The Typical Ground Feature Extraction Using High-resolution Remotely Sensed Imagery From Google Earth

With the development of earth observation technologies, the spatial resolution of remote sensing images has been improved a lot. Because the continuous improvement of the resolution, object detection has become one of the most hot technologies in remote sensing applications area. Object detection has been widely used in the field of military, agriculture, space science, water management, etc. Because of the improvement of the spatial resolution of the images, the objects are also more elaborate in images. On the one hand, obeject detection are closely related with the spatial resolution. There are usually mixed pixels in low-resolution images. So traditional classification methods based on pixel are unable to extract the small and distributing discrete obejects, which need to use high-resolution images to detect and extract. On the other hand, there are some weakness of high-resolution images which include high prices, large amounts of data, slow processing efficiency and requiring a lot of human control in the classification process. So there is little related work for extracting the objects in large area. In this context, this paper presents an automated data acquisition and processing method, which combined with object-based features to extract this kind of typical obejects.In this paper, the major research contributions of extracting typical objects are as follows:1. The paper achived a technology based on Google Earth that aqcuire remote sensing data automatically from Google Earth. And it aquired the high resolution data of the lower-stream of Heihe River(area is 28859.7km2) which is 0.6m resolution. Then, it proposed a method to compute the geographic coordinates and projection parameters of images for the 3454 datas of researching area. Finally, 34 regions were mosaicking through these informations.2. The paper adopted the extraction method that conbined decision tree algorithm and object-based classification algorithm. It analysis all of the object-based feature of the objects in the image on the basis of decision tree,and maximize features among each objects. In the end, it generates a extration ruleset of the typical objects and traverse that to extract the objects.3. Based on the researching data that proccessed automatically before, the paper implemented the populus extration in lower-stream of Heihe River. The extration area is about 30.9km2. Also, the result was evaluated through confusion matrix. The overall accuracy achieve more than 80%, even reach up to 87% in the some area. Both of the rate of leakage points and error rate are lower than 19%.The research method of this paper has a high practical value, it can save millions of dollars for the data research. Also it is easy to be extended to other remote sensing applications.In addition to typical object extraction, such as vegetation fine cover classification.The implementation of the method provides a feasible technical solution to detect and extract national and global thematic information with automation,low-cost,high-precision.And it will lead a group of related applications in the future, which is able to provide high-precision, high-resolution essential data at a very low cost for state economy and social development.