中文摘要：

分析了几种常用海洋声学仪器信号的基本特征，提出一种基于误差反向传播（back propagation，BP）神经网络，以实现对信号特征参数进行分类、识别的方法．该方法采用短时傅里叶变换提取信号特征参数，运用Levenberg—Marquardt算法训练BP神经网络．以实测海洋声学仪器信号的特征参数进行训练后，采用实测和仿真样本对BP神经网络的识别能力进行测试．实验结果表明，BP神经网络能够有效地区分不同海洋声学仪器的信号，识别准确率达到95％以上，且虚警率低于5％．该研究成果可用于识别海域中不同海洋声学仪器，检测海洋中声学仪器的工作状态．该识别方法对于其他海洋声信号的识别研究也有一定的参考价值．

英文摘要：

This paper analyzes the basic characteristics of several familiar marine acoustic instruments＇ signals, and presents a BP neural network （BPNN） based method for signal recognition and classification,which uses short time fourier transform（STFT） for characteristics extraction,and Levenberg Marquardt algorithm for BPNN training. After training with real acoustic signals,we evalu- ate the classification ability of BPNN with real and simulated samples. Experimental results show that BPNN is able to categorize dif ferent marine acoustic instruments efficiently,and the recognition accuracy is more than 95 ~ while the false alarm probability is less than 5 ~. In general, this method can be used to identify a variety of marine acoustics instruments and detect their working status. Al-so, it may provide references for recognizing other marine acoustic signals.