基于海天线引导与卷积神经网络的舰船目标检测  

Ship Target Detection Based on Sea Antenna Guidance and Convolution Neural Network

作　　者：梅升阳 田金文 MEI Shengyang;TIAN Jinwen(National Key Laboratory of Science and Technology on Multi-spectral Information Processing Technology,School of Automation,Huazhong University of Science and Technology,Wuhan 430074)

机构地区：[1]华中科技大学自动化学院多谱信息处理技术国家级重点实验室,武汉430074

出　　处：《计算机与数字工程》2018年第4期721-726,共6页Computer & Digital Engineering

基　　金：国家自然科学基金项目(编号:61273279)资助

摘　　要：红外舰船目标检测在自动目标识别系统中具有重要作用。海天背景下的舰船目标检测通常可以先检测海天线,从而减小目标检测的计算量,提高目标检测的效率。论文结合海天线的分界特性与直线特性,提出基于多特性融合的海天线检测方法,取得较好的效果。在海天线检测基础上,将卷积神经网络的方法引入舰船检测任务中来。通过自行设计网络结构,自主标记训练样本,实现基于卷积神经网络的舰船目标检测方法。实验表明,海天线引导降低了神经网络算法的计算量,使得检测过程更具有实用性。神经网络的引入也让舰船目标检测任务有了更通用更简单的方案。Infrared ship target detection plays an important role in automatic target recognition system.Shipboard target detection in the background of sea sky can usually detect the sea antenna,thus reducing the calculation of the target detection,improving the efficiency of target detection.Based on the boundary characteristics and the straight line characteristics of the sea antenna,this paper proposes a sea antenna detection method based on multi-characteristic fusion,and obtains good results.On the basis of sea antenna detection,the convolution neural network method is introduced into the ship detection task.Through the self-design of the network structure and self-labeled training samples,ship target detection method based on the convolution neural network is achieved.Experiments show that the sea antenna guidance reduces the computational complexity of the neural network algorithm,making the detection process more practical.The introduction of the neural network also allows the ship target detection task to have a more general and simpler solution.

关 键 词：舰船目标检测 海天线 卷积神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]