Aerial Identification of Fruit Maturity in Amazonian Palms via Plant-Canopy Modeling

Abstract

UAV-based multispectral imagery was used to characterize and associate the canopy features of the Moriche palm with the maturity state of its fruits, by correlating variations in the palm’s reflectance at different wavelengths throughout the phenological cycle. Several approaches for feature extraction were compared based on vegetation indices and graph-based models. A comprehensive dataset was collected and labeled, containing spatial–temporal variations in the features. Experimental results reported an accuracy of 72% in the estimation of the fruit maturity state, applying the proposed system to the dense forests of Colombia Amazonian region. Also, this UAV-based vision system enables monitoring, inventorying, palm identification, and fruit maturity identification, providing support to the local indigenous organizations of the Amazon.

Keywords

• UAV
• deep learning
• vegetation indices
• graph-based models
• mauritia flexuosa palm
• dense forests