基于多维特征融合的海面目标检测  

Sea Surface Target Detection Based on Multi-dimensional Feature Fusion

作　　者：黄胜彬 潘大鹏[1] 陈涛[1] HUANG Shengbin;PAN Dapeng;CHEN Tao(Harbin Engineering University,Harbin 150001,China)

机构地区：[1]哈尔滨工程大学,黑龙江哈尔滨150001

出　　处：《舰船电子对抗》2024年第3期84-91,共8页Shipboard Electronic Countermeasure

摘　　要：为提高海面目标的检测概率,提出一种多维特征融合的目标检测方法。首先,在时域的基础上,提取赫斯特(Hurst)指数和信息熵(IE)2种特征。其次,在频域的基础上,提取频域方差与均值比(FVAR)作为特征。然后,在时频域的基础上,提取短时傅里叶变换(STFT)后的时频图作为特征。最后,使用2支改进的卷积神经网络(CNN)分别训练一维特征和二维特征,之后联合2支网络,得到最终的目标检测结果。进一步与传统卷积神经网络作对比,结果显示在相同训练条件下,所提方法的计算时间缩短40%以上。To improve the detection probability of sea surface targets,a multi-dimensional feature fusion method for target detection is proposed.Firstly,based on the time domain,two features:Hurst exponent and information entropy(IE)are extracted.Secondly,based on the frequency domain,the frequency variance to average ratio(FVAR)is extracted as a feature.Then,based on the time-frequency domain,the time-frequency map after short time fourier transform(STFT)is extracted as the feature.Finally,two improved convolutional neural networks(CNNs)are used to train one-dimensional and two-dimensional features respectively.The two networks are then combined to obtain the final target detection result.Further comparison with traditional convolutional neural networks shows that the proposed method reduces computation time by more than 40%under the same training conditions.

关 键 词：海杂波 目标检测 特征融合 

分 类 号：TN971[电子电信—信号与信息处理]