用于水声目标分类的加权免疫克隆样本选择算法

A new weighted immune clonal algorithm for underwater acoustic target classification

提出了一种新的用于水声目标分类的加权免疫克隆样本选择算法(weight Immune Clonal Instance Selection,wI-CISA)。算法利用Adaboost算法给予每个样本一权值,每代中根据样本权值计算抗体亲合度和克隆数,且根据克隆复制、抗体更新(海明距离交叉和加权简化最近邻变异)和克隆选择操作指导种群进化。提取了实测3类水声目标的时域波形结构特征、小波分析特征和听觉谱特征,进行样本选择和分类仿真实验,结果表明:wICISA可以选出有效样本子集,使样本数目减少82%左右,并且支持向量机分类器的正确分类率能提高约2%;wICISA具有较好的收敛性、稳定性,所得优化样本子集具有较好的泛化能力且能明显减少分类的时间。

A new adaptive weighted immune clonal instance selection algorithm （wlCISA） for underwater acoustic tar- gets classification is proposed in this paper. The wICISA gets the weight of instances through the Adaboosting algorithm, in each iteration, the affinity of antibodies and the number of clones of antibodies are calculated with the weight of in- stance. And then, new generations are generated through repetitive application of clonal copy, antibody update （Ham- ming distance crossover and Weight Reduced Nearest Neighbor mutation） and clonal selection operation. The time wave structure features, wavelet analysis features and auditory spectrum features are extracted from 3 classes of un- derwater targets, and are used in instance selection and classification experiments. Experimental results show that wI- CISA can select the subsets of efficient instances with about 2% increase in the accuracy of SVM classifier when the number of instances decreases by about 82%. wICISA has good convergence and stability, and the instance subset ob- tained by wICISA achieves with good generalizability, and it can reduce the classification time remarkably.