首页 | 本学科首页   官方微博 | 高级检索  
     

基于特征增强的SAR图像舰船小目标检测算法
引用本文:严春满,王铖.基于特征增强的SAR图像舰船小目标检测算法[J].控制与决策,2023,38(1):239-247.
作者姓名:严春满  王铖
作者单位:西北师范大学 物理与电子工程学院,兰州 730070
基金项目:国家自然科学基金项目(61961037);甘肃省教育厅产业支撑计划项目(2021CYZC-30).
摘    要:针对合成孔径雷达(SAR)图像中小目标舰船检测困难的问题,提出基于单次多盒检测器的一种特征增强小目标检测算法.首先提出一种混合多特征提取模块,采用并行的普通卷积、不同空洞率的空洞卷积以及非对称卷积形成与舰船目标相匹配的感受野,以提高浅层网络对复杂形状小目标的特征提取能力;然后提出一种邻近多特征融合模块,将特征信息进行更科学的深层次融合,对小目标特征进一步增强;最后根据SAR图像单通道的特性,缩减特征提取网络VGG-16的冗余特征通道.在公开的SSDD数据集上与其他检测算法进行对比实验,实验结果表明,所提出方法将平均精确度提升至93.44%,检测速度提升至41.8FPS,参数量减少为18.74M,综合性能优于其他检测算法.

关 键 词:合成孔径雷达  小目标检测  舰船  卷积神经网络  单次多盒检测器  特征增强

A ship small target detection algorithm based on feature enhancement in SAR image
YAN Chun-man,WANG Cheng.A ship small target detection algorithm based on feature enhancement in SAR image[J].Control and Decision,2023,38(1):239-247.
Authors:YAN Chun-man  WANG Cheng
Affiliation:College of Physics and Electronic Engineering,Northwest Normal University,Lanzhou 730070,China
Abstract:To solve the problem of small target ship detection in synthetic aperture radar (SAR) image, a feature enhanced small target detection algorithm based on the single shot multi box detector is proposed. The algorithm proposes a hybrid feature extraction module, which uses general convolution, dilated convolution with different dilation ratio and asymmetric convolution to form a receptive field matching the ship target, so as to improve the feature extraction ability of shallow network for small targets with complex shape. Then, a neighbor feature fusion module is proposed, which integrates the feature information more scientifically and deeply, and further enhances the features of small targets. Finally, according to the characteristics of the single channel of the SAR image, redundant feature channels of the feature extraction network VGG-16 are reduced. It is compared with other detection algorithms on the public SSDD data set. The experimental results show that the proposed method improves the average accuracy to 93.44$%$, the detection speed is improved to 41.8FPS, and the number of parameters is reduced to 18.74M, which is superior to other detection algorithms in comprehensive performance.
Keywords:
点击此处可从《控制与决策》浏览原始摘要信息
点击此处可从《控制与决策》下载全文
设为首页 | 免责声明 | 关于勤云 | 加入收藏

Copyright©北京勤云科技发展有限公司  京ICP备09084417号