CASIA OpenIR  > 毕业生  > 硕士学位论文
基于深度学习的遥感图像飞机检测与识别研究
蔡健
Subtype硕士
Thesis Advisor杨一平 ; 马雷
2019-05-23
Degree Grantor中国科学院大学
Place of Conferral中国科学院自动化研究所
Degree Name工程硕士
Degree Discipline计算机技术
Keyword遥感图像 目标检测 深度学习 旋转矩形框 显著性 飞机识别
Abstract

遥感图像飞机检测识别是遥感目标检测的研究热点之一。随着卫星传感器分辨率的不断提高,图像细节越来越丰富,使高精度的飞机检测与识别成为可能。同时,仅依靠人工判读难以满足海量遥感图像中飞机目标检测的需求,因此基于深度学习的飞机检测与识别工作越来越重要。

本文在深入分析国内外飞机检测识别相关文献的基础上,总结已有工作的不足,围绕提高飞机检测效率和精度展开研究。主要研究内容包含以下几点:

1. 飞机目标显著性区域检测算法研究

为了提高飞机目标检测的速度,实现了一个基于Shuffle-Net的遥感图像飞机显著性区域过滤算法,该算法具有更少的参数、更快的检测速度和更高的精度。

2. 基于朝向矩形框的飞机检测算法研究

针对遥感图像中飞机刚体变换的性质,提出了一种新的朝向矩形框目标表示方法,并结合经典检测模型,提出了基于朝向矩形框的飞机检测模型(OBB-CNN)。与之前的方法相比,所提出的方法不仅能够提供飞机的位置信息,还能够估计飞机的朝向角度,且有着更高的精度。

3. 飞机细粒度分类算法研究

为了实现飞机型号的细粒度分类,本文建立了一个遥感图像飞机型号级数据集(FGRSIA2019),并根据朝向矩形框检测模型的检测结果,设计了一个飞机角度对齐的细粒度分类模型,与不带角度对齐的模型相比,所提出的方法有着更高的分类准确率。

Other Abstract

Aircraft detection and recognition in remote sensing images is one of the research hotspots of remote sensing object detection. With the increase of the resolution of satellite sensors, the image details are becoming more and more abundant, which makes high-precision aircraft detection and recognition possible. At the same time, the accumulated massive images are difficult for human to interpret. Therefore, deep-learning based methods are becoming more and more important.

After an in-depth analysis of the present relevant literatures on aircraft detection and recognition algorithms and summaries of the shortcomings existing work, this paper conducts research on improving the efficiency and accuracy of aircraft detection. The main research content includes the following points:

1. Research on object saliency detection 

In order to improve the speed of objection detection, a Shuttle-Net-based remote sensing image aircraft saliency region filtering algorithm is implemented. The designed model has fewer parameters, faster detection speed and higher accuracy.

2. Research on oriented-bounding-box based aircraft detection

with the nature of rigid transformation of aircraft in remote sensing images, a new object representation method of oriented bounding box is proposed. Combined with the classical detection model, a new aircraft detection model is proposed. Compared with previous methods, the proposed method can not only give the position information of the aircraft, but also give the orientation angle of the aircraft and has higher precision.

3. Research on Fine-grained aircraft classification

In order to realize the fine-grained classification of aircraft models, a new remote sensing image military aircraft data set is established. According to the detection result of the oriented-bounding-box based detection model, a fine-grained classification model of aircraft angle alignment is designed. Compared with the model without angle alignment, the proposed method has higher classification accuracy.

Pages86
Language中文
Document Type学位论文
Identifierhttp://ir.ia.ac.cn/handle/173211/23813
Collection毕业生_硕士学位论文
Recommended Citation
GB/T 7714
蔡健. 基于深度学习的遥感图像飞机检测与识别研究[D]. 中国科学院自动化研究所. 中国科学院大学,2019.
Files in This Item:
File Name/Size DocType Version Access License
Thesis.pdf(9740KB)学位论文 暂不开放CC BY-NC-SAApplication Full Text
Related Services
Recommend this item
Bookmark
Usage statistics
Export to Endnote
Google Scholar
Similar articles in Google Scholar
[蔡健]'s Articles
Baidu academic
Similar articles in Baidu academic
[蔡健]'s Articles
Bing Scholar
Similar articles in Bing Scholar
[蔡健]'s Articles
Terms of Use
No data!
Social Bookmark/Share
All comments (0)
No comment.
 

Items in the repository are protected by copyright, with all rights reserved, unless otherwise indicated.