Andy Desman Lo, Elvin Nur Furqon, Junaidul Islam, Isack Farady, Kahlil Muchtar, Ronnie Concepcion, Chih-Yang Lin
Early detection of plant responses to stress and pathogen stimuli is essential for maintaining crop health and optimizing yield. Electrophysiological signals provide a non-invasive, real-time method for monitoring plant response. This study explores the application of a one-dimensional autoencoder for detecting abnormal patterns in electrophysiological signals recorded from plant branches. The autoencoder achieves this by learning to compress the input signals and reconstruct them from the compressed signal. Signals that show high reconstruction errors are considered to significantly differ from the learned normal patterns and are flagged as potential anomalies. Signals from multiple plant species such as Araucaria, Cyperus, Hedera, Mentha, Ocimum, Plectranthus, Ruta, Salvia, Solanum were preprocessed digitally using detrending, temporal shifting, and normalization techniques. Model performance was evaluated using Mean Absolute Error (MAE) and Intersection over Union (IoU) metrics. Results indicate that reliable anomaly detection was achieved for four species Ocimum (IoU: 0.615), Plectranthus (0.701), Ruta (0.631), and Solanum (0.670), while other species exhibited lower IoU scores, likely due to weaker or less distinguishable electrophysiological responses. The result also shows that while autoencoder approach can detect abnormalities in some plant species, their performance is highly correlated on species-specific signal pattern. © 2025 IEEE.
National Central University, Taoyuan, Taiwan; Yuan Ze University, Taoyuan, Taiwan; Universitas Syiah Kuala, Banda Aceh, Indonesia; De La Salle University, Manila, Philippines
Research at a Glance
Register to unlock
Topics & SDG Alignment
Register to unlock
Collaboration
Register to unlock
Author Profile (Selected)
Register to unlock
References Overview
Register to unlock
Journal & Source
Register to unlock
Metadata & Integrity
Register to unlock