With the rapid development of modern remote sensing technology, a great amount of remote sensing images have been available for earth observation, covering different types of sensors, different spatial resolutions, different spectral resolutions and different temporal resolutions. In recent years, object detection and navigation based on remote sensing images have drawn increasing attention from military and academia, and received a great deal of progress. However, because of the variety of applications, many problems still remain open. Our work mainly focuses on these two respects: object detection and image-aided navigation. As for object detection, we investigate two typical objects: oil tanks and urban roads; and as for navigation, we focus on how to improve the environmental applicability of current image-aided navigation systems. The main contributions of this thesis include: 1 For oil tank detection, we propose an approach by fusing Synthetic Aperture Radar(SAR)and optical images. The presented algorithm takes advantage of the corner cube reflector property of oil tanks, which always makes them bright in SAR images. The distinct geometrical shape of oil tanks in optical satellite images is also an important clue for oil tank detection. Furthermore, prior knowledge, such as the valid range of their physical dimensions, homogenous and group emerging characteristics, is integrated into the automatic detection process. Experimental results demonstrate that the proposed algorithm has an encouraging detection performance. 2 A new framework based on machine learning is proposed for extracting urban roads automatically from high resolution satellite images. Based on urban road properties in high resolution optical images, many features covering intensity, geometrical and textural properties are extracted for learning process, through which not only a classifier is obtained but also effective features for urban road recognition are selected. These selected features are helpful for guiding urban road extraction process and for analyzing the urban road physical properties. Experimental results demonstrate the effectiveness of the proposed method. 3 From the point view of environmental applicability of image-aided navigation, we propose a new approach under the particle filter framework, which can reduce the fastidious demands for matching area selection. Experiments show that it can achieve fairly good results.
修改评论