摘要
针对传统目标识别方法资源消耗大、精度和可靠性低、泛化能力不强的问题,提出了一种基于改进YOLO(You Only Look Once)模型的舰船目标识别方法。通过精简YOLO模型,设计了一个10层的卷积神经网络用于舰船目标的自动特征提取和分类识别,模型训练过程中引入迁移学习的概念防止模型过拟合并加速模型参数的训练。在自建舰船目标图像测试集上的实验分析结果表明,该方法能够正确识别出航母、除航母外的其余军舰及民船三类舰船目标,识别精度达到93.7%且识别效率较高,验证了所提舰船目标识别方法的有效性。
For the problems of the traditional target recognition method,such as high resource consumption,low precision and reliability,poor generalization ability,a ship target recognition method based on convolutional neural network(CNN) is proposed.By simplifying the You Only Look Once(YOLO) model,a 10-layer CNN model is designed to extract the ship features and recognize different ship targets automatically.In the process of model training,the concept of transfer learning is introduced to prevent model overfitting and accelerate the training of model parameters.The results of experiment on the self-built ship target image testing set show that,this method can correctly recognize three types of ship targets,including the aircraft carriers,the remaining warships except the aircraft carriers,and the civilian ships.The recognition accuracy reaches 93.7 % and the recognition efficiency is high,which verifies the effectiveness of the proposed method.
作者
马啸
邵利民
金鑫
徐冠雷
MA Xiao;SHAO Limin;JIN Xin;XU Guanlei(Department of Navigation,Dalian Naval Academy,Dalian 116018,China)
出处
《电讯技术》
北大核心
2019年第8期869-874,共6页
Telecommunication Engineering
基金
国家自然科学基金资助项目(61471412,61771020)
