Target-oriented Image Feature Extraction And Change Detection Method

Remote sensing technology is a comprehensive subject rising in recent years, feature extraction and change detection is hot issue in the field of remote sensing research. Feature extraction is the basics of the pattern recognition and machine learning, and also is an indispensable precondition of target identification. Change detection is the technique of dynamic research of the development and change of the ground target, which has been widely used in the civilian and military.There are lots of targets on remote sensing image; this paper selects the building as research object which is closely related to human production and life. This paper mainly complete the following contents:(1)Goal-oriented model is applied to remote sensing image field, goal-oriented model considers remote sensing image as a classification layer system, which proposed a goal-oriented image feature extraction method, this method extract texture features of the building segment the building from the image. Goal-oriented image feature extraction method firstly dusting remote sensing images using the improved K-mean clustering algorithm which comprehensive wavelet transform and K-mean clustering algorithm, which perform wavelet transform decomposition on unclassified remote sensing image, obtain three high-frequency and a low frequency, and then selected twelve typical ground target region from remote sensing image and perform wavelet transform decomposition, this region include building region, vegetation region and water region. Calculate the Euclid Distance of the four components of the remote sensing image and the four components of these small region and executing K-mean clustering, segment the building regions and extract texture feature of the building after obtain the classification results, segment the texture feature images using the difference of the texture features of building region and other regions, and obtain binary image of the building and make false-color processing.(2)Proposed a comprehensive pixel-level and feature-level change detection method. Pixel-level change detection and feature-level change detection belong to different levels of change detection method, the pixel-level change detection can detect change information of all ground target, but can not distinguish changes between the target, feature-level change detection can detect change information based on ground target feature, but must calculate more feature statistics, this paper will comprehensive the two method to achieve the purpose of combine theirs advantages. Firstly performing pixel-level change detection on two phases remote sensing images, obtain ground target change region, which including building changes, and then superimposed this change region on the original remote sensing image of two phases, extract texture features and tone feature of this region, and set an appropriate threshold range carved out of the building area, finally performing pixel-by-pixel comparison and obtain the building change information.The proposed method of this paper are verified the performance by compared with other methods with experiment.