An Approach for Linear Feature Detection from Remote Sensing Images withHigh Spatial Resolution Based on Sparse Decomposition

Objective Linear feature detection from remote sensing data is an importance step in automatic targetrecognition.A high spatial resolution remote sensing image is highly detailed,which causes problemsfor some linear feature detection methods with edge fractures and fuzzy,salt and pepper noise.In thispaper,a novel linear feature detection approach for high spatial resolution remote sensing imagesbased on sparse decomposition is proposed.First,an over-complete dictionary for linear feature detec-tion was designed with a K-SVD algorithm.Using the sparse decomposition model,high frequencycomponents were extracted from high spatial resolution remote sensing images,realizing initial detec-tion of linear features from high spatial resolution remote sensing images.Then,denoising with aCurvelet and hierarchical adaptive threshold was applied to the high frequency component,which im-proved the linear feature detection effect..Finally,a simulation based on QuickBird data was execu-ted.Experimental results verified that the proposed method is superior to the Canny and Sobel meth-ods for line continuity,low contrast line detection,and salt and pepper noise elimination.