FPGA和多DSP系统的并行RX探测算法  

Parallel RX algorithm implementation based on the FPGA and multi-DSP system

作　　者：赵宝玮[1] 相里斌[1,2] 吕群波[2] 张桂峰[2] 曾晓茹[2] 郭文记[2] 

机构地区：[1]西安电子科技大学物理与光电工程学院,陕西西安710071 [2]中国科学院光电研究院,北京100094

出　　处：《西安电子科技大学学报》2014年第3期152-156,共5页Journal of Xidian University

基　　金：国家杰出青年科学基金资助项目(61225024)

摘　　要：针对高光谱图像异常目标探测过程运算量大、结果不能实时应用的问题,在FPGA和多数字信号处理的硬件平台上实现了RX并行处理算法.结合RX算法的原理与特征,研究了该算法的并行特性;通过计算机仿真验证了RX算法并行化的可行性及特点;该平台通过FPGA完成高光谱图像数据立方体的奇异值分解降维,降低了数据处理量和传输量;通过多数字信号处理完成RX算法的并行化,实现了高光谱图像异常目标探测的快速处理.用该系统处理64波段280×800大小的高光谱图像数据,得到探测结果仅需4.86s,能够满足高光谱遥感应用中异常目标探测的载荷平台在线处理和探测结果的快速获取及应用的需求.Hyperspectral imagery is hard to apply for real-time detection of abnormal targets due to its huge amount of computation . This paper proposes a fast processing method by parallelizing the RX detection algorithm on a hardware platform based on FPGA and multi-DSP . According to the principle and characteristics of the RX algorithm , computer simulation is completed to confirm the feasibility and characteristics of parallelization . On this basis , the parallel processing RX algorithm is implemented on this platform , where FPGA implements the SVD for dimensionality reduction in hyperspectral imagery to reduce the amount of data processing and transmission , and multi-DSP speeds up the process of abnormal target detection by parallelizing the RX algorithm . For hyperspectral data including 64 bands of images at size 280 ＆#215; 800 , the proposed system only took 4.86 s to obtain the detection result . The demonstrated ability can satisfy the requirement for online processing and fast result acquisition in applications of hyperspectral remote sensing .

关 键 词：并行处理 高光谱图像 目标探测 RX算法 多数字信号处理 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]