复合深度神经网络在直升机声目标识别中的研究  

Research on combined deep neural network in acoustic helicopter target recognition

作　　者：郭洋 周翊[2] 管鲁阳[1] 鲍明[1] GUO Yang;ZHOU Yi;GUAN Luyang;BAO Ming(Key Laboratory of Noise and Vibration Research,Chinese Academy of Sciences,Beijing 100190,China;Chongqing University of Posts and Telecommunications,Chongqing 400065,China)

机构地区：[1]中国科学院噪声与振动重点实验室(声学研究所),北京100190 [2]重庆邮电大学,重庆400065

出　　处：《应用声学》2019年第1期8-15,共8页Journal of Applied Acoustics

摘　　要：针对直升机探测中目标运动过程连续识别的鲁棒性问题,提出了一种基于复合深度神经网络的直升机声学特征提取和识别框架。复合深度神经网络由卷积神经网络和长短时记忆神经网络以并行结构组合,进行直升机声学特征的优化,完成直升机类型识别。针对直升机声信号特性,对卷积神经网络进行了改进,使得该复合深度神经网络在信号短时谱基础上优化声信号特征表征并提取前后帧之间的相关信息,弥补通常声目标识别方法不能充分利用目标信号时间历程信息的缺陷。真实外场实验数据测试结果显示:相较于传统识别方法,该算法显著提升了直升机进入有效探测范围后连续识别的鲁棒性和目标识别正确率。To improve the performance of continuous recognition of acoustic targets,a novel combined deep neural network was proposed to extract features and recognize helicopters.In the framework of the combined deep neural network,a modified convolutional neural network and a long short-term memory neural network were combined primarily in a parallel manner to optimize the representation of helicopter’s acoustic characteristics and implement helicopter type recognition.The optimized feature pattern extracted by the combined deep neural network included the current spectral characteristics and time series information hidden in the input short-term spectrum.It was designed to overcome the lack of time information of the target signal in the conventional acoustic target recognition methods.The proposed method was tested using the real helicopter acoustic signals from the field experiments.The results indicate that the proposed combined deep neural network significantly improves the recognition accuracy and the robustness of the continuous acoustic target recognition when the target is within the detection range.

关 键 词：深度神经网络 声目标识别 直升机识别 

分 类 号：TB535.3[理学—物理]