CNN海况等级分类方法的性能  

Performance of CNN Sea State Class Classification Method

作　　者：张梦雨 王中训[1] 李飞 刘宁波 董云龙 ZHANG Meng-yu;WANG Zhong-xun;LI Fei;LIU Ning-bo;DONG Yun-long(School of Physics and Electronic Information,Yantai University,Yantai 264005,China;91423 Chinese People's Liberation Army,Yantai 265202,China;Information Fusion Institute,Naval Aviation University,Yantai 264001,China)

机构地区：[1]烟台大学物理与电子信息学院,山东烟台264005 [2]中国人民解放军91423部队,山东烟台265202 [3]海军航空大学信息融合研究所,山东烟台264001

出　　处：《烟台大学学报（自然科学与工程版）》2023年第2期196-203,共8页Journal of Yantai University(Natural Science and Engineering Edition)

基　　金：国家自然科学基金资助项目(61871392,62101583)。

摘　　要：采用经典的LeNet网络对实测IPIX雷达数据、SPPR-50雷达数据、一型试验雷达数据通过多种组合方式进行交叉实验验证,分析训练集和测试集采用相同类型实测数据、不同类型实测数据时对分类准确率的影响,从而分析网络对雷达信号层数据处理的泛化能力。实验结果表明,不同类型的实测数据对分类准确率影响较为明显,并且当数据具有一定的相似性时分类效果显著。因此,在样本数据集不足时,可考虑通过迁移学习来加快并优化模型的学习效率,提高神经网络模型分类的准确率。The classical LeNet is used for cross experimental verification of the measured IPIX data,SPPR-50 data,and Type 1 test data through various combinations,to analyze the impact on classification accuracy when the training set and testing set adopt the same and different types of measured data,and further to analyze the generalization ability of the network for radar signal layer data processing.A series of simulations show that the impact on the classification accuracy of different types of measured data is obvious,and when the data have a certain similarity,the classification effect is particularly significant.Therefore,when the sample data set is insufficient,transfer learning can be considered to accelerate and optimize the learning efficiency and improve the accuracy of the neural network model classification.

关 键 词：卷积神经网络 雷达数据 分类 迁移学习 

分 类 号：TN957.51[电子电信—信号与信息处理]