A novel adaptive immune multi-filed feature selection algorithm (AIFSA) for underwater acoustic tragets classification is proposed. The AIFSA is proposed to address the problem that the classification performance in classifying underwater acoustic targets declines as the dimension of feature set increases, and that classifying underwater acoustic targets is a small-sample-size classification problem. The AIFSA generates an initial population using prior knowledge, and then generates new generations through repetitive application of mutation, crossover, and adaptive immune operator. In each iteration, individuals with less number of features and with high classification accuracy are given higher fitness values. The advantages of AIFSA include; using of prior knowledge, fast convergence, and small size of optimal feature subset. The multi-field features are extracted from 4 classes of underwater targets and used in feature selection and classification extracted. Experimental results show that the AIFSA can select the subset of efficient features, and there is only a small decline in the accuracy of SVM classifier when the number of features is decreased about 60%. Compared with the genetic algorithm, the AIFSA is more stable and achieves better converge speed, and the feature subset obtained by AIFSA achieves better class ification performance and generalizability.%针对水声目标小样本识别中样本数目有限而特征数目不断增加,导致分类系统分类性能下降的问题,提出了一种新的自适应免疫特征选择算法(AIFSA).该算法先利用先验知识生成初始种群,接着利用交叉、变异和新的自适应免疫算子指导种群进化,每代中对分类贡献大且选择特征数目少的个体适应度值高.AIFSA具有可以利用先验知识、收敛速度快以及优化特征子集维数小的优点.提取了实测4类水声目标的多域特征,进行特征选择和分类识别仿真实验,结果表明:AIFSA可以选择有效特征子集,在特征维数下降60％的情况下,支持向量机分类器的平均正确分类率下降很小；AIFSA与标准遗传算法相比,收敛快、稳定,所得优化特征子集具有更高的正确分类率和更好的范化性能.
展开▼
