Based On The Mean Shift Algorithm Of Remote Sensing Image Feature Extraction And Classification

With the rapid development of remote sensing technology, remote sensing image has been widely applied to all walks of life. Due to remote sensing image content contains a wide range, large data, concerned content in different ministries, so how the various elements of remote sensing image classification and accurate extraction in order to become the research hot spot.Scholars commonly used classification methods mainly include the maximum likelihood method, minimum distance method, support vector machine (SVM) method, etc, but these methods are not fully consider the like of the positional relationship between the pixel, to lead to "with different spectrum" and "Foreign bodies with spectrum" phenomenon appears, make the final classification accuracy is not high. However mean shift algorithm is a kind of based on kernel density estimation of nonparametric kernel density estimation algorithm, is not dependent on the probability density function of selection and parameter estimation.Mean shift algorithm has the advantages of computational complexity is small, easy to implement, in the image smooth, image segmentation and tracking of target has been widely used in such aspects. So this article according to the clustering characteristics of mean shift algorithm, two improvements of remote sensing image classification methods is given. The main works and achievements are as follows:(1) Based on the clustering characteristics of mean shift algorithm for remote sensing image as an example, discussed are the two parameters of the kernel function algorithm for image segmentation.(2) Is given in this paper, two kinds of improved method of remote sensing image classification is a kind of mean shift algorithm combined with support vector machine (SVM) classification method, and the other is a mean shift algorithm combined with minimum distance method. And compared with common classification methods from three aspects:the kappa coefficient, the confusion matrix and classification of time.The experimental results show that, in this paper, we give two methods of improvement are in effect and the time has significant advantages.(3) Are presented in this paper an improved remote sensing image element extraction method, the method by introducing contour, binarization and morphology method to improve. The effect of the method in extracting rivers and have good performance on operation time.