Study On Object-Oriented Remote Sensing Image Classification System

With the development and successful launch of China’s environmental satellite in China, study and application of Remote Sensing Technology have made remarkable progress, high resolution remote sensing image has played an important role on the relevant application fields. The traditional classification method which based on a single pixel cannot satisfy the demand of high resolution remote sensing image classification. Therefore, The object-oriented classification is essential.The method of object-oriented remote sensing image classification is not analyzes pixels singly,it takes a set of homogeneous pixels as unit to classify map features. This method eliminates the "salt and pepper effect" caused by the individual pixels. object-oriented classification is based on image segmentation to get homogeneous image object, then analysis spectrum, shape, texture features according to the classification target, to classify and extract the feature information. Object-oriented remote sensing image classification is mainly include two key steps:multi-scale segmentation and feature selection. Multi-scale segmentation is based on the analysis of the shape characteristics and the spectrum characteristics, make the homogeneous pixels as objects of different size, and the internal consistency and heterogeneity with the neighbor object are both maximum. Feature selection is to select the spectral features, shape and texture features to classify according to the classification target, and high-precision information extraction results are obtained.This article designed and developed an object-oriented remote sensing image classification system based on Microsoft Visual Studio2008 software platform and ArcGIS Engine component library. The system include two key technologies:multi-scale segmentation algorithm and image classification.First, according to the principle of multi-scale segmentation algorithm, Designed and written the code to realize the remote sensing image segmentation, and better segmentation results were obtained. Then on the basis of image segmentation, studied the calculation method of image objects’spectral, shape and texture characteristics, realized object-oriented classification by choosing suitable characteristic of image object. This paper extracted the nature reserve human interference information of Nuluer Hushan National Nature Reserve with high-resolution remote sensing image as data sources, using the designed object-oriented remote sensing image classification system, then checked by the total accuracy of classification result reached to 89.29% and Kappa coefficient was 0.86,so the accuracy meet application requirements. And the system was tested in Nabanhe River Watershed Nature Reserve, Xishuangbanna as demonstration zone, then the system was checked by the total accuracy of classification result reached to 88.58% and Kappa coefficient was 0.77. Results show that the system can effectively extract the information in the nature reserve, and can reduces the time and save the manpower, realized high resolution remote sensing monitoring of the human interference information in nature reserve high precision, fast high resolution remote sensing monitoring.