混沌离散粒子群优化的高光谱影像端元提取算法

Endmember Extraction Algorithm on Hyperspectral Remote Sensing Images Based on Chaotic Discrete Particle Swarm Optimization

针对离散粒子群优化算法在进行端元搜索时易陷入局部最优值等缺陷,提出了混沌机制扰动下的离散粒子群优化的端元提取算法(CDPSO-EE)。研究了混沌理论特有的随机性、遍历性以及对初始值敏感性等特点,将混沌理论引入到DPSO端元提取算法的初始化阶段,优化了初始种群质量;将混沌变量附加到粒子自身历史最优位置获得扰动新位置,对比混沌扰动前后粒子所在位置的适应度函数值,选择最优位置为粒子新的位置,让粒子有能力跳出局部最优值。结果表明CDPSO-EE在高光谱影像端元提取方面具有更好的端元提取质量。

For the discrete particle swarm optimization easily to lost in local optimum during endmember search, the endmember extraction based on chaotic discrete particle swarm optimization（ CDPSO-EE） was proposed. The properties of randomness, ergodicity and sensitivity to initial value of chaos theory was researched. The chaos theory was introduced into initialization phase of the DPSO endmember extraction algorithm to optimize the quality of the initial population, then the chaotic variables was attached to the personal best positions of particles, then obtained the new disturbed position. Comparison the fitness function value of particles position before and after chaotic disturbance, the optimal position was chosen for new particles, so that the particles had the ability to escape from local optimum. Finally, the result showed that CDPSO-EE has a better quality in hyperspectral image endmember extraction.