Genetic Algorithms neural networks based wooden panel superficial defects CCD recognition system

Wooden panel superficial defects detection is a key step of the timber manufacturing procedure. This paper presents a wooden panel superficial defects Charge Coupled Device (CCD) recognition system based on Genetic Algorithms (GAs) Neural Networks. The system employs a nonlinear dual-window operator to extract the image segments according to the large transition region of defects image. Compared with other detection methods, the recognition system has the advantages of low cost, flexibility, high security etc. The gray-level mean, variance, circularity and length to width ratio of defects are selected as input parameters for the GA Neural Networks. And GAs is employed to solve the problem of the convergence rate and the local minimum in Back Propagation (BP) Neural Networks. The system realizes the recognition of both the natural and parasitical wooden panel defects effectively. The experiments demonstrate that the recognition accuracy rating of the main defects, such as loose knots, tight knots, decay and wormholes is 95%; the recognition time is less than 0.1s; the resolution of defects is better than 1mm(2).