遥感图像飞机目标检测与识别

CASIA OpenIR > 毕业生 > 硕士学位论文

	遥感图像飞机目标检测与识别
其他题名	Aircraft Detection and Recognition from Remote-Sensing Images
	巴君
	2016-05
学位类型	工程硕士
中文摘要	随着遥感卫星成像技术的发展，遥感卫星的成像分辨率得到了大幅度的提高。大量遥感卫星的成功发射，产生了海量的遥感数据，很难依靠人工去判读。如何在遥感图像中进行目标检测和识别成为一个具有重要意义的课题。同时，随着深度学习领域相关理论与技术的不断发展，目标检测和识别的相关算法得到了飞速发展，推动了整个领域应用性能的极大提升。本文在深入分析国内外研究现状的基础上，围绕着提高目标检测的精度和效率两方面展开研究，主要工作体现在以下几个方面：遥感图像飞机目标检测算法研究为了提高检测精度和检测速度，提出了基于Single Shot MultiBox Detector （SSD）的遥感图像目标检测算法。为了降低存储空间，同时提出了一个改进版本的检测算法。改进版本相比于基础版本有更快的检测速度和更少的参数个数。遥感图像飞机多类别检测与识别算法研究针对遥感图像多类别检测与识别的特点，在分析现有算法优缺点的基础上，提出了基于SSD的多类别识别算法。与之前的方法相比，所提出的方法有更高的检测精度（mAP）和更快的检测速度。星载图像处理器为了满足遥感图像实时检测的需求，提出了基于星载处理器遥感图像目标检测方案。该方法实现了对图像的在轨检测、地面解析等一系列功能。该方法基于应用场景的特殊需求，在满足检测精度的条件下，对检测时间进行了多方位的优化。
英文摘要	With the development of remote sensing satellite imaging technology, remote sensing satellite imaging resolution has been greatly improved. A large number of remote sensing satellites have been successfully launched and a huge amount of image data will be produced, which cannot be read and interpreted by human in a short time. How to carry on the object detection and recognition in remote sensing image is of great importance. At the same time, the deep learning related theory and technology has made a huge progress, and target detection and recognition algorithms got rapid development. Based on in-depth analysis on the present research, this dissertation studied the key techniques of aircraft detection. The main body of our research can be summarized into the following three pieces of work: 1. Research on the aircraft detection from remote sensing images For improving the detection precision and efficiency, a remote sensing image aircraft detection method based on the Single Shot MultiBox Detector (SSD) is proposed. To reduce the memory space, an improved version is also proposed. The improved version has a higher efficiency and fewer arguments than the basic one. 2. Research on the multi-categories aircraft detection from remote sensing images Based on the characteristics of remote sensing image multi-categories detection and recognition and the pros and cons of the existing algorithms, a multi category recognition algorithm based on the SSD is proposed. The method proposed has a higher mAP and efficiency than previous methods. 3. Research on on-board computer based objection detection n order to meet the requirements of remote sensing image real-time detection, a target detection system is proposed, which is based on the space borne remote sensing image processor. This method implemented the in-orbit target detection, data parsing and a series of functions. The method is based on the specific requirements of the application scenario, which has optimized the detection time on various aspects.
关键词	目标检测飞机舰船星载处理器卷积神经网络 Objection Detection Aircraft Ship On-board Computer Convolutional Neural Network
语种	中文
文献类型	学位论文
条目标识符	http://ir.ia.ac.cn/handle/173211/11516
专题	毕业生_硕士学位论文
作者单位	中科院自动化研究所
第一作者单位	中国科学院自动化研究所
推荐引用方式 GB/T 7714	巴君. 遥感图像飞机目标检测与识别[D]. 中国科学院自动化研究所. 中国科学院大学,2016.

条目包含的文件
文件名称/大小	文献类型	版本类型	开放类型	使用许可
毕业论文-终版-巴君-20160602.（3208KB）	学位论文		限制开放	CC BY-NC-SA