摘要
为了发挥合成孔径雷达(synthetic aperture radar,SAR)检测系统在军事和航海领域的重要作用,提升高分辨率SAR图像目标检测的精度,提出了基于贝叶斯分类器与轻量伪孪生网络(pseudo-siamese network,PSN)的SAR舰船目标检测算法。针对大部分卷积神经网络(convolutional neural networks,CNN)目标检测模型因参数多、模型大而不利于移动端嵌入式使用的特点,利用视觉显著结合贝叶斯分类,获得舰船疑似切片,降低数据量。引入一维直方图信息,在PSN的基础上自主设计了SAR舰船目标检测框架。基于HRSID数据集进行检测实验,并与各种基于深度学习的目标检测算法进行比较,检测结果表明了所构建模型的有效性。
In order to make full use of synthetic aperture radar(SAR)detection system in military and navigation,a SAR ship target detection algorithm based on Bayesian classifier and lightweight pseudo-siamese network(PSN)was proposed,which could improve the accuracy of high-resolution SAR image target detection.In view of most convolutional neural networks(CNN)target detection models have many parameters and large models,which are difficult to apply to small mobile devices,the visual saliency technology and Bayesian classifier were used to obtain suspected target slices and reduce the amount of data.By introducing one-dimensional histogram information,a SAR ship target detection framework was designed on the basis of PSN.The experiment was based on the HRSID standard data set and was compared with various target detection algorithms which applied the deep learning.The target detection result shows the validity of the proposed model.
作者
张潘东
韩玉兵
管礼
ZHANG Pandong;HAN Yubing;GUAN Li(School of Electronic and Optical Engineering,Nanjing University of Science and Technology,Nanjing 210094,China)
出处
《中国科技论文》
CAS
北大核心
2022年第11期1296-1302,共7页
China Sciencepaper
