粒子群优化的高光谱图像快速稀疏分解算法

Fast Sparse Decomposition Algorithm for Hyperspectral Images Based on Particle Swarm Optimization

获取高光谱图像的稀疏表示,能够降低原图像信号的数据量,便于对其进行后续处理。正交匹配追踪算法的计算复杂度高、不能满足实时处理要求,借鉴粒子群算法和Hermitian求逆引理的思想,提出了针对高光谱图像的粒子群快速稀疏分解算法。该算法依靠粒子群的局部寻优能力和Hermitian递归求逆对正交匹配追踪算法的匹配过程和残差更新过程进行改进,提高稀疏分解过程的执行效率。实验结果表明,与正交匹配追踪算法相比,在获得相同的重构图像峰值信噪比条件下,所提算法能将计算效率提高约25倍。

Obtaning the sparse representation of hyperspectral image signals can reduce the amount of original data and facilitate the subsequent processing. Since the computational complexity of traditional orthogonal matching pursuit algorithm is too high to be applied to real-time processing,in this paper,a fast sparse decomposition algorithm for hyperspectral images based on the idea of particle swarm optimization and Hermitian inversion lemma is proposed. The proposed algorithm relies on the local search ability of particle swarm optimization and the Hermitian recursive inversion to optimize the matching process and residual updating process in orthogonal matching pursuit algorithm,which leads to improve the computational efficiency. Experimental results show that compared with orthogonal matching pursuit algorithm,the proposed algorithm can improve the computational efficiency by about 25 times under the same reconstructed peak signal-to-noise ratio of hyperspectral images.