Study On High Resolution Image Change Detection And Its Application In The Emergency Disaster Evaluation

In recent years, with the continuous development of related techniques and the rapid growth of application requirements, remote sensing technique is increasingly applied in the field of disaster detection and assessment. Compared with low or moderate resolution images, high resolution remote sensing satellite images contain more information, so they has been more widely used in this field. When the traditional pixel-based image processing methods are applied to high resolution images, there are many problems, so object-oriented image processing technique is proposed and shows significant advantages. Thus, in this dissertation, multi-temporal remote sensing image change detection and its evaluation application in the emergency disaster detection and assessment are researched for high resolution remote sensing satellite images using object-oriented image processing technique.First, the characteristics of high resolution remote sensing images are analyzed and the interested area detection methods are studied. Generally, the populated areas are interested when detecting and assessing the damage caused by disaster using remote sensing images, so urban area detection method based on Gabor filter is modified and fast urban area detection method is obtained in this dissertation. In this method, the local feature points are extracted and the voting matrix is analyzed. Urban areas can be extracted and marked from high resolution remote sensing images which cover large area fast and effectively using this method, thus the following process time can be saved.Then, multi-scale segmentation methods for high resolution images are researched and statistic region merging method based on heterogeneity is proposed. A necessary condition for object-oriented image processing is to get accurate image objects. By multi-scale segmentation, image objects with different scales can be obtained and each land-cover type can be extracted and analyzed in an appropriate scale. Multi-scale segmentation method based on region growing makes use of the abundant shape and texture features of high resolution images, but it is slow because of multiple merging processing. Multi-scale segmentation method based on statistic region merging is faster, however, it only uses the color information rather than the shape information and texture information. Statistic region merging method based on heterogeneity combines the advantages of the two methods and by which accurate image objects can be acquired in appropriate scales quickly.After the multi-scale segmentation, object-oriented change detection method based on feature fusion is studied in this dissertation. Traditional change detection methods only take single pixels into account, while the object-oriented change detection method based on feature fusion uses the gray feature, shape feature and texture feature of images and can detect building changes more accurately than traditional methods. After that, the damage assessment of buildings is done using the correlation coefficient and the location and magnitude of the damage can be obtained.Finally, the algorithms above are integrated into a disaster detection and assessment system based on high-resolution images. As a disaster reduction application demonstration of multi-source remote sensing resources integrated application and management system, it has important application value.