Research On Target Recognition Methods For Building Detection In High Spatial Resolution Remote Sensing Images

Since the nineties of last century,high-resolution earth observation technology has entered a rapid development stage.Not only the spatial resolution has been greatly enhanced,but also the amount of the data has been increased rapidly.The high spatial resolution remote sensing images have been widely used in surveying and mapping,agriculture,forestry, urban planning,land resource management,geology and mineral resources survey,military and other fields.However,compared with the well-developed earth observation technology,the image processing and information extraction technology develops slowly.Especially in high spatial resolution remote sensing image processing,which can reflect the fine features in spatial structure and distribution of information,and has great potential application value in the field of urban planning,no accurate and effective identification method has been developed yet.The visual interpretation,which is time-consuming,laborious,and hard to guarantee the accuracy,is still the most widely used interpretation method.Thus,the large-scale application of the high spatial resolution remote sensing images does not come true,and a lot of image data is wasted.Based on this,this article summarizes domestic and international target recognition methods which are commonly used in remote sensing images,and target recognition methods for buildings in high spatial resolution remote sensing images,puts forward that the object-oriented object recognition method for remote sensing images has a habit of identifying that is closest to the human interpretation,can lead to a strong application prospects.However,the he object-oriented object recognition method is not perfect yet,it still remands high artificial participation,and the extraction accuracy is not ideal.Against these problems,this article uses SINCE2008 as the experimental platform,conducts a series of experiments with QuickBird data,proposes the coupling GA and SVM target recognition method and verifies its accuracy.The main works of this article are as follows:(1) In SINCE2008 platform,apply the pretreatment to QuickBird data of an area of Wuhan,using a variety of image fusion and image enhancement methods,find the best method by the segmentation results.(2) Select two sets of experimental data,using mean-shift segmentation method,find the optimal segmentation scale and the segmentation parameters.(3) Summarize the spectrum,texture,shape topology and other characteristics buildings have shown in high-resolution remote sensing images,develop the categories program for building recognition.(4) Sum up the feature optimization and classification methods commonly used in building recognition,improve the traditional genetic algorithm,put forward the coupled GA and SVM target recognition method, make the experiments and compare the results with the original method.(5) After the recognition,design a Primitive merging algorithm to merge all the Building Primitives.