Automatic extraction of economically important forest tree species using deep learning and image processing techniques

Trees are crucial for the structure and functioning of forests and play a significant ecological and economic role.Certain tree species in forests have a key economic impact by providing timber and non-timber forest products (NTFPs) and contributing to the economic productivity and the livelihood of forest-dependent communities.Therefore, it is vital to accurately identify and quantify these economically important species to prioritise resource allocation and support rural livelihoods.While previous studies have focused on identifying tree species in more uniform temperate forest regions, few studies have concentrated on tropical, heterogeneous forests.Trees like Madhuca longifolia (Mahua), Dendrocalamus strictus (Bamboo), Tectona grandis (Teak), and Hardwickia binata (Anjan) are widely found in the Central regions of India.This study aimed to automatically identify economically important tree species using object-based image processing techniques, such as segmentation, object-based image classification, graphic analysis of spectral reflectance curve, and deep learning methods applied to Pliades Neo Satellite data.Trees possess specific crown characteristics, tone, texture, pattern and shadows, which produce distinctive spectral and spatial signatures.The study automatically extracted the targeted trees from India's highly diverse tropical forests, achieving an intersection over union (IoU) of 0.806.The Kappa coefficient turned out to be 0.81321, with omission error of 3% and commission error of 8%.