模拟退火机制下优化离散粒子群的端元提取算法

Endmember Extraction Algorithm on Hyperspectral Imagery by Simulated Annealing Discrete Particle Swarm Optimization

针对离散粒子群优化(Discrete Particle Swarm Optimization,DPSO)端元提取算法初始种群质量差、收敛性能低且易于陷入局部最优,本文将模拟退火算法引入到DPSO的不同阶段,模拟退火算法能以一定的概率接受和舍弃新状态,使种群内粒子渐趋有序、达到平衡,收敛到全局最优,有效避免了搜索陷入局部最优。因此,该算法不仅保持了DPSO的全局组合优化特点,克服了初始种群质量差、易陷入局部最优等缺点,而且还提高了收敛速度和端元提取精度。

In view of the existing endmember extraction research defects that the Discrete Particle Swarm Optimization （DPSO） algo- rithm has poor quality of initial particle swarm, low convergence performance and easily falls into the local optimum. So simulated an- nealing algorithm is introduced into the different stages of DPSO. Using the Simulated Annealing algorithm can accept and give up a new state at a certain probability so that the particles can be gradually ordered to achieve balance and ultimately converged to global optimum, thus it can avoid search falling into the local optimum. Therefore, the new algorithm can keep the DPSO characteristics of global combinatorial optimization, also overcomes the disadvantages of poor quality of initial particle swarm and easily failing into the local optimum, as well as improves the convergence speed and the precision of endmember extraction.