Enhancing bark texture analysis and species classification with deep learning models: Wide residual networks and convNeXt

Abstract

This study delves into the analysis of bark texture images using deep learning methods to efficiently classify different wood species. With applications spanning from construction to furniture manufacturing, efficient and precise wood species classification is vital for effective forestry management and the timber trade. The research centers on a dataset featuring images of 50 distinct wood species, each characterized by unique texture patterns. Two deep learning models, Wide Residual Networks (WRN) and ConvNeXt, are employed and compared for their analysis purposes. Results consistently demonstrate WRN's superior performance, attributed to its architectural design and effective training strategy in capturing intricate texture patterns. Notably, WRN achieves impressive efficiency alongside high accuracy, precision, and recall rates of 97.23%, 97.29%, and 97.23%, respectively. WRN's success over the pre-processed dataset underscores its versatility and robustness in handling complex texture patterns. Overall, the study showcases the transformative potential of deep learning in revolutionizing tree species classification.