Research On Image Classification Application Of Landsat8 Based On Support Vector Machine

With the rapid development of remote sensing technology,the enhancement of the resolution of satellite remote sensing image has provided us with the richer information about the surface.The appearance of high-resolution remote sensing satellite makes us acquire geometric information and texture information except ground object spectral information better especially in recent years.The feature classification of remote sensing image is the most essential part of remote sensing image processing.It plays an important role in many fields such as land utilization,environmental monitoring and geological survey.As a result,the extraction of ground information from remote sensing image has always been an issue which is worth doing long-term research.How to use better classification technology to classify ground objects for remote sensing images has a direct impact on the application of remote sensing technology.So many scholars in related fields have kept on trying to polish up the existing sorting technology and exploring new methods in order to improve the accuracy and speed of remote sensing image classification.This dissertation has chosen Landsat8 images of Zhoushan archipelago and some areas in Ningbo as the experimental data of classification and has used many remote sensing images with rich ground features as training samples and classifies remote-sensing images based on SVM in machine learning theory.It extracts colour features and textural features from remote sensing images with rich feature types and normalizes the two features into a general feature of feature samples,then uses various feature types and each feature's different sample data to train the multi-classification SVM classifier.It also makes use of the sliding window to select the images of child window and takes advantage of trained classifiers to distinguish which feature type this window image belongs to,and finally,merges adjacent windows with same surfacefeatures in the whole image and outputs the classification result of the whole image.By comparing results from the experiment with the one through ENVI Software to classify the remote sensing images via the method of maximum likelihood,it can be found that though SVM algorithm costs more time than ENVI in feature extraction and sample training,but the accuracy can be better than maximum likelihood classification.And the trained SVM classifier which is more practical can classify any remote sensing images.