The Method And Implementation Of Semi-Automatic Extraction From High Resolution Remote Sensing Image Based On Object Oriented And Ensemble Learning

High resolution remote sensing image could provide much detailed information of the surface of the ground. It becomes a hotpot of remote sensing research to use the image analysis methods to fulfill object extraction and update the geography information database. Traditional pixel-level approaches merely utilize the spectral characteristic,but the high resolution remote sensing image such as QuickBird and IKONOS have a lot of characteristics such as spectral, shape, texture and context and so on , compared to the other remote sensing data. It will result in not only reducing the extraction accuracy but also making the spatial data redundant and wasting the resource when the single traditional extraction method is applied to the high resolution remote sensing image.So, with the QuickBird data as the main source, this thesis analyses and verifies the great application potential of the object-oriented and ensemble learning technology in high resolution remote sensing image information extraction. The new algorithm of feature selection for ensemble is brought forward based on the research before and applied to the road extraction from high resolution QuickBird image. This dissertation includes the following research products and originalities:1) Study two key technology including image segmentation and object-oriented classification of object-oriented information extraction technology of high resolution remote sensing image. Apply the support vector machine and neural network to the high resolution remote sensing image classification based on the object. Through the experiment, verify that the object-oriented information extraction technology is superior to traditional pixel-based method on high resolution remote sensing image classification.2) Introduce the basic concepts and principles of ensemble learning and several main ensemble learning methods. Verify the validity of ensemble learning through heterogeneous and homogeneous classifier ensembles.3) Describe the spectral, shape, texture features of objects quantificationally and introduce the genetic algorithm used to solve the combination and optimization problems. The new algorithm of feature selection for ensemble based on the genetic algorithm is brought forward at the same time and the model of road has been extracted through applying the algorithm.4) Introduce the semi-automatic extraction framework from high resolution remote sensing image based on object-oriented ensemble learning and the genetic ensemble feature selection algorithm, then use this framework to experiment on road extraction, verify the feasibility of applying this method for extracting.