s
###
DOI:
飞控与探测:2022,(2):30-37
本文二维码信息
码上扫一扫!
基于改进Faster R-CNN的失效卫星部件检测方法
(1.东南大学 仪器科学与工程学院;2.微惯性仪表与先进导航技术教育部重点实验室;3.上海航天控制技术研究所;4.上海市空间智能控制技术重点实验室)
An Improved Faster R-CNN Detection Method for the Failed Satellite Components
(1.School of Instrument Science and Engineering, Southeast University;2.Key Laboratory of Micro-Inertial Instrument and Advanced Navigation Technology, Ministry of Education;3.Shanghai Aerospace Control Technology Institute;4.Shanghai Key Laboratory of Aerospace Intelligent Control Technology)
摘要
相似文献
参考文献
本文已被:浏览 70次   下载 36
    
中文摘要: 基于光学图像对失效卫星部件的精确检测可以为失效卫星的定位与捕获等任务提供支撑。然而,失效卫星部件多为密集小目标,且其光照条件变化较大,这导致一般主干网络出现特征表征分辨率低,小目标漏检等问题。针对上述问题,提出了一种基于改进Faster R-CNN的失效卫星部件检测方法。该方法在Faster R-CNN的基础上,融合高分辨网络构建新的特征提取主干网络,以获得可靠、高分辨率的特征表达式。其次,在模拟真实空间环境的条件下,利用1:1的嫦娥卫星模型构建了一个信息丰富的失效卫星数据集。用该数据集进行验证,结果表明:本文方法的平均精度为93.6%,其与Faster R-CNN和Cascade R-CNN相比,对小部件检测的准确率与召回率分别平均提高了9.8%与5.4%。该方法可有效检测失效卫星部件。
Abstract:Accurate detection of failed satellite components based on optical images can provide support for aerospace missions such as the location and capture of failed satellites. However, most of the failed satellite components are dense and small targets, and their illumination conditions change greatly, which leads to the problems of low accuracy of feature representation and missing detection of small targets in the general backbone network. To solve the above problems, a failure satellite component detection method based on improved Faster R-CNN is proposed. Firstly, a new feature extraction backbone network based on Faster R-CNN is constructed by fusing high resolution network to obtain reliable and high-resolution feature expression. Secondly, under the condition of simulating the real space environment, a failure satellite dataset with rich information is constructed by using the 1:1 Chang'e satellite model. The results show that the average accuracy of this method is 93.6%. Compared with Faster R-CNN and Cascade R-CNN, the accuracy of small target component detection is improved by 4.2% and the recall is improved by7.8%. This method can effectively detect failed satellite components.
文章编号:20220204     中图分类号:TN911.73; TP391.9    文献标志码:A
基金项目:上海航天科技基金(SAST2020-058);中国航天科技集团公司第八研究院产学研合作基金资助项目
引用文本:
曹 毅,程向红,李丹若,刘宗明,牟金震.基于改进Faster R-CNN的失效卫星部件检测方法[J].飞控与探测,2022,(2):30-37.

分享按钮