Research On TM Classification Based On K-means Of Wavelet Transform

With the development of remote sensing technology and computer technology, remote sensing data has been playing a more and more important role for analysis in the fields of geological disasters, military target recognition, agricultural production and geographic area statistics and other fields. In the study of remote sensing image data processing, remote sensing image classification has important research significance as an effective method to dig the image feature information and foundation of other advanced application.Remote sensing image classification is a technical process dividing all pixels in the image into several classes according to its nature. The vegetation characteristics of remote sensing image are directly and intelligently extracted and classified on computer. And different theories have different methods. On the basis of the necessity of foregone category and its training samples provided by classification personnel for computer classifier to be trained and supervised, the computer classification is divided into unsupervised classification and supervised classification. Supervised classification method is a process that the classifier is training by learning. It often need enough prior knowledge. While the unsupervised classification is a process of clustering without the need of manually training samples select.The traditional K-means clustering algorithm is one of the most widely used algorithms in remote sensing image classification. But it is still has the inevitable shortcomings: 1) low classification accuracy; 2) diversity of classification results caused by different initial values in random selection of the initial clustering center. For such shortcomings, this article introduced the traditional unsupervised classification based on the mean- variance K-means algorithm in detail and it also proposed the improve K-means algorithm based on wavelet transform. 7 classes of sample image are selected from the research area. Using wavelet theory to carry out two-dimensional wavelet transform processing on the TM remote sensing image to highlight or strengthen the different terrain characteristics. Energy feature vectors are used as the initial samples classification center for K-means classification. The problem of random initial value sensitivity is avoid, and at the same time supervised classification method thought is effectively integrated in unsupervised classification method to have the advantages of the two methods and improve the classification accuracy.Experimental results show that: 1) After single scale two-dimensional discrete wavelet transform to the unclassified image, the noise within the same object types caused by soil intensity is weakened, different terrain types of the edge difference is improved. Thus it is conducive for computer intelligent classification for remote sensing image; 2) Remote sensing images are classified by the two kinds of algorithms. From analysis and evaluation of different sides, the performance of the proposed algorithm is superior to the traditional K-means algorithm.