To improve the reconstruction quality of hyperspectral compression sensing, a hyperspectral reconstruction algorithm based on dual-band prediction is proposed. K-means clustering algorithm is introduced to adaptively complete the band grouping, and the dual reference bands in the group are determined. Dual band prediction model is employed to obtain the initial prediction image. Based on the predicted image, the modified reconstruction and weighted fusion method are utilized to achieve highprecision image reconstruction. The results show that at the same sampling rate, the reconstructed image of this method has better Peak Signal to Noise Ratio(PSNR) and Structural SIMilarity(SSIM) compared with that of the existing reconstruction method.