摘要
针对合成孔径雷达(SAR)图像中的舰船检测与分类问题,常规的图像处理技术或者机器学习方式难以准确检测出海上舰船的类别以及当前舰船运行状态。因海上舰船目标具有相对于SAR图像尺寸较小、方位难以测算以及容易与其他目标混淆的特点,针对上述问题设计了端到端的海上舰船分类与状态感知模型,加入特征金字塔以达到拥有对于微小目标提取其深度特征,同时又保留其相对位置的目的;使用残差结构以解决特征融合网络层数增加导致的梯度消失问题;最后加入舰船状态感知模块,使其最终可以得到海上舰船目标相对于图像的角度值。使用公开SAR卫星图像进行了多次实验,最终体现出提出的端到端的模型具有较高的识别率以及良好的舰船状态估计能力。
Limited by ship detection and classification in synthetic aperture radar(SAR)images,the types and operating status of marine ships in sea are difficult to be accurately recognized by conventional image processing techniques or machine learning methods. Compared with the original SAR images,the marine ship targets in sea are smaller in size,difficult to measure in azimuth,and tend to be confused with other targets. In order to overcome these challenges,this paper proposes an end-to-end marine ship classification and state perception model. We construct a feature pyramid to extract the depth features of tiny targets while retaining their relative positions. Meanwhile,the residual structure is used to solve the gradient disappearing problem caused by the increase in the layer number of the feature fusion network. At the end,a ship status perception module is composed to derive the relative angle values of the ship targets in SAR images. Several experiments are conducted with the public SAR image dataset,and the results prove that the proposed end-to-end model has a high recognition rate and good ship state estimation capability.
作者
崔雷
庄磊
张泽栋
魏松杰
CUI Lei;ZHUANG Lei;ZHANG Zedong;WEI Songjie(Shanghai Institute of Satellite Engineering,Shanghai 200240,China;School of Computer Science and Engineering,Nanjing University of Science and Technology,Nanjing 210094,Jiangsu,China)
出处
《上海航天（中英文）》
CSCD
2022年第6期75-83,共9页
Aerospace Shanghai（Chinese&English）
基金
国家自然科学基金(61802186,61472189)
上海航天科技创新基金(SAST2019-033)。
关键词
合成孔径雷达图像
目标检测
特征叠加网络
深度神经网络
synthetic aperture radar(SAR)image
target detection
feature stacking network
deep neural network
