空间相关性约束联合子空间追踪的高光谱图像稀疏解混

通过深入分析高光谱图像空间相邻数据之间的空间相关性,提出一种利用空间相关性进行约束的联合子空间追踪解混(Spatial correlation constrained simultaneous subspace pursuit,SCCSSP)方法。该方法首先基于分块思想将高光谱图像进行分块处理,然后在图像块的端元提取步骤中,结合空间相关性特征对端元的提取进行约束,从而确保当前端元支撑集相对于高光谱图像残差是最优的。在丰度估计中将图像块的端元集合合并作为整幅图像的端元支撑集,通过求解非负性约束的最小二乘法获得丰度重建图像。模拟图像数据实验结果表明,本文方法在同等条件下能够获得更高的信号重构误差,且解混运算时间低于凸优化算法。在实际图像数据实验中,本文方法丰度图像稀疏度最低,取得了仅次于SUnSAL-TV算法的图像重建误差,其所得到的丰度重建图像也取得了更好的视觉效果。实验结果验证了本文方法具有更高的解混精度。

Spatial Correlation Constrained Simultaneous Subspace Pursuit for Sparse Unmixing of Hyperspectral Imagery

Based on the analyses of the interpixel correlation in the hyperspectral imagery, a spatial correlation constrained simultaneous subspace pursuit (SCCSSP) method is proposed. The method uses a block-processing strategy to divide the whole hyperspectral imagery into several blocks. In each block, the spatial correlation information is added to improve the accuracy of endmember selection, and ensures that the estimated endmember set is optimal to the current hyperspectral image residuals. The endmembers picked in each block is associated as the endmember sets of the whole hyperspectral imagery. Finally, the abundances are estimated by the nonnegative least squares method with the obtained end member sets. The results of simulated images experiment show that the proposed method can obtain higher signal reconstruction error under the same condition, and the time of unmixing operation is lower than the convex optimization algorithms. In the real images experiment, this method has the lowest sparsity of the abundance images, and is second only to the SUnSAL-TV algorithm in image reconstruction error. In addition, the reconstructed images obtained by this method obtain better visual effects. To sum up, experimental results on both simulated images and real images indicate that the hyperspectral unmixing accuracy of the SCCSSP algorithm is higher than that of the traditional methods.