Solving the problem of identifying the annual rings of lumber in images of the ends using a neural network

Abstract

An urgent task for the timber processing industry is the analysis of annual rings in images of the ends of lumber. The successful use of neural networks in this task partly depends on the quality of training data preparation and on well-designed training quality metrics. The article discusses the methodology for preparing a data set, as well as the use of the DexiNed neural network for the task of recognizing tree rings. An approach is proposed for assessing the quality of training. The purpose of the study is to improve the quality of recognition of annual rings in images of the ends of lumber.