CALYE: a deep learning approach for calamansi fruit yield estimation through CNN-based flower detection
TL;DR
An automated yield prediction system using drone imagery, deep learning, and predictive analytics enables accurate yield estimation and supports improved harvest planning and decision-making and showed that manual upload provided more reliable detection than real-time analysis.
Abstract
Accurate yield estimation is important for improving agricultural productivity and farm management, particularly for calamansi (Citrofortunella microcarpa), a key citrus crop in the Philippines. Traditional methods, such as manual counting, are labor-intensive and error-prone. This study developed an automated yield prediction system using drone imagery, deep learning, and predictive analytics. A dataset of 18,628 flower and 11,322 fruit samples from 100 trees was used to train three YOLOv8 models, with YOLOv8m achieving the best performance (precision = 1.00, recall = 0.71, F1 score = 0.58, mAP = 0.556). The model was integrated with ByteTrack for accurate detection and counting and deployed in a Django-based web application for automated yield prediction using ridge regression. Results showed that manual upload provided more reliable detection than real-time analysis. Overall, the system enables accurate yield estimation and supports improved harvest planning and decision-making.