Research On Region Target Semi-automatic Extraction Of RS Images Based On Texture Features

With the development of technologies of sensor, aerial and spatial platform, and data communication, remote sensing technique has entered into a new era that can obtain various earth data actively, rapidly, accurately, and through multi-way. It possesses omni-directional and all-weather ability to acquire space information. Therefore, it is an important data source to update spatial information, and have extensive application potential in national economy and military affairs. As far as the interpretation of spatial images and the digitization work in surveying and mapping field are concerned, this paper has researched a method, based on texture features of remote sensing images, to semi-automatically extract canonical area symbol. The key contents of the method are texture feature extraction, feature matching, and region extraction etc.The major works of this paper are listed as follows:1. The concept, significance, difficulty of remote sensing image interpretation is introduced. Current status and developing trend of area symbol extraction from remote sensing images is analyzed. Based on texture features, relating technologies of area symbol extraction are discussed.2. Vegetation extraction methods, which are based on Radon transformation and geometrical cente moment, are extensively analyzed and improved. Then, new algorithm, which combines Log-polar transformation and Zernike moment, is proposed. Experimental results prove that the algorithm can ideally extract settlement places from panchromatic remote sensing images of different scales.3. Using scale-based concurrent matrix and Laws texture energy, vegetation texture information is deeply mined on various frequency bands. The dynamic information between different scales and the static information of the same scale is combined and verified to sufficiently express vegetation texture information.4. Based on Gabor wavelet transformation, texture feature extraction techniques are analyzed and investigated. According to the characteristics of settlement places, the techniques are improved to extract settlement places precisely and semi-automatically, and therefore, possess great practicality.