UAV-based precision agriculture solution for orchard tree detection and biomass estimation

Overview

This solution combines unmanned aerial vehicle (UAV) imagery with advanced computer vision and deep learning to help smallholder orchard farmers monitor tree growth, estimate biomass, and make timely in-field decisions. It addresses the challenges of irregular planting patterns and hillside terrain common in small-scale plantations, enabling affordable precision agriculture for crops such as avocado and citrus.

Technical specifications

Key features:

• Morphological-based transform that enhances tree localization in irregular planting patterns and hillside terrain
• Local maxima summation method for seed selection that integrates information from multiple heights, improving individual tree detection accuracy
• Convolutional neural networks designed for image phenotyping tasks, optimized for deployment on low-cost computing devices
• Digital surface model analysis capable of handling overlapping canopies, varying canopy heights, and low-resolution imagery
• Planned U-net deep learning architecture for fast and accurate image segmentation
• Transfer learning approaches to compensate for limited training data
• Python-based implementation using Keras and TensorFlow APIs for replicability

Technology readiness level

The methodology has been evaluated on digital surface models from three diverse farms covering avocado trees with regular spacing, citrus trees with overlapping canopies at different stand ages, and regions with varying canopy heights on irregular and hillside terrain. Future validation will focus on data collection and preprocessing in the first six months, followed by deep learning model development over the remaining period. The team is actively seeking partnership support and funding to advance the research toward deployment-ready solutions for smallholder farmers.