Imaging system and method based on compressed sensing block complementation

Abstract

The invention provides an imaging system and method based on compressed sensing block complementation, and belongs to the technical field of imaging based on compressed sensing block complementation. The technical problem to be solved is to provide an imaging system structure based on compressed sensing block complementation and improvement of an imaging method. According to the technical scheme, the device comprises a telescope used for receiving scattered signals, and the rear end of the telescope is sequentially provided with a first lens and a fast steering mirror; the system further comprises a computer and a microcontroller, the microcontroller controls the fast steering mirror to deflect to image scattering signals to a modulation area of the digital micromirror device by sending rotation signals to the fast steering mirror, and a second lens is further arranged between the fast steering mirror and the digital micromirror device. An array detector is also arranged above the digital micromirror device; the method is applied to image data processing.

Description

Technical Field

[0001] The present invention provides an imaging system and method based on compressed sensing block complementation, belonging to the technical field of imaging based on compressed sensing block complementation. Background Art

[0002] Research on weak target detection and recognition is crucial for early warning and accurate response to potential threats. The array detectors currently used are limited by their complexity and high cost when performing image recognition. Such detectors cannot quickly and accurately detect and identify targets when dealing with large-scale monitoring tasks. In addition, the single-pixel compressed sensing imaging technology used, combined with the computational imaging algorithm, can reconstruct large-pixel images with only a number of acquisitions far less than the number of images. Although the single-pixel compressed sensing imaging technology has the advantages of simple system structure, low development cost, high detection sensitivity and high resolution, it enhances the detection capability of small targets at a long distance. However, in the face of complex scenes, single-pixel compressed sensing imaging still requires a large number of samples. The increase in the number of samples and pixels causes a significant increase in the time required to collect coded information and reconstruct the entire image, which will result in the imaging speed not meeting the requirements in dynamic scenes, affecting the ability of real-time monitoring and rapid response. In practical applications, especially in scenarios where fast-moving objects need to be monitored and tracked, lengthy sampling and reconstruction times will become limiting factors.

[0003] In order to overcome the above technical defects, a scheme for image recognition based on block compressed sensing technology has been proposed. This technology combines the advantages of array imaging and single-pixel compressed sensing imaging technology and has broad application prospects. Compared with traditional single-pixel compressed sensing imaging, the block compressed sensing method greatly reduces the number of modulation masks and avoids processing high-dimensional data. This technology uses less low-dimensional data for image reconstruction and is more suitable for real-time, energy-limited imaging applications. At the same time, since each sub-block image is processed independently, the image reconstruction process can be accelerated. By using high-resolution digital micromirror devices to encode the scene in parallel blocks and using low-resolution detectors to sample compressed data, an efficient reconstruction algorithm can be used to reconstruct a high-resolution image. This framework based on block compressed sensing can be applied to image acquisition systems, which can reduce the memory and energy consumption burden of sensors and reconstruction algorithms, realize high-resolution imaging based on low-resolution detectors, and greatly reduce sampling time, equipment cost and amount of transmitted data. In the parallel single-pixel imaging process, each block area should be accurately received by each detection unit.

[0004] However, the above imaging method is difficult to achieve precise alignment between the block area and the detection unit in real application scenarios, which will lead to problems such as block artifacts and information loss in the reconstructed image, seriously affecting the imaging quality. Therefore, it is necessary to design another new imaging system and method to overcome the above defects. Summary of the invention

[0005] In order to overcome the deficiencies in the prior art, the present invention aims to solve the technical problem of providing an improved imaging system structure and imaging method based on compressed sensing block complementarity.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is: an imaging system based on compressed sensing block complementarity, comprising a telescope for receiving scattered signals, a first lens and a fast steering mirror are sequentially arranged at the rear end of the telescope; It also includes a computer and a microcontroller, wherein the microcontroller controls the fast steering mirror to deflect and image the scattered signal to the modulation area of ​​the digital micromirror device by sending a rotation signal to the fast steering mirror, and a second lens is also arranged between the fast steering mirror and the digital micromirror device; An array detector is also arranged above the digital micromirror device. The computer sends a modulation signal to the digital micromirror device, so that the digital micromirror device performs encoding modulation on the imaging target, and then emits the modulation signal to the outside, which is received by the array detector. A third lens is also arranged between the digital micromirror device and the array detector. The array detector transmits all response data to a computer via a wire; The microcontroller is connected to the signal output end of the digital micromirror device through a wire.

[0007] A method for imaging based on a compressed sensing block complementary imaging system includes the following imaging steps: Step 1: Collect imaging target signals: Step 1.1: The telescope collects the scattered signal of the imaging target resolution plate, and the collected scattered signal is converged at the signal receiving end of the fast steering mirror through the first lens; Step 1.2: The microcontroller outputs two voltage signals with voltage values ​​of 0V and 0V respectively, controls the deflection angle of the fast steering mirror to be at the initial position, and reflects the scattered signal and images it into the modulation area of ​​the digital micromirror device through the second lens; Step 1.3: The computer sends a preset modulation signal to the digital micromirror device, so that a set of large mask information consisting of multiple identical Hadamard masks is loaded on the digital micromirror device, and the imaging target is coded and modulated. Then the digital micromirror device emits the modulated signal outward, and the image is formed on the array detector through the third lens; Step 1.4: By adjusting the position of the array detector, the modulation signal of each block modulation area on the digital micromirror device is collected into a single block area corresponding to the array detector, and all response data of the array detector are transmitted to the computer; Step 1.5: When the microcontroller detects that the modulation mask playback count of the digital micromirror device has completed a set, the microcontroller outputs two voltage signals with voltage values ​​of V1 and V2 respectively, controls the fast steering mirror to change the reflection angle, and images the scattered signal to the digital micromirror device; Step 2: Reconstruct the two stitched images: Step 2.1: Preprocess the data acquired by the computer after the array detector responds. Segment the acquired data according to the block area of ​​the array detector to obtain the corresponding data in each block area. Arrange the data in each block into a column. Each row represents the response of a single pixel detector under a certain mask. After processing the i×i block area, obtain i 2 Column data, each column of data corresponds to a response in a block area, where i is a positive integer; Step 2.2: Use the block Hadamard mask matrix and TVAL3 algorithm in parallel to perform single pixel reconstruction on the processed column data in each block; Step 2.3: seamlessly stitch together each block area image after reconstruction according to the column data source to obtain a complete multi-pixel stitched image as the initial image; Step 2.4: Change the output voltage of the microcontroller to change the deflection angle of the fast steering mirror, image the scattered signal to the modulation area of ​​the digital micromirror device again, collect all the response data by the computer, pre-process the data obtained by the computer, and calculate the light intensity value corresponding to each mask in a single pixel in the array detector at different deflection angles of the fast steering mirror; The image reconstruction network and the light intensity value corresponding to a single mask in each pixel of the array detector are used to restore the sub-image of each block area under two deflection angles, and then each sub-image is spliced ​​into a complete image according to the block position; Step 3: Use the trained U-net neural network to fuse two stitched images of the same target with different fields of view to restore a high-quality, high-pixel image.

[0008] The specific method of controlling the rapid steering mirror to change the deflection angle in step 2.4 is: The microcontroller is used to output two voltage values ​​V1 and V2 to control the deflection of the fast steering mirror, so that the lateral and longitudinal deflection angles α1 and α2 respectively satisfy: ; When the digital micromirror device has finished playing a set of Hadamard masks, the output voltage value of the microcontroller is increased by θ , and control the lateral and longitudinal deflection angles α3 and α4 of the fast steering mirror to satisfy: .

[0009] The specific method of stitching each sub-image into a complete image in step 2.4 is: Set the number of mask blocks to i×i, and let each block contain a Hadamard mask. The Hadamard masks in different blocks of each mask are consistent, and obtain a group of block imaging masks. Adjust the number of masks by changing the compression rate, and let the single mask be 2 n ×2 n Pixels, number of masks is 16 m The complete mask is a set of compression ratio 2 4m-2n % of 2 n+2 ×2 n+2 Pixel image, where i, n, and m are all positive integers; The modulation area of ​​the digital micromirror device is evenly divided into i×i blocks without overlap, and each detection unit of the array detector collects the light intensity information of the corresponding block area; The array detector pixels are divided into i×i blocks, and the light intensity information collected in each block is processed using the TVAL3 reconstruction algorithm to obtain the reconstructed image of each area. 2 Single pixel reconstruction of 2 n *2 n Pixel image, and then complete the i×i image stitching according to the location of the block area.

[0010] The specific method of restoring a high-quality high-pixel image by fusing the stitched images in step 3 is: Two images of the same target before and after the field of view shift n+2 ×2 n+2 The stitched images of pixel size and i×i blocks are grouped together. The U-net neural network is used with the target image group as input and the standard image as output. The feature information is extracted from each group of images before and after the field of view shift, and the network model is trained. Then, the trained network model is used to reconstruct a group of stitched low-quality images into high-quality images, eliminating image block artifacts and supplementing missing information.

[0011] The beneficial effects of the present invention compared with the prior art are as follows: the imaging system based on compressed sensing block complementation provided by the present invention realizes image reconstruction of different positions of the imaging target by slightly transforming the imaging field of view, and uses the complementary information in the two directly reconstructed pictures to restore the high-quality large-pixel image, thereby providing more effective information for the recognition and display of the imaging target, and solving the technical problem that the block artifacts in the block compressed sensing imaging seriously affect the imaging quality; compared with the existing block compressed sensing imaging technology, the present invention uses a pre-trained fusion imaging network to synthesize two low-quality target images at different positions, thereby effectively reducing the random noise in the imaging process and the influence of abnormal noise points in the system on the imaging quality, improving the noise suppression capability of the imaging system, and being able to ensure the stability and reliability of the imaging system under the conditions of low signal-to-noise ratio and system abnormality. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The present invention will be further described below in conjunction with the accompanying drawings: Figure 1 is a schematic structural diagram of an imaging system of the present invention; Figure 2 A schematic diagram of a block-based compressed sensing network used in the present invention; Figure 3 A schematic diagram of a block complementary network used in the present invention; Figure 4 A comparison diagram of simulation effects of a target under different block conditions in an embodiment of the present invention; Figure 5 The effect diagram of the simulation imaging experiment for verifying the effectiveness of the embodiment of the present invention; Figure 6 The following is an imaging experiment effect diagram for verifying the effectiveness of the embodiment of the present invention; In the figure: 101 is a telescope, 102 is a first lens, 103 is a fast steering mirror, 104 is a second lens, 105 is a digital micromirror device, 106 is a third lens, 107 is an array detector, 108 is a computer, and 109 is a microcontroller. DETAILED DESCRIPTION

[0013] like Figures 1 to 3As shown, the present invention discloses an imaging system and method based on compressed sensing block complementation. The adopted imaging system includes a field of view offset module, a detection imaging module and a high-quality image synthesis module. In the field of view offset module, the optical lens converges the imaging target on the fast steering mirror, and the optical signal is reflected from different angles by flipping the fast steering mirror to change the overall detection field of view; in the detection imaging module, the optical signal after the field of view offset module is focused by the lens to the digital micromirror device, and the digital micromirror device is divided into regions and loaded with multiple Hadamard masks to encode and modulate the optical signal, and the modulated optical signal is focused again on the array detector through the lens; the block reconstruction image of the target is obtained by parallel calculation using an image reconstruction network, and the block reconstruction image is spliced ​​into the overall target image according to the block position; in the high-quality image synthesis module, the initial image and the offset image are synthesized by a fusion imaging network, and finally a high-quality target image is restored; the present invention uses a compressed sensing block complementation method, and through the complementary characteristics of different field of view imaging target information, the information gap between the detection units of the array detector is filled, the block artifacts are effectively solved, the random noise and the inherent noise of the system can be effectively suppressed, and the large-pixel high-quality image reconstruction is realized.

[0014] The present invention proposes an imaging system based on compressed sensing block complementation, which mainly uses a fast steering mirror to adjust the imaging field of view to obtain two large-pixel stitched images of the same target at different positions, and restores the two low-quality images to a high-quality target image through a fusion imaging network, effectively solving the problems of image block artifacts and information loss caused by pixel mismatch and inaccurate image stitching, and improving the image reconstruction quality. In order to solve the above technical problems, the technical solution adopted by the present invention is: The present invention collects the scattered signals of the imaging target through a telescope 101 and converges them to a fast steering mirror 103 through a first lens 102. The microcontroller 109 outputs a voltage to control the deflection angle of the fast steering mirror 103 to image the scattered signals to a modulation area of ​​a digital micromirror device 105 through a second lens 104. The computer 108 sends a modulation signal to the digital micromirror device 105, and a group of block Hadamard masks are loaded in the digital micromirror device 105 to encode and modulate the target image; the modulation signal is focused to an array detector 107 through a third lens 106, and the data of the array detector 107 after being triggered by the digital micromirror device is transmitted to the computer 108 and preprocessed, and the acquired data is converted into a digital micromirror device 105. The data is segmented according to the detector block area, and the corresponding data in each block area is obtained and arranged in a column; the block Hadamard mask matrix and TVAL3 algorithm are used to reconstruct the data in each block by single pixel; the reconstructed block images are spliced ​​according to the location of the block data source to obtain a complete spliced ​​image, which is called the initial image; the output voltage of the microcontroller 109 is changed to change the deflection angle of the fast steering mirror 103, and the above steps are repeated to obtain a spliced ​​image of the same target after the field of view is shifted, which is called the shifted image; finally, a well-trained U-net neural network is used to fuse two low-quality spliced ​​images of the same target in different fields of view to restore a high-quality image.

[0015] The embodiment of the present invention provides an imaging method based on compressed sensing block complementation, comprising the following control steps: Step 1: Collect imaging target signals: Step 1.1: The scattered signals of the imaging target resolution plate are collected by the telescope 101 and converged on the fast steering mirror 103 through the first lens 102; Step 1.2: The microcontroller 109 outputs two voltage values ​​of 0V and 0V respectively, controls the deflection angle of the fast steering mirror 103 to be at the initial position, and images the scattered signal to the modulation area of ​​the digital micromirror device 105; Step 1.3: The computer 108 sends a preset modulation signal to the digital micromirror device 105. A set of (512) large masks consisting of multiple identical Hadamard masks are loaded on the digital micromirror device 105 to perform coding modulation on the imaging target. The modulation signal is vertically emitted from the digital micromirror device 105 and then imaged on the array detector 107 through the third lens 106. Step 1.4: By adjusting the position of the array detector, the modulation signal of each block modulation area of ​​the digital micromirror device 105 is collected into a single block area corresponding to the array detector 107, and all response data of the array detector 107 are transmitted to the computer 108; Step 1.5: After the microcontroller 109 detects that the modulation mask playback count of the digital micromirror device 105 has completed a set, it adjusts the output voltage value to V1, V2, controls the change of the reflection angle of the fast steering mirror 103, and images the scattered signal to the digital micromirror device 105. The position of the resolution board scattered signal in the modulation area of ​​the digital micromirror device 105 is located in the upper left corner of the initial position.

[0016] Step 2: Reconstruct two stitched images: Step 2.1: Preprocess the data acquired by the computer 108 after the array detector 107 responds. The acquired data is divided according to the block area of ​​the detector to obtain the corresponding data in each block area. The data in each block is arranged into a column. Each row represents the response of a single pixel detector under a certain mask condition. After processing the i×i block area, i 2 Column data, each column of data corresponds to a response in a block area, where i is a positive integer; Step 2.2: Use the block Hadamard mask matrix and TVAL3 algorithm in parallel to perform single pixel reconstruction on the processed column data in each block; Step 2.3: seamlessly stitch together each block area image after reconstruction according to the column data source to obtain a complete multi-pixel stitching image; Step 2.4: After changing the output voltage of the microcontroller 109 to change the deflection angle of the fast steering mirror 103, repeat steps 2.1-2.3 to collect a second set of data to obtain a reconstructed image of the same target after the field of view is shifted.

[0017] The specific method of controlling the fast steering mirror 103 to change the deflection angle is: The microcontroller 109 is used to output two voltage values ​​V1 and V2 to control the deflection of the fast steering mirror 103, so that the lateral and longitudinal deflection angles α1 and α2 respectively satisfy: ; When the digital micromirror device 105 has finished playing a set of Hadamard masks, the output voltage value of the microcontroller 109 is increased by θ , and control the lateral and longitudinal deflection angles α3 and α4 of the fast steering mirror 103 to satisfy: .

[0018] Finally, each sub-image is stitched into a complete image as follows: Set the number of mask blocks to i×i, and let each block contain a Hadamard mask. The Hadamard masks in different blocks of each mask are consistent, and obtain a group of block imaging masks. Adjust the number of masks by changing the compression rate, and let the single mask be 2 n ×2 nPixels, number of masks is 16 m The complete mask is a set of compression ratio 2 4m-2n % of 2 n+2 ×2 n+2 Pixel image, where i, n, and m are all positive integers; The modulation area of ​​the digital micromirror device 105 is evenly divided into i×i blocks without overlapping, and each detection unit of the array detector 107 collects the light intensity information of the corresponding block area; The array detector 107 pixels are divided into i×i blocks, and the light intensity information collected in each block is processed using the TVAL3 reconstruction algorithm to obtain the reconstructed image of each area. 2 Single pixel reconstruction of 2 n *2 n Pixel image, and then complete the i×i image stitching according to the location of the block area.

[0019] Step 3: Finally, a well-trained U-net neural network is used to fuse two stitched images of the same target with different fields of view to restore a high-quality, high-pixel image. The specific method is as follows: Two images of the same target before and after the field of view shift n+2 ×2 n+2 The stitched images of pixel size and i×i blocks are grouped together. The U-net neural network is used with the target image group as input and the standard image as output. The feature information is extracted from each group of images before and after the field of view shift, and the network model is trained. Then, the trained network model is used to reconstruct a group of stitched low-quality images into high-quality images, eliminating image block artifacts and supplementing missing information.

[0020] The effectiveness of the technical solution of the present invention is verified by experiments below: like Figure 4 The figure shows the changes in the imaging quality of the reconstructed image under different block conditions when the compression rate is 12.5%. Using a random sparse matrix, the left figure shows the imaging target, and the right figure shows the image reconstruction quality under the corresponding block conditions. The results show that as the number of blocks increases, the signal-to-noise ratio in a single block area of ​​the imaging target is higher, and the overall reconstruction imaging effect is improved. Figure 4 Taking the seventh reconstruction target in the figure as an example, the quality of the reconstructed image with a block number of 10×10 is better than that with a block number of 1×1, and the PSNR (solid line) and SSIM (dashed line) are increased by 7.27 dB and 0.168 respectively. In addition, the increase in the number of blocks is accompanied by a decrease in imaging detection time and reconstruction time. In different application scenarios, the appropriate number of blocks can be selected according to the imaging quality and speed requirements to achieve optimized processing of the imaging target.

[0021] In order to verify the feasibility of block-based complementary imaging based on compressed sensing in the presence of noise, the following takes 4×4 blocks as an example. An imaging target is set in the experimental environment, noise is added to the digital micromirror device, and the imaging process under different noise interference is simulated.

[0022] like Figure 5 The figure shows the effects of direct block reconstruction and complementary fusion restoration under different noise conditions. In the experiment, the Hadamard mask compression rate was set to 12.5%, and the reconstructed image was 256×256 pixels. The left picture shows the imaging target, and the right picture corresponds to three cases: no noise, adding random noise, and adding abnormal noise points. The method of adding abnormal noise points is to add 10-20 bright spots with a diameter of 5 pixels at any position on the basis of random noise. With the addition of noise, the quality of the reconstructed image decreases. The PSNR and SSIM of the image reconstructed under the condition of adding abnormal noise points decrease by 8.57 dB and 0.523 respectively. The image quality restored by block complementary fusion shows a significant improvement in quality compared with the directly reconstructed image under the three noise conditions. Even when adding abnormal noise points, the PSNR is still improved by 11.64 dB and the SSIM is improved by 0.529. The results show that the proposed method can effectively fill the image information loss and artifacts caused by the detector unit gap, detector anomalies and noise introduced in the single-pixel reconstruction process, and can reconstruct high-quality images with PSNR≥31.5 dB and SSIM≥0.85 under several different noise conditions.

[0023] like Figure 6 The results of high-quality image restoration of the imaging target image under different noise conditions are shown. The results show that the PSNR of the image restored based on compressed sensing block complementarity is ≥32 dB and SSIM ≥0.96 in the case of random noise and abnormal point noise. In the case of random noise, the PSNR and SSIM of the image restored by dynamic complementarity are improved by 25.8 dB and 0.939 respectively compared with the direct block reconstruction image. When abnormal noise points are added, the PSNR and SSIM increase by 23.74 dB and 0.928 respectively. The results show that when facing the abnormal noise points of the detector, the dynamic complementary imaging method based on block compressed sensing can eliminate noise, fill in missing information, and greatly improve image quality.

[0024] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An imaging system based on compressed sensing block complementation, characterized in that: It comprises a telescope (101) for receiving scattered signals, wherein a first lens (102) and a fast steering mirror (103) are sequentially arranged at the rear end of the telescope (101); It also includes a computer (108) and a microcontroller (109), wherein the microcontroller (109) controls the fast steering mirror (103) to deflect and image the scattered signal onto a modulation area of ​​a digital micromirror device (105) by sending a rotation signal to the fast steering mirror (103), and a second lens (104) is also provided between the fast steering mirror (103) and the digital micromirror device (105); An array detector (107) is also provided above the digital micromirror device (105); the computer (108) sends a modulation signal to the digital micromirror device (105), causing the digital micromirror device (105) to perform coding modulation on an imaging target, and then transmit the modulation signal to the outside, which is received by the array detector (107); and a third lens (106) is also provided between the digital micromirror device (105) and the array detector (107); The array detector (107) transmits all response data to a computer (108) via wires; The microcontroller (109) is connected to the signal output end of the digital micromirror device (105) via a wire.

2. The imaging method according to claim 1, characterized in that: The imaging steps include the following: Step 1: Collect imaging target signals: Step 1.1: The telescope (101) collects scattered signals from the imaging target resolution plate, and the collected scattered signals are converged at the signal receiving end of the fast steering mirror (103) through the first lens (102); Step 1.2: The microcontroller (109) outputs two voltage signals with voltage values ​​of 0V and 0V respectively, controls the deflection angle of the fast steering mirror (103) to be at an initial position, reflects the scattered signal, and images it onto the modulation area of ​​the digital micromirror device (105) through the second lens (104); Step 1.3: The computer (108) sends a preset modulation signal to the digital micromirror device (105), so that a set of large mask information composed of multiple identical Hadamard masks is loaded on the digital micromirror device (105), and the imaging target is coded and modulated. Then, the digital micromirror device (105) emits the modulated signal outward, and the signal is imaged on the array detector (107) through the third lens (106); Step 1.4: by adjusting the position of the array detector (107), the modulation signal of each block modulation area on the digital micromirror device (105) is collected into a single block area corresponding to the array detector (107), and all response data of the array detector (107) are transmitted to the computer (108); Step 1.5: When the microcontroller (109) detects that the modulation mask playback count of the digital micromirror device (105) has completed a set, the microcontroller (109) outputs two voltage signals with voltage values ​​of V1 and V2 respectively, to control the fast steering mirror (103) to change the reflection angle, so as to image the scattered signal onto the digital micromirror device (105); Step 2: Reconstruct the two stitched images: Step 2.1: Preprocess the data acquired by the computer (108) after the array detector (107) responds, divide the acquired data according to the block area of ​​the array detector (107), obtain the corresponding data in each block area, arrange the data in each block into a column, and each row represents the response of a single pixel detector under a certain mask condition. After processing the i×i block area, obtain i 2 Column data, each column of data corresponds to a response in a block area, where i is a positive integer; Step 2.2: Use the block Hadamard mask matrix and TVAL3 algorithm in parallel to perform single pixel reconstruction on the processed column data in each block; Step 2.3: seamlessly stitch together each block area image after reconstruction according to the column data source to obtain a complete multi-pixel stitched image as the initial image; Step 2.4: changing the output voltage of the microcontroller (109) so that the fast steering mirror (103) changes its deflection angle, imaging the scattered signal again onto the modulation area of ​​the digital micromirror device (105), collecting all response data by the computer (108), pre-processing the data acquired by the computer (108), and calculating the light intensity value corresponding to each mask in a single pixel of the array detector (107) at different deflection angles of the fast steering mirror (103); Using the light intensity value corresponding to a single mask in each pixel of the image reconstruction network and the array detector (107), the sub-image of each block area at two deflection angles is restored, and each sub-image is then spliced ​​into a complete image according to the block position; Step 3: Use the trained U-net neural network to fuse two stitched images of the same target with different fields of view to restore a high-quality, high-pixel image.

3. The imaging method according to claim 2, characterized in that: The specific method of controlling the rapid steering mirror (103) to change the deflection angle in step 2.4 is: The microcontroller (109) is used to output two voltage values ​​V1 and V2 to control the deflection of the fast steering mirror (103), so that the lateral and longitudinal deflection angles α1 and α2 respectively satisfy: ； When the digital micromirror device (105) has finished playing a set of Hadamard masks, the output voltage value at both ends of the microcontroller (109) is increased by θ , and controlling the lateral and longitudinal deflection angles α3 and α4 of the rapid steering mirror (103) to respectively satisfy: 。 4. The imaging method according to claim 2, characterized in that: The specific method of stitching each sub-image into a complete image in step 2.4 is: Set the number of mask blocks to i×i, and let each block contain a Hadamard mask. The Hadamard masks in different blocks of each mask are consistent, and obtain a group of block imaging masks. Adjust the number of masks by changing the compression rate, and let the single mask be 2 n ×2 n Pixels, number of masks is 16 m The complete mask is a set of compression ratio 2 4m-2n % of 2 n+2 ×2 n+2 Pixel image, where i, n, and m are all positive integers; The modulation area of ​​the digital micromirror device (105) is evenly divided into i×i blocks without overlap, and each detection unit of the array detector (107) collects light intensity information of the corresponding block area; The array detector (107) pixels are divided into i×i blocks, and the light intensity information collected in each block is processed using the TVAL3 reconstruction algorithm to obtain the reconstructed image of each area. 2 Single pixel reconstruction of 2 n *2 n Pixel image, and then complete the i×i image stitching according to the location of the block area.

5. The imaging method based on compressed sensing block complementation imaging system according to claim 2, characterized in that: The specific method of restoring a high-quality high-pixel image by fusing the stitched images in step 3 is: Two images of the same target before and after the field of view shift n+2 ×2 n+2 The stitched images of pixel size and i×i blocks are grouped together. The U-net neural network is used with the target image group as input and the standard image as output. The feature information is extracted from each group of images before and after the field of view shift, and the network model is trained. Then, the trained network model is used to reconstruct a group of stitched low-quality images into high-quality images, eliminating image block artifacts and supplementing missing information.

Images