Compact computer vision system for tropical wood species recognition based on pores and concentric curve

In Indonesia, there are more than 400 different species of wood commodity which commonly verified by trained evaluator, who is well trained and experienced to differentiate any different species of wood. The verifying process is carried out by sensing some general wood characteristics e.g. color, texture, fiber, conjecture, odor, rigidity and wood anatomical features. This study develops a portable vision system, which could recognize different wood species based on its pores and concentric curve, in order to replace the role of human evaluator. Firstly, ten different species of Indonesian-wood samples are provided as the objects. Secondly, the wood anatomical feature images captured by the CCD microscope with 50× magnification followed by the pre-processing method in order to obtain the physical characteristics of the wood. Next, wood's feature extractions are obtained based on multichannel Gabor filter on the wood pores and concentric curves. As final step, an artificial neural network with back propagation method of multilayer perceptron (MLP) was used to classify wood species. The training and validation process is carried out by using 20 testing data for each wood species. The total recognition of 10 wood species on multiple π/4 orientations is 95%, on multiple π/6 orientations is 95.5% and multiple π/8 orientations is 96.5%.