Automatic detection of annual rings and pith location along Norway spruce timber boards using conditional adversarial networks

Abstract In the woodworking industry, detection of annual rings and location pith in relation to timber board cross sections, how these properties vary longitudinal direction boards, is relevant for many purposes such as assessment shape stability prediction mechanical timber. The current work aims at developing a fast, accurate operationally simple deep learning-based algorithm automatic surface growth along knot-free clear wood sections Norway spruce boards. First, individual that are visible four sides scanned boards detected using trained conditional generative adversarial networks (cGANs). Then, locations determined, on basis rings, by multilayer perceptron (MLP) artificial neural network. proposed was solely based raw images surfaces obtained from optical scanning applied total 104 with nominal dimensions $$45\times 145\times 4500\,\hbox {mm}^{3}$$ 45 × 145 4500 mm 3 . results show scanners method allow fast For located within section, median errors 1.4 mm 2.9 mm, x- y-direction, respectively, were obtained. sample outside section most positions board, discrepancy between automatically estimated manually determined 3.9 5.4 respectively.