Tree Species Classification from Remotely Sensed Data: From Hyperspectral to RGB скачать в хорошем качестве

Tree Species Classification from Remotely Sensed Data: From Hyperspectral to RGB

Tree Species Classification from Remotely Sensed Data: From Hyperspectral to RGB 11 April 2025 by Dr Matheus Pinheiro Ferreira Universidade de São Paulo (USP), Department of Forest Sciences at the Luiz de Queiroz College of Agriculture, Brazil Tree species classification is crucial for managing and conserving highly diverse tropical forests. However, spatially explicit information on tree species distribution is typically limited to small-scale field inventories, making large-scale monitoring challenging. Remote sensing (RS) offers a powerful alternative, particularly very-high-resolution imagery (GSD ≤ 1 m), which enables species classification at the level of individual tree crown (ITC) level. Hyperspectral imaging have demonstrated the potential of fine spectral resolution for distinguishing tree species based on subtle biochemical and structural differences. More recently, deep learning, especially convolutional neural networks (CNNs), has emerged as a promising approach to automate tree species classification using ultra-high-resolution drone-acquired RGB images. Additionally, unsupervised deep learning methods are being explored to augment ITC datasets, addressing the challenge of scarce labeled data in tropical forests. In this talk, I will explore how different RS data sources—from hyperspectral to high-resolution RGB—can be leveraged to improve tree species classification in tropical environments. I will discuss recent advances in CNN-based classification and the role of unsupervised learning in enhancing ITC datasets. Webinar sponsored by IEEE GRSS UK & Ireland Chapter