Aerial Identification of Amazonian Palms in High-Density Forest Using Deep Learning

Abstract

This paper presents an integrated aerial system for the identification of Amazonian Moriche palm (Mauritia flexuosa) in dense forests, by analyzing the UAV-captured RGB imagery using a Mask R-CNN deep learning approach. The model was trained with 478 labeled palms, using the transfer learning technique based on the well-known MS COCO framework©. Comprehensive in-field experiments were conducted in dense forests, yielding a precision identification of 98%. The proposed model is fully automatic and suitable for the identification and inventory of this species above 60 m, under complex climate and soil conditions.

Keywords

• Amazon Palm Moriche (<i>Mauritia flexuosa</i>)
• dense forests
• precision agriculture
• Mask R-CNN model
• deep learning
• convolutional neural networks