An Object-based Forest Type Classification Research Based On GF-1 Remote Sensing Data

With the rapid development of remote sensing technology, it has become important and difficult to use remote sensing image to extract information in the field of remote sensing. Because of ignoring the richness of high resolution remote sensing image space and texture feature information, traditional remote sensing image information extraction technology can not get satisfactory result of extraction. It is a new direction for classification of remote sensing technology that object-oriented classification method is proposed.This study used a new remote sensing image GF-1 remote sensing image data for the study of base data, and according to the characteristics of GF-1 data for atmospheric correction, geometric correction, and image fusion for image preprocessing. Adopts the object-oriented classification method to extract forest information in Walagan forest farm of Tahe, give full consideration to the spectrum and texture, geometry and to join the information such as vegetation index. Use the multi-scale segmentation technology, and use the decision tree classification method and the adjacent classification method for classification process, finally it realizes the classification of forest type. The research emphasis that it is multi-scale segmentation parameter acquisition, feature information filtering and object-oriented classification method. The following conclusions:(1)According to the principle and algorithm of multi-level multi-scale, use different scale to segment different terrain objects. Through many experiments, it determines segmentation scale, shape factor, smoothness, and firmness, and combined with visual interpretation, choosing the optimal segmentation scale.(2) Using different spectral features, shape, texture characteristics, and application of decision tree model for remote sensing image feature extraction, and it completes the study area of the classification of forest type, at the same time, compared to a single level of the adjacent classification method, the practical conclusion proves that application of multi-level segmentation model of decision tree classification accuracy is 84.1%, the Kappa coefficient is 0.815, and the single level of the adjacent classification method classification accuracy reached 76.2%, the Kappa coefficient of 0.703. The former classification accuracy is higher than the latter, thus it proves the classification of the multi-level multi-scale method is more effective for forest types classification and recognition for high resolution remote sensing image classification processing, it also provides reliable evidence for the extraction of forest type.