中文摘要：

特征选择是复杂模式分类系统中重要预处理过程。针对filter模式下遗传算法特征选择精度不高，wrapper模式特征选择时间代价较高的缺点，提出了一种新的特征选择算法。该算法设计了搜索性能较好的链式智能体遗传算法为搜索算法，引入多个评价准则进行轮询式选择．实验将算法与filter模式下多种单准则特征选择算法以及wrapper模式下特征选择算法进行了比较。实验结果表明，此算法具有比filter模式下单评价准则选择精度更高的特点，同时选择时间代价远远低于wrapper此模式下的特征选择算法，因此，该算法可用于设计实用高识别正确率的模式分类系统。

英文摘要：

Feature selection is a pretreatment process for complex pattern classification systems. According to the low precision of feature selection under filter mode and high time cost of feature selection under wrapper mode, a new feature selection algorithm was proposed. This algorithm designed chain-like agent genetic algorithm as searching algorithm, introduced several evaluation criteria for poll mode selection. The experiments were done to compare this algorithm and several other feature selection algorithms. The experimental results show that this algorithm can obtain better precise selection result than several single evaluation criterion feature selection algorithms under filter mode, and less selection time cost than feature selection algorithm under wrapper mode. Therefore, this algorithm can be used for designing feasible pattern classification system with high recognition rate.