Research On Key Technology Of Single Pixel Imaging

Different from traditional array-based CCD or CMOS cameras,single-pixel imaging system uses only one pixel/detector to acquire scene image information.The special imaging mode has many advantages.For example,array-based infrared,terahertz,and single-photon cameras that manufacture high spatial resolution are expensive,while using a single pixel for imaging can greatly reduce the manufacturing cost of detectors.In addition,the single-pixel camera only uses the overall intensity signal of the scene to obtain the characteristics of the target image,making it widely used in cutting-edge imaging fields such as imaging through scattering media and non-line-of-sight imaging.Based on these excellent characteristics,single-pixel imaging technology has important application prospects in defense,military,remote sensing,biomedical and other fields.Although the advantages of single-pixel imaging technology compared to traditional cameras are significant,its drawbacks are also evident.Since only a single pixel is used for imaging,capturing a high-resolution image using a single-pixel camera requires multiple samplings,which makes imaging slow.Reducing the number of samples to improve imaging efficiency can also lead to a decrease in imaging quality.The problem of the trade-off between imaging quality and efficiency of single-pixel cameras has become a major constraint in their practical applications.This paper starts with the challenge that single-pixel imaging technology is difficult to balance in terms of imaging quality and efficiency and improves the imaging performance by improving the sampling method,image reconstruction algorithm,and coding mode,which are the three core elements of a single-pixel camera.The main innovative work is as follows:s:1.A heuristic high-low frequency hybrid sampling method and a fast high-quality image reconstruction algorithm are proposed.This paper addresses the problems of insufficient high-frequency sampling in traditional fixed-path sampling methods,spectral truncation,and high-frequency sampling that is too randomly dispersed in variabledensity random sampling strategies under low sampling rate conditions and proposes a new heuristic high-low frequency hybrid sampling method.The algorithm first samples the low-frequency spectrum of the image using the fixed-path sampling method and then samples the high-frequency spectrum using the important spectral distribution in the low-frequency spectrum,effectively solving the problems of insufficient high-frequency sampling and spectral truncation caused by traditional fixed-path sampling strategies and avoiding the problem of high-frequency spectral sampling being too dispersed in variable-density random sampling strategies.Compared with existing sampling methods,the proposed method obtains clearer and more detailed image details.Secondly,a new fast high-quality image reconstruction algorithm is proposed to solve the problem that the image reconstruction quality and reconstruction speed of existing image reconstruction methods are difficult to achieve simultaneously.By transforming the reconstruction model into the image gradient space and deriving a new closed-form solution model based on the alternating direction optimization architecture,an image reconstruction algorithm that balances quality and speed is obtained.Experimental results show that the proposed method guarantees imaging quality while compressing image reconstruction time to 20 milliseconds,greatly improving the practicality of the algorithm.2.A Fourier single-pixel video imaging method that combines spatial-temporal variable-density random sampling strategy with continuous frame image spatial sparsity and temporal local low-rank constraint is proposed.Traditional single-pixel reconstruction methods based on total variation regularization will cause ”staircase effects”in the image at low sampling rates,resulting in the loss of some image details.To address this problem,this paper introduces a second-order Hessian norm-based constraint to replace total variation constraints,effectively suppressing the ”staircase effects” and improving image details.In addition,to address the problem of image blurring caused by existing Fourier single-pixel video imaging methods failing to fully utilize the temporal redundancy information in the image,a new imaging algorithm that combines spatialtemporal variable-density random sampling methods with temporal local low-rank priors is proposed.By extending the traditional variable-density random sampling method to the spatial-temporal domain to fully sample the redundant information between continuous frame images and combining image temporal local low-rank priors to recover the image from sampling information,high-quality Fourier single-pixel video imaging is achieved.Finally,a corresponding closed-form solution algorithm is derived based on the above model to efficiently solve the imaging model.Simulation and optical experiments show that the proposed method improves imaging quality by 4-7 d B and achieves fast image reconstruction compared to existing methods.3.A sampling method based on S-basis pattern coding and S-transform spectrum distribution of images is proposed.To address the problem of insufficient information expression of Fourier basis patterns and low imaging quality under low sampling rate conditions,S-basis patterns are introduced to improve existing single-pixel imaging methods:first,a fast generation method for S-basis patterns is derived from the two-dimensional discrete S-transform,and the modulation strategy of S-basis patterns for scene information is improved based on the 2-simplex method,effectively reducing the number of samples.Secondly,to address the problem that the image S-transform spectrum is too dispersed and traditional fixed-path sampling methods cannot take advantage of the superiority of S-basis patterns in information expression,a new frequency spectrum sampling method suitable for S-basis patterns is proposed.By dividing the image S-transform spectrum into orders and sampling from low to high frequencies in each partition,the proposed method achieves a more uniform distribution of sample points in the S-transform domain and obtains higher-quality images with fewer samples.Overall,this article proposes a series of innovative solutions starting from the three core elements of single pixel imaging technology,including encoding,sampling,and reconstruction algorithms,which greatly improve the imaging performance of single-pixel imaging technology and provide important technical references for single-pixel imaging.