A Novel Remotely Sensed Image Interpretation Method MS-SVMS

Support vector machines (SVMs) is a statistical learning method with good performance when the sample size is small, due to their excellent performance, SVMs are now used extensively in pattern classification applications and regression estimation, Unfortunately, it is currently considerably slower in test phase caused by number of the support vectors, which has been a serious limitation for some application such as remotely sensed data classification. To overcome this problem, we introduced mean shift (MS) algorithm to select the feature vectors. Through the MS algorithm, the modes of data are real input vectors and the number of modes is controlled by three physical meaning parameters. Remotely sensed data has spatial and spectral characters and it has several million pixels in one image generally. Therefore, how to reduce the complexity of the data becomes a crucial problem in remotely sensed data classification based on SVM method. In order to solve such problem, we proposed MS-SVMs classification method. MSSVMs is the combined process of segmenting an image into regions of pixels based on mean shift algorithm, computing attributes for each region to create objects, and classifying the objects based on attributes, to extract features with SVMs supervised classification. This workflow is designed to be helpful and intuitive, while allowing you to customize it to specific applications. In order to verify the feasibility and effectiveness of proposed method, Landsat ETM image is adopted as original data, and experiments proved the proposed method is robust and efficient, further more, it helps improve classification speed and accuracy observably.