基于粒子群优化算法的波段选择方法

Band Selection Method Based on Particle Swarm Optimization

针对现有的高光谱遥感波段选择方法应用于目标检测时效果不高的问题,提出了一种基于粒子群算法的波段选择(Band Selection Based on Particle Swarm Optimization,BS-PSO)新方法。首先,将能够有效衡量目标检测效果的ROC(Receiver Operating Characteristic)曲线下面积(Area Under the Curve,AUC)作为波段选择准则,并将之构造为适应度函数。接下来,利用群智能优化算法粒子群算法对波段选择搜索过程进行优化,最终得到拥有较高检测效果的波段组合。实验结果表明:以曲线下面积作为衡量波段的指标,能够有效地选出具有较高目标检测性能的波段组合,与其它几种典型的波段选择方法相比,所选波段的目标检测性能平均提升5%;利用粒子群算法优化波段搜索过程,其波段搜索时间较传统的波段搜索方法缩短了一个数量级。新方法具有较强的针对性,能够在较短的时间内,选出具有较高目标检测性能的波段组合,达到降低数据维数及改善目标检测效果的目的。

To solve the problem of low detection efficiency of present hyperspectral band selection methods, a new Band Selection Based on Particle Swarm Optimization （BS-PSO）） was proposed in this paper. First, the we used the area under the Receiver Operating Characteristic （ROC） curve as a band selection criterion and constructed a fitness function based on this criterion. Then, we used particle swarm optimization to optimize the band selection. Finally, the band subset with better target detection result can be obtained. Experimental results on a real world hyperspectral data indicate that, （ 1 ） using AUC as the index of band selection can effectively select the bands with high perform- ance of target detection, compared with other typical band selection methods, the selected bands make an improve- ment of 5% in detection performance; （2） Compared with other typical band search method, the use of particle swarm optimization in band search processing can shorten the search time by an order of magnitude. The bands with higher detection performance can be selected in a short period of time by the proposed method. Both data dimension- ality reduction and improvement of target detection result can be realized at the same time. Experimental results on a real world hyperspectral data show the efficiency of the proposed method.