Marc P. Laureta, Laurice Anne I. Laureta, Arthur Allison B. Gesmundo, and Simon Castro, New Era University, Philippines
Coffee cultivation is vital to global agriculture, yet assessing crop health often relies on time-consuming visual inspections prone to human-error. To address this, our study introduces an automated deep learning framework that evaluates coffee leaf health using two Convolutional Neural Networks (CNNs). One classifies nitrogen, phosphorus, and potassium deficiencies, while the second assesses stress severity. Both models were trained on the CoLeaf dataset using custom architectures enhanced with Swish activation, batch normalization, and dropout regularization. The system achieved 95% test accuracy for nutrient deficiency classification and 90% for stress severity. Despite high accuracy, borderline cases revealed a few challenges, incorporating advanced feature extraction and attention mechanisms could refine future detection. This research offers a scalable, automated solution for coffee plant health monitoring, advancing precision agriculture, and sustainable farm management.
Coffee plant health, Nutrient deficiency, Stress severity, Convolutional Neural Networks (CNNs), Deep Learning, Machine Learning, Image Classification. .
