摘要
针对Faster R-CNN算法检测舰船目标存在的不足,提出基于深度特征金字塔和级联检测器的舰船检测算法。先利用小目标数据增强算法对数据进行扩充,使检测模型学习足够的特征;再使用深度特征金字塔网络改进原目标检测算法的特征提取网络,抑制相干斑噪声,有效提取舰船特征;并根据合成孔径雷达(SAR)图像中舰船目标稀疏的特点使用级联结构调整网络。基于上述改进,选取舰船目标检测数据集中部分图像及2月份渤海湾的SAR图像进行实验,实验结果表明:所提算法均取得了良好的检测效果,证明了所提算法的有效性。
Faster R-CNN algorithm cannot achieve accurate ship detection.Therefore,a ship detection algorithm based on a deep feature pyramid and cascade detector is proposed in this study.First,the small-target data enhancement algorithm is used for expanding the data to ensure that sufficient features are learned by the detection model.Then,the deep feature pyramid network is used for improving the feature extraction network of the original target detection algorithm,suppressing the coherent speckle noise,and effectively extracting the ship features.Further,a cascading structure is adopted to adjust the improved network according to the sparse features of the ship targets obtained from the synthetic aperture radar(SAR)images.Based on the aforementioned improvements,some images from the ship target detection dataset and the SAR images of the Bohai Bay captured in February are selected for performing the experiments.Experimental results show that,the proposed algorithm achieves good detection results,proving its effectiveness with respect to ship detection.
作者
赵云飞
张宝华
张艳月
谷宇
王月明
李建军
赵瑛
Zhao Yunfei;Zhang Baohua;Zhang Yanyue;Gu Yu;Wang Yueming;Li Jianjun;Zhao Ying(School of Information Engineering,Inner Mongolia University of Science and Technology,Baotou,Inner Mongolia 014010,China)
出处
《激光与光电子学进展》
CSCD
北大核心
2020年第12期229-234,共6页
Laser & Optoelectronics Progress
基金
国家自然科学基金(61962046,61663036,61841204)
内蒙古杰青培育项目(2018JQ02)
内蒙古草原英才、内蒙古青年科技创新人才项目(第一层次)
内蒙古自治区自然科学基金(2015MS0604,2018MS06018)
内蒙古自治区高等学校科学技术研究项目(NJZY145)
内蒙古自治区研究生科研创新资助项目(S20191187Z)。
关键词
图像处理
目标检测
数据增强
深度特征金字塔
级联
合成孔径雷达图像
image processing
object detection
data enhancement
deep feature pyramid
cascade
synthetic aperture radar images
