Extraction Of Thematic Information Based On Svm Remote Sensing Data

As the development of aeronautics and computer science and their technologies, extracting special information from RS (Remote Sense) data has already become a very popular research area in information times of today. But extracting weak information such as mineralizing information from RS data is still a new and important subject in geography. After briefly introducing research contents, approaches and tendency of extracting special information from RS, this paper analyzes principles of extracting mineralizing information, and discusses, appliedly, SVM's (Support Vector Machine) algorithm principles, classification, automatic selection of modal, optimization of algorithm's actualization, problems existed in the algorithm and their solutions etc. and proposes a new approach of extracting weak information from TM RS pictures and proves its accuracy and practicability through the experiment.In the experiment, first, with RGB values as input feature vectors of three-dimensions, the picture is recognized by SVM and the accuracy of SVM in recognition together with the efficiency of algorithm of LOO (Leave-One-Out) automatic modal selection are tested. Then TM RS data of Lianglan district (Dulan county and Wulan county) in Qinhai Province is tested. Its training set, with sand area and mountain of 32x32 pixel separately as its subsets, and its testing set, the district about 236 square kilometers of 512x512 pixel, are classified and recognized satisfactorily; whereas if its training set, with mineralizing of copper as its subset, the same district cannot be classified better perhaps because of the purity problem of the training set.Lastly, this paper concludes the problem in extracting weak information from RS data by application of SVM and suggests some improvements of this algorithm.The major research achievements of this paper:1. For the first time, proposing a new approach of extracting mineralizing information from TM RS pictures by application of SVM algorithm and getting some satisfactory effects.2. With R.G.B color values as input feature vectors, testing the accuracy of recognizing picture by SVM algorithm and the efficiency of algorithm of LOO automatic selection, and this providing the basis for the accuracy of recognition of RS pictures.3. Through SVM algorithm, solving the building problem of input sample feature vector (weak information sample) in the process of extracting mineralizing information from RS data.4. By the way of extracting mineralizing information abovementioned, testing the recognition of RS pictures of the specific and small area, and qualitatively analyzing and proving its practicability, and finding some effect of this approach. 5. In order to get high accuracy of recognition, proposing furtherimprovements of this approach.