The Study Of High Resolution Remote Sensing Image Classification Based On Extreme Learning Machine

Remote sensing technology has the characteristics of macro, comprehensive, efficient, dynamic, large amount of information, so that it can be widely used in various fields. With the development and operation model of remote sensing data, a large number of remote sensing images have been obtained, including more abundant information. Characteristics of complex details, scale of high spatial resolution remote sensing image and massive data dependent lead to that it is difficult to deal with high resolution remote sensing images. The high resolution remote sensing image, such as QuikBird and WorldView displayed more object information, including spectral, shape, color, texture and etc. The classification of remote sensing image is a basic method for extracting information from them.The traditional classification methods for remote sensing image based on pixel only relate to spectral features for image processing. However, the limited information can be obtained only considering the spectral characteristics of remote sensing image. Therefore, classifying remote sensing data by the traditional methods will waste a lot of information included in high spatial resolution remote sensing image. It is a difficult and challenging problem how to realize classification of remote sensing image classification method in the case of employing information contained within them.One of the main tasks of spatial data mining is the extraction of information from remote sensing image, and to make an interpretation for remote sensing image based on knowledge driven. In order to solve this problem, many machine learning algorithms have been introduced to the remote sensing image processing. In this dissertation, the algorithm for high resolution remote sensing image classification based on extreme learning machine is improved. The HYDICE aerial image is chosen to test the performance of the proposal. These aerial images are the standard high resolution image data used to test the effectiveness of classification algorithms. The test results show that the classification algorithm based on extreme learning machine has made improvements both on precision and speed. Then we use the ELM algorithm to extract the greenbelt information from WorldView 2 images. The greenbelt information obtained by the proposed method is very accurate and effective. It will be useful to establish a greenbelt classification system or provides a good data source to other auxiliary decision system. The research proves that the algorithm based on extreme learning machine for high resolution remote sensing image classification has good overall performance and practical value.