基于改进TCN模型的野外运动目标分类  

Classification of Moving Targets in Fields Based on Improved TCN Model

作　　者：范裕莹 李成娟 易强 李宝清[1] FAN Yuying;LI Chengjuan;YI Qiang;LI Baoqing(Key Laboratory of Microsystem Technology,Shanghai Institute of Microsystem and Information Technology,Chinese Academy of Sciences,Shanghai 201800,China;University of Chinese Academy of Sciences,Beijing 100049,China)

机构地区：[1]中国科学院上海微系统与信息技术研究所微系统技术重点实验室,上海201800 [2]中国科学院大学,北京100049

出　　处：《计算机工程》2021年第9期106-112,共7页Computer Engineering

基　　金：微系统技术重点实验室基金(6142804190304)。

摘　　要：野外运动目标信号的背景噪声复杂,利用单模态声音信号进行野外目标分类识别率低且鲁棒性差。针对该问题,提出一种基于声震多模态融合的网络模型。借鉴DenseNet网络密集连接的思想改进时域卷积网络,从而对四通道声音信号和单通道震动信号进行深层次的特征提取,并将两种信号相互融合得到最终的目标分类结果。同时,使用带权重的损失函数解决因数据不均衡导致的泛化性能差的问题。实验结果表明,融合网络的识别准确率达到92.92%,较单模态输入网络提高了6.63%~9.46%,且该网络具有较强的鲁棒性。Due to the complex background noises of the moving target signals in the fields,the classification methods of moving targets based on single-mode sound signals is limited by the low recognition rate and poor robustness.To address the problem,a network model is proposed based on multi-modal fusion of sound and vibration signals.The new model is constructed based on the Temporal Convolutional Network(TCN)model,which is modified by using the idea of dense connection in DenseNet.On this basis,the deep features of the four-channel sound signals and the single-channel vibration signals are extracted.Then the two kinds of signals are fused to obtain the final target classification result.At the same time,this paper uses the weighted loss function to solve the poor generalization performance caused by data imbalance.Experimental results show that the recognition accuracy of the proposed model reaches 92.92%,which is 6.63%~9.46% higher than that of the single-mode input network models,and the model has higher robustness.

关 键 词：声震信号 多模态融合 时域卷积网络 密集连接 运动目标分类 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]