Olive Tree Biovolume from UAV Multi-Resolution Image Segmentation with Mask R-CNN

Area of Science:

• Agricultural Science
• Remote Sensing
• Computer Vision

Background:

• Olive cultivation is a significant global economic activity, particularly in rainfed, scattered groves.
• Accurate biovolume measurement is essential for monitoring olive tree performance, health, and yield.
• Traditional methods for biovolume estimation are often labor-intensive and less precise.

Purpose of the Study:

• To develop and evaluate a deep learning-based method for estimating the biovolume of individual olive trees.
• To assess the effectiveness of Unmanned Aerial Vehicle (UAV) imagery and various spectral indices for this task.
• To improve the accuracy and efficiency of biovolume measurement in scattered olive groves.

Main Methods:

• Utilized the Mask R-CNN deep learning instance segmentation model for olive tree crown and shadow segmentation (OTCS).
• Employed UAV imagery with different spectral bands (RGB, Near Infrared) and vegetation indices (NDVI, GNDVI).
• Evaluated model performance at various spatial resolutions (3 cm/pixel and 13 cm/pixel).

Main Results:

• Mask R-CNN models achieved high performance in tree crown segmentation, with F1-measures ranging from 95% to 98% using fused NDVI and GNDVI data.
• Estimated biovolume showed an average accuracy of 82% when compared to ground truth measurements.
• NDVI and GNDVI spectral indices demonstrated effectiveness for biovolume estimation in scattered trees.

Conclusions:

• The proposed deep learning approach using UAV imagery and spectral indices (NDVI, GNDVI) enables accurate biovolume estimation of scattered olive trees.
• This method offers a valuable tool for precision agriculture, enhancing monitoring of olive production and tree health.
• The findings support the integration of advanced remote sensing techniques in agricultural management practices.