【作者】 凡志邈； 李海林； 夏伟杰； 杨添文； 赵世瑾； 鲍育鸿；

【Author】 FAN Zhi-miao;LI Hai-lin;XIA Wei-jie;YANG Tian-wen;ZHAO Shi-jin;BAO Yu-hong;Key Laboratory of Fundamental Science for National Defense-Advanced Design Technology of Flight vehicle,Nanjing University of Aeronautics & Astronautics;Ministry of Education Key Lab of Structure Mechanics and Control for Aircraft,Nanjing University of Aeronautics & Astron Underwater Autics;

【机构】 南京航空航天大学国防先进飞行器设计技术基础科学重点实验室； 南京航空航天大学飞机结构力学与控制教育部重点实验室；

【摘要】 成像声纳能够提供前方区域较高清晰度的水声图像,对图像中的目标进行分类识别是水声信号处理的关键难点之一,也是成像声纳走向智能化的关键技术之一。针对这一问题,本文将深度学习引入声呐图像的识别中,首先用声呐获取目标图像,构成深度学习的训练样本;然后选择inception_v3模型实现目标图像的分类,该模型不仅可以降低参数数量,也可以减轻过拟合现象;最后选用当前流行的检测模型——SSD模型来实现目标图像的检测。测试集上的验证结果表明了本文方法的有效性。更多还原

【Abstract】 Imaging sonar can provide high-resolution acoustic images in the front area.It is one of the key difficulties in underwater acoustic signal processing to classify and recognize the targets in the image,and it is also one of the key technologies for imaging sonar to become intelligent.In order to solve this problem,this paper introduces deep learning into sonar image recognition.Firstly,sonar is used to acquire target image and form training samples for deep learning.Then inception_v3 model is selected to classify target image.This model can not only reduce the number of parameters,but also reduce the phenomenon of over-fitting.Finally,SSD model,a popular detection model,is selected to realize the target detection.The validation results on the test set show the effectiveness of the proposed method.更多还原