A region-based block compressive sensing algorithm for plant hyperspectral images

In order to improve the reconstruction effect of plant hyperspectral images, a region-based block compressive sensing (RBCS) algorithm is proposed. Local means and local standard deviations (LMLSD) criterion is used to select the optimal band in the hyperspectral images. The k-means clustering algorithm is introduced to extract the tea regions from the optimal band. And spatial adaptive blocking strategy is involved to realize the optimized spatial blocking only for tea regions in the hyperspectral images. Then discrete cosine transform (DCT) sparse basis and random gaussian measurement matrix are combined to compress the data. Finally, stagewise orthogonal matching pursuit (StOMP) algorithm is used to reconstruct plant hyperspectral images. Peak signal to noise ratio (PSNR), spectrum curve and spectral angle mapper (SAM) and the error of spectral indices are used to evaluate the reconstructed performance in the spatial and spectral domains. Experimental results show that the reconstructed performance of RBCS is significantly better than that of single spectral compressive sensing (SSCS) and block compressive sensing (BCS) at different sampling ratios.