摘要
为了提升合成孔径雷达(Synthetic Aperture Radar,SAR)图像舰船目标检测的精度和速度,对卷积神经网络(Convolutional Neural Network,CNN)在SAR图像舰船目标检测上进行了研究。通过改进OTSU方法对SAR图像进行分割,并且用最小外接矩形将疑似目标标记出来;依据矩形中心在原始图像上提取出固定大小区域作为候选区域;将提取的目标通过训练好的卷积神经网络进行判定,去除虚警目标并将检测结果在原图中标记出来。实测数据的实验结果表明,该算法在降低虚警的同时提升了检测速度。
In order to improve the ship targets detection precision and speed for SAR image,the convolution neural network was studied in ship target detection for SAR images. Firstly,segmentation the SAR image by using the improved OTSU method,and mark the suspected target by the minimum bounding rectangle;secondly,according to the rectangular center extract the fixed size region on the original image as a candidate area;finally,determine the extracted target by trained convolution neural network,remove the false target and mark the detection result on the original image. The experimental results by the measured data show,this algorithm improves the detection speed and reduces false alarm at the same time.
作者
曲长文
刘晨
周强
李智
李健伟
QU Chang-wen;LIU Chen;ZHOU Qiang;LI Zhi;LI Jian-wei(Naval Aviation University,Yantai 264001,China)
出处
《火力与指挥控制》
CSCD
北大核心
2019年第1期40-44,共5页
Fire Control & Command Control
基金
国家自然科学基金资助项目(60874112
61571454)
关键词
合成孔径雷达
卷积神经网络
目标检测
图像分割
候选区域提取
synthetic aperture radar
convolutional neural network
target detection
image segmentation
candidate area extraction
作者简介
曲长文(1963- ),男,山东济南人,教授,博士生导师。研究方向:信息融合,雷达成像,阵列信号处理,电子对抗。
