动态测量的高光谱图像压缩感知

Based on the strong spectral correlation between adjacent bands of hyperspectral images, we proposes a compressed sensing algorithm that uses edge information to design dynamic measurement rate to improve the reconstruction effect of compressive sensing in hyperspectral images. First, each image block is sampled at a fixed measurement rate by a random projection block-compressive sensing method, a single-band image is reconstructed as a priori information of other bands, and an image edge region is extracted therefrom; then, the measurement values are adaptively assigned according to the richness of the edge information of each image block. With a certain total number of measurements, different number of measurements is assigned to different image blocks. Finally, the remaining wave bands are collected and reconstructed with the assigned measurements. The simulation results show that under the same total number of measurements, the hyperspectral image quality(Peak Signal to Noise Ratio(PSNR)) reconstructed by the dynamic measurement algorithm proposed in this paper is 1-4 dB higher than the traditional fixed-measurement compressive sensing strategy. Moreover, the reconstruction time is also reduced, and the image quality at the detail is further enhanced based on the successful reconstruction of the hyperspectral images.