EXPERIMENTS WITH SOM BASED INSPECTION OF WOOD

摘要: We have devised a non-supervised clustering based approach for detecting and recognizing defects in lumber boards. The solution is simple to train, supports knots other by using multidimensional feature vectors containing texture color cues from small non-overlapping regions the image. key idea employ Self-Organizing Map (SOM) discriminating between sound wood defects. An almost identical scheme employed classifying Human involvement needed training minimal. still under development, although it approaching application level maturity. In this paper, we investigate dependence size of SOM false alarm, error escape, correct classification rates. Based on tests demanding real-world material, rather (12*8 nodes) SOMs provide attractive performance, approximately 31% alarm 5% escape rate defect detection. classification, accuracy better than 72 % with material used. All results are respect human region labelings, can be considered excellent. new inspection, tested similar food steel inspection.