中文摘要：

由于水下噪声较为复杂，传统的通过信号处理与变换技术来提取水下目标特征的方法在某些程度上都存在不足，导致水下目标识别效果不理想。考虑到音色在分类识别中显示出的优势，主要对与音色相关的各种特征进行提取，并且针对提取特征维数较高导致的系统结构复杂、运行时间较长的问题，采用遗传算法和误差反向传播神经网络相结合的方法来对特征进行优化选择，并对结果进行了物理解释与分析。

英文摘要：

Since the underwater noise is very complex, the traditional method mainly uses signal processing and the converter technique to extract feature of underwater acoustic targets, but this method has insufficiency in certain degrees, causing low recognition rate of the underwater acoustic target. Considered of the superiority of timbre in the target recognition, kinds of characteristics related to timbre are extracted mainly. What＇s more, a new method is proposed for feature selection that uses Back Propagation Network （BP） combined with genetic algorithm （GA）. With this method the dimension of training data is reduced while the classification accuracy is almost the same. Therefore, the system structure becomes simpler, and the running time is shorter. Finally, the physical interpretation and analysis to the result are done.