深度学习技术及其在舰船目标识别领域的应用  

Deep Learning Technology and Its Application in Ship Target Recognition

作　　者：马啸 邵利民[1] 金鑫[1] 徐冠雷[1] MA Xiao;SHAO Li-min;JIN Xin;XU Guan-lei(Department of Navigation,Dalian Naval Academy,Dalian 116018,China)

机构地区：[1]海军大连舰艇学院航海系

出　　处：《计算机技术与发展》2019年第12期141-147,共7页Computer Technology and Development

基　　金：国家自然科学基金(61471412,61771020)

摘　　要：近年来,中国海洋权益争端日益频繁,对海上舰船目标进行识别和监视显得尤为重要。图像技术的发展为舰船目标识别提供了新的感知来源,随着获取图像数据量的增加,传统通过人工判读识别舰船目标的方法资源消耗大且难以保证目标识别的精度和可靠性,迫切需要引入新的技术和方法以节省人力资源,提高舰船目标识别的精度和可靠性。大数据背景下,深度学习技术在语音识别、图像识别等领域的发展为舰船目标识别技术的突破提供了新思路。文中阐述了具有代表性的深度学习模型,介绍了主流的基于深度卷积神经网络的目标识别方法,将其中较典型的两种目标识别方法Faster RCNN和YOLO应用于舰船目标识别领域,通过客观分析比较两种方法在舰船目标识别中的优劣性。舰船目标识别结果表明,Faster RCNN的准确率和召回率高于YOLO,但其运行效率远低于YOLO的运行效率。由此,提出下一步的工作方向。In recent years,the conflicts of maritime rights and interests have become more and more frequent in China,therefore it is particularly important to recognize and monitor the ship target.The development of image technology provides a new source of perception for ship target recognition.With the increase of image data acquisition,the traditional ship target recognition method by manual interpretation consumes a lot of resources and is difficult to ensure the accuracy and reliability of target recognition,so it is an urgent to introduce new technologies and methods to save human resources and improve the accuracy and reliability of ship target recognition.Under the context of big data,the development of deep learning technology in speech recognition,image recognition and other fields provides a new idea for the breakthrough of ship target recognition.We describe some representative deep learning models and introduce the mainstream methods of target recognition based on deep convolution neural network.Two typical target recognition methods,Faster RCNN and YOLO,are chosen to be applied in ship target recognition.Through objective analysis to compare the advantages and disadvantages of these two methods in ship target recognition,the ship target recognition result shows that the accuracy and recall rate of the Faster RCNN are higher than YOLO,but its operation efficiency is much lower than YOLO.Therefore,the next step of work is proposed.

关 键 词：深度学习 模型 卷积神经网络 目标识别 舰船 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]