遗传与互信息混合算法用于水声目标特征选择

特征选择是目标分类的一项重要步骤，直接影响到分类器的设计和性能。本文利用实际水声目标辐射噪声数据，对遗传算法和互信息算法两种特征选择方法分别作了分析。在特征维数较大的情况下，两种方法都需要很长的计算时间，为此，提出一种遗传与互信息混合算法，旨在降低计算时间。最后，分类器用三种选择后的特征子集作为输入进行分类，并与任意选择的特征子集作为输入的分类结果作了比较。

Hybrid Genetic /Mutual Information Algorithm Used for Underwater Acoustic Target Feature Selection

Feature selection is an important process in a target classification program and directly affects the design and performance of the classifier. Our research will analyze the two methods, genetic algorithm and mutual information algorithm, used for feature selection of a realistic underwqter target-radiated noise data set. In the high feature dimemsion case, they both need very long computation time. Therefore a hybrid genetic/mutual information algorithm is proposed in order to reduce the computation time. At last targets are classified by a classifier with the feature subsets selected by the three algorithms. The results are compared each other and to that for a feature subset selected arbitrarily.