Study On Adaptive Sampling Technology For Single Photon Compressive Imaging

The single photon compressive imaging,which combines the single pixel imaging with the photon counting technology,can realize imaging under extremely weak light,and has wide application in biomedical imaging,astronomical detection,multispectral imaging and other fields.Because the sampling time of the single pixel imaging is limited by the measurements and the modulation frequency of spatial light modulator,the single pixel imaging often reduces the measurements and the sampling time by compressive sensing sampling.However,the reconstruction algorithm of compressive sensing usually involves iterative optimization,has high computational complexity,and the time spent increases exponentially with the image size.Under this background,the adaptive sampling techniques of the single photon compressive imaging are studied in the spatial domain and the transform domain respectively.The main research contents and results are as follows:(1)We study the adaptive sampling method using the spatial domain information of the image,and proposes an adaptive compressive sampling method based on the region of interest.The main method is to locate the region of interest through visual saliency.Then,non-uniform resolution masks are generated based on these regions to image the scene.The experimental results show that the adaptive compressive sampling based on the region of interest can obtain better imaging quality in the region of interest under the same measurement times.(2)Using the sparsity of the image in the transform domain,the image can be compressed and sampled directly in the transform domain,and then the image can be reconstructed by inverse transform.In order to reduce the storage of observation matrix and the computational complexity of compressed sensing,a block adaptive compressed sampling technique based on wavelet domain is proposed.By comparing the reconstruction time and image quality of different algorithms through experiments,it is verified that the scheme can effectively reduce the image reconstruction time and the storage of observation matrix,which is beneficial to relieve the hardware pressure.(3)A fast-adaptive target tracking method based on cross correlation is proposed for the single photon compressive imaging system.According to the method,a corresponding mask is adaptively loaded on the DMD,and a point detector is used to directly measure the spatial correlation between the target and the scene so as to track the target.The experimental results show that the method can track the target quickly and effectively,and only a few measurements are needed to update the target location information.