Deep learning approaches for accurate wood species recognition

Abstract

Wood species identification is a crucial task in various industries, including forestry, woodworking, and conservation. Traditional methods rely on manual expertise, which can be time-consuming and error prone. Hence, an automatic wood species recognition system is developed in this study using deep learning (DL) models. In this study, three deep convolutional neural network (CNN) architectures, SqueezeNet, GoogLeNet, and ResNet-50 was tailored for wood species classification. The accuracy of the DL models was evaluated in recognizing fifty different wood species. Additionally, the wood species images were altered using JPEG Compression, Gaussian Blur, Salt and Pepper, and Speckle noises to assess the models' performance in identifying the wood species from the distorted images. Results show that the ResNET-50 based wood recognition system is the most accurate model to recognise the wood species. The implications of this research extend to forestry management, quality control in woodworking industries, and the preservation of endangered wood species in conservation efforts.