Automatic wane detection in the images of planks using a neural network

Optimal cutting of planks into components requires all defect types and their position to be known. This paper reports on the detection of the wane on images of through sawn planks. A fuzzy neural network operated as a window filter is used in order to determine the network structure and its initialisation from problem knowledge. The initial set of network weights is calculated from rules extracted from domain specific knowledge. Post-processing the output of the network results in a single pixel width line indicating the position of the wane edge. The system detects 97% of the wane and that wane edge detected has a transverse position standard deviation of less than 1.25 pixels (3.75 mm)