Objects Detection In Remote Sensing Images

Detecting objects in remote sensing images can be used in obtaining precise information of the war field in time, catching the strike targets, providing precise qualitative and quantitative information and so on. In the field of civilian use, detecting objects also plays an important role, such as resources exploration, environmental monitoring, city planning. We mainly discuss water extraction and bridge detection in remote sensing images, the paper consists of the following three parts:(1) A river extraction method based on seed point in remote sensing images is proposed. This method firstly determine the water conditions considering the initial seed point selected by manual, then divide the river into two parts and scan the two parts respectively. In the process of scanning, update the seed point and scanning mode if there is embranchment or bridge truncation. The method is especially suitable for large images ( the number of pixels is more than 106 ) , even though the area of river is large.(2) An approach for detecting bridges over water bodies from high resolution SAR images is presented. Firstly a new feature - porosity is proposed and a method based on porosity combined with an edge mending algorithm utilizing canny edges is used to extract water bodies. By considering the ubiety of bridges and water bodies, the regions of interest are detected. Then the false bridge regions are removed by line detection using Radon transform and the geometric characteristics of bridges. Finally bridges are located in confirmed bridge regions according to water extraction result. The proposed approach has been tested with SAR images that have a spatial resolution of 1 meter. The experimental results demonstrate our method can detect bridges effectively.(3) An approach based on treelet transform and knowledge for detecting bridges over water bodies is proposed. Firstly we establish bridge knowledge base and segment water by using treelet transform to extract features. After the extraction of water, we connect water bodies using mathematic morphological operation, possible bridge segments and bridge candidate regions are then obtained. Finally bridges are detected by feature matching. The proposed approach has been tested with panchromatic images that have a spatial resolution of 2.5 meters. The experimental results show the method is effective.