Research On Landscope Objects Extraction In Remote Sensing Image Based On Conjunction Of Spectrum And Spatial Feature

High spectral and spatial resolution remote sensing datas which enrich more information forinterpretation and application of spatial images, provide more and more specific knowledge oflandscope objects such as tone, texture, shape, structure and so on for image interpretation andanalysis. However, the results of landscope elements extraction turn worse, and efficency lower,with traditional image intelligentized interpretation technologies, as local information in imagebecomes more complicated and heterogeneity in object region is enhanced. Summarizing thetechnologies of object extraction and feature expression, this paper proposes an appreciablystable object recognition model for median or high spatial resolution, multi-souce spatial imagesbased on spectral-spatial conjunct feature with different levels of feature knowledge, andtherefore accomplishes the extraction of landscope elements integrating the spectrum with spatialinformation. The major works implemented are listed as follow:1. The concept,development,difficulty of remote sensing image intelligentized interpretationwere introduced. Current technologies of landscope elements extraction from RS images withspectrum or spatial information were analyzed. The significance of landscope object extractionintegrating the spectrum with spatial information was then outlined.2. A delineation of spectral-spatial conjunct feature and its otaining methods were proposedbased on various methods of obtaining different features in different kinds of RS datas, as themanifestation and delineation of spatial and spectral feature attributes of diverse landscopeelements were lucubrated. The experiments showed that conjunct feature turned a betteradaptability and more robust.3. A novel feature-weighed object discriminant method was concluded while the objectdiscriminant function and rule was lucubrated in feature space. Then the classic seed regiongrowing algorithm was modified accordingly as the use of three adjustive parameters optimizedconjunct feature clustering and image segmentation process. This method resulted a betterextraction effection to diverse object regions and greater feature difference.4. The kernel fisher discriminant analysis was introduced into the landscope objectextraction based on conjunct feature as far as kernel fisher object discriminant and featureclustering methods were concerned. The experiments indicated an optimized effection oflandscope elements extraction based on conjunct feature.