Mauritia flexuosa palm trees airborne mapping with deep convolutional neural network.

Area of Science:

• Forestry Science
• Remote Sensing
• Computer Vision

Background:

• Accurate tree species mapping is vital for forest inventory, especially in dense forests.
• RGB imagery presents challenges for automated species identification due to spectral similarities.
• The Buriti palm (Mauritia flexuosa) is ecologically and culturally significant in South America and indicates water resources.

Purpose of the Study:

• To develop a deep learning approach for detecting and geolocating the Mauritia flexuosa palm tree in high-density forest environments using aerial RGB imagery.
• To address the limitations of spectral similarity in RGB scenes for automated tree mapping.

Main Methods:

• A Convolutional Neural Network (CNN) was employed to identify and geolocate individual Mauritia flexuosa trees.
• The method was tested in a complex forest environment using aerial RGB imagery.

Main Results:

• The developed CNN approach achieved a mean absolute error (MAE) of 0.75 trees.
• An F1-measure of 86.9% was obtained, demonstrating high accuracy in species detection.
• The method surpassed Faster R-CNN and RetinaNet under identical experimental conditions.

Conclusions:

• The deep learning method is effective for mapping single tree species in dense forest scenarios.
• Accurate geolocation of Mauritia flexuosa is achievable, offering benefits for ecological and resource management.
• This approach provides a foundation for future advanced forest mapping frameworks.