光学学报, 2020, 40 (1): 0111020, 网络出版: 2020-01-06   

基于改进旋转区域生成网络的遥感图像目标检测 下载: 2144次

Object Detection of Remote Sensing Image Based on Improved Rotation Region Proposal Network
作者单位
武汉大学电子信息学院, 湖北 武汉 430072
摘要
为了实现遥感图像中目标的快速准确检测,解决遥感图像目标带有旋转角度的问题,在卷积神经网络理论的基础上,将旋转区域网络生成融入到Faster R-CNN网络中,提出了一种基于Faster R-CNN改进的遥感图像目标检测方法。相对于主流目标检测方法,本文算法针对遥感图像中的大多数目标都具有方向性不定且相对聚集的特点,在区域候选网络中加入了旋转因子,以便能够生成任意方向的候选区域;同时,在网络的全连接层之前增加一个卷积层,以降低其特征图参数,增强分类器的性能,避免出现过拟合。将本文算法与几种主流目标检测方法进行对比分析后可知,本文算法因融合了多尺度特征及旋转区域网络的卷积神经网络所提取的特征,能得到更好的检测结果。
Abstract
In this study, the integration of the rotation region proposal network with Faster R-CNN network along with an improved remote sensing image object detection method based on the convolutional neural network is proposed. The aim is two-fold: 1) to realize rapid and precise detection of remote sensing image objects; 2) to address the problem caused by objects with rotated angle. Compared to the mainstream target detection methods, the proposed method introduces the rotation factor to the region proposal network and generates proposal regions with different directions, aiming at the characteristics of variable direction and relative aggregation of most targets in the remote sensing image. The addition of a convolution layer before the fully connected layer of the Faster R-CNN network has the advantages of reducing the feature parameters, enhancing the performance of classifiers, and avoiding over-fitting. Compared with the state-of-the-art object detection methods, the proposed algorithm is able to combine the features extracted by the convolutional neural network in the rotation region proposal network with the multi-scale features. Therefore, significant improvement in remote sensing image object detection can be achieved.

戴媛, 易本顺, 肖进胜, 雷俊锋, 童乐, 程志钦. 基于改进旋转区域生成网络的遥感图像目标检测[J]. 光学学报, 2020, 40(1): 0111020. Yuan Dai, Benshun Yi, Jinsheng Xiao, Junfeng Lei, Le Tong, Zhiqin Cheng. Object Detection of Remote Sensing Image Based on Improved Rotation Region Proposal Network[J]. Acta Optica Sinica, 2020, 40(1): 0111020.

本文已被 13 篇论文引用
被引统计数据来源于中国光学期刊网
引用该论文: TXT   |   EndNote

相关论文

加载中...

关于本站 Cookie 的使用提示

中国光学期刊网使用基于 cookie 的技术来更好地为您提供各项服务,点击此处了解我们的隐私策略。 如您需继续使用本网站,请您授权我们使用本地 cookie 来保存部分信息。
全站搜索
您最值得信赖的光电行业旗舰网络服务平台!