Automatic Wood Species Classification Using Network's Architecture Model Based on Convolutional Neural Network.

We propose a variant of the convolutional neural network for the classification of tree species by extracting features from microscopic images containing wood vessels. State-of-the-art methods and algorithms were deployed. As a result, an algorithm for tree species classification is implemented. The obtained results using images of tangential and radial sections confirm the results of the scientific studies, namely that the transverse section contains the most characteristic of individual tree species information. The highest accuracy of 99.24% was achieved using cross-section images with dimensions $100\times 100$ pixels. The accuracy results for the tangential section and the radial section images, with the same resolution, are 89.20% and 87.35% respectively.