A comparison on texture classification algorithms for remote sensing data

Ground penetrating radar (GPR) systems are widely used to inspect underground structures. We have collected GPR images to detect and identify structures beneath rail tracks. Several approaches, including a statistical edge detector and machine learning methods, have been applied to analyze the images. Our results show that the edge detector efficiently enhance the layer boundaries that separate ballast, subballast, and clay. The machine learning approaches are promising means to detect these layers automatically.