1". h If the coefficients are linearly independent, the target terrain x can be reconstructed by solving the compressed sensing principle y = Φx, and the coefficient matrix C is not unique.

4. The method for optimizing the modulation matrix of multi-region compressed sensing imaging according to claim 1, characterized in that, A non-zero basis matrix P ij The modulation masks generated by matrices such as the Hadamard matrix that satisfy the finite isometry property (RIP property) and thus provide a solution to the compressed sensing equation are all applicable to this invention.

5. The method for optimizing a modulation matrix for multi-region compressed sensing imaging according to claim 1, characterized in that, The system uses an optimized modulation matrix mask loaded onto a DMD to modulate the echo. The modulated echo is received by a single-pixel detector, and the two-dimensional information of the target is reconstructed using a compressed sensing algorithm. It can be combined with arbitrary ranging technology to obtain the depth information of the target, and finally achieve three-dimensional imaging.

Images