Road extraction algorithm based on region segmentation of remote sensing image
|School||Kunming University of Science and Technology|
|Course||Cartography and Geographic Information Engineering|
|Keywords||remote sensing image segmentation algorithm road extraction mathematical morphology|
Remote sensing image is a comprehensive reflection of the Earth’s surface features. Image analysis and information extraction is the way to transfer the remote sensing image data into information which can serve for the intermediate links and core technology of various industries. In recent years, the technology of remote sensing about earth observation developing rapidly, because of the remote sensing data which has high spatial resolution, high spectral resolution, and high phase resolution appearing, traditional information extraction technology is difficult to adapt to the requirements of current remote sensing data processing and application. Especially, the imperfection of the technology about automatic target recognition in high precision, high efficiency which has become a "bottleneck" to the application of the remote sensing data in large-scale, which also led to a large number of remote sensing data that can not be fully utilized. Therefore, in order to solve the problem, to develop intelligent information extraction technology about remote sensing image is urgent and effective, which also is one of the hot research field of remote sensing in current.The main purpose of the thesis is how to implement high-quality segmentation and automatically distinguish technology and extract road in high-resolution remote sensing image. In the course of the study, take the information extraction based on object-oriented as the main application target, the thesis analyses the principles and methods about image segmentation of remote sensing, and proposes a segmentation algorithm could suit for high-resolution remote sensing images, then carries out experiment on the application of remote sensing image processing, and extracts the road area in image effectively. The specific research work mainly reflects in:(1) Current main principles and methods of technology about remote sensing image information extraction are systematically summarized. A variety of surface features about thematic information extraction, including spectrum features, texture features, shape features, spatial relationship features, are introduced. Based on these characteristics, the technology about object-oriented remote sensing image information extraction is analyzed. Main segmentation methods and thematic information extraction methods about eCognition which is commercial software based on object-oriented image analysis are discussed in detail.(2) Multi-feature and multi-band regional segmentation algorithm (MM-RSA) is proposed. First, a texture image from single band is extracted and combined with original multi-spectral image to generate a combined multi-band image. Second, fuzzy c-means clustering algorithm is utilized to process the combined multi-band image into a single-band clustering image is used to generate the seed region. Third, an integrated criterion including the spectral feature and shape information is utilized to process the region growing and merger. According to different extracted objects, MM-RSA can segment the image into a variety of scales. MM-RSA is based on the segment model, so it has strong expansibility.(3) QuickBird image and World View image are utilized respectively for segmentation experiment, evaluating the results that generated by MM-RSA. Two evaluated methods are used for the QuickBird image objects:one based on object metrics evaluated method and the other based on classification evaluated method. Only one method that based on classification evaluated is used for the WorldView image objects. Experimental result proves that MM-RSA could generate high quality segmentation results and it is a feasible and effective segmentation method.(4) Mathematical morphology is adopted to extract road image. Start with the basic theory of mathematical morphology, the method about extracting road image is proposed, including morphological opening operation, morphological closing operation, and morphological reconstruction. A theory that uses the structural elements to calculate the numerical ratio is put forward to connect the disconnected road network. Then this method is utilized to optimize the rough road images which are extracted respectively from QuickBird image and WorldView image.