Parameter Optimization in Multi-scale Segmentation of High Resolution Remotely Sensed Image and Its Application in Object-oriented Classification

Abstract

With the widely use of high resolution remotely sensed image, the methodology of object-oriented classification emerged and became an active research area. Two basic steps of objectoriented classification method is segmentation and classification, and the effect of segmentation will directly affect the accuracy of classification. An optimal parameters method of multi-scale segmentation was described in this paper with the SPOT-5 high spatial resolution image of Pingtan Island in Fujian province, P. R. China. The cross validation technology was employed to compared the optimal value of the 4 parameters as scale, shape, color and compactness. The classification results of the traditional pixel-based supervised classification method and the object-oriented classification method were also checked using error matrix to qualify which method would be better. The result shows that object-oriented classification can make full use of spectral information, geometric structure and texture features of the image, and it has a better prospect in classification of the high spatial resolution remote sensing image. Keywords-segmentation; parameter optimization; objectoriented classification; high resolution remotely sensed image