Application of Edge Detection on Large Lump Detection

Detecting large lumps of oil sand is a critical step which protects exploitation facilities from jamming. This report presents a method that detects large lumps with edge detection and edge classification. The approach applies Gaussian based local scale control algorithm to detect edges in oil sands, then extracts intensity features as well as geometry features for edge classification. Trained with manually selected ground truth, the k nearest neighbour classifier will distinguish edges of large lumps out of edges of fine fragments. Finally, large lumps of oil sands can be located with classified edges. Since large lumps are often merged in a pile of fine fragments, it is quite difficult to get the complete shape of this kind of oil sands. The main contribution of this approach is detecting large lumps only with edge information, which is much easier to get.