基于FPGA的高光谱图像奇异值分解降维技术

为了解决高光谱图像维数高、数据量巨大、实时处理技术实现难的问题,提出了高光谱图像实时处理降维技术.采用奇异值分解(SVD)算法对高光谱图像进行降维,又在可编程门阵列(FPGA)芯片中针对这一算法划为自相关模块、特征求解模块、特征提取模块和降维实现模块4个模块进行编程实现、仿真和验证.仿真结果表明,高光谱图像降维后数据量为降维前的1/3,而降维后的分类像素点误差为0.2109%,证明了奇异值分解算法进行高光谱图像降维算法的有效性.

FPGA Implement of SVD for Dimensionality Reduction in Hyperspectral Images

To solve hyperspectral image's problems of the high dimensionality, the huge amount of data, and the real-time solution and so on, a real-time hyperspectral dimensionality reduction method is brought forward. Based on singular value decomposition (SVD) method, hyperspectral dimensionality is reduction, and finish the design of the chip system with top-down method. The chip system is divided into autocorrelation module, SVD module, feature extraction module and dimensionality reduction module. It completes the design, simulation and verification of these modules. The results indicate that the hyperspectral image reduced to 1/3, classification error is only 0.2109 percent after the dimensionality reduction. All of this show, the SVD method for hyperspectral dimensionality reduction is effective.