PhytoClip: Multimodal Wearable Sensing and Online Machine Learning for Real-Time Plant Health Monitoring and Early Stress Detection

Wearable plant sensing systems for simultaneous biochemical and physical monitoring with real-time multimodal data analysis remain limited. Here, we present PhytoClip, a multimodal wearable patch that continuously monitors leaf temperature, humidity, three volatile organic compounds (VOCs) with high selectivity, and microenvironmental light intensity and CO2 concentration. PhytoClip features a bookmark-inspired design for secure attachment to leaves of diverse morphologies, supported by a flexible printed circuit board for data acquisition, wireless communication, and cloud-based monitoring. We develop PhytoSense, an open-source machine learning (ML) framework for sensor importance ranking, multi-stress classification, and early stress detection. The integrated PhytoClip-PhytoSense platform detects and classifies nine biotic and abiotic stresses in tomato plants with 92% accuracy. Notably, P. infestans on tomato was detected within 15.5 h post-inoculation, earlier than quantitative polymerase chain reaction (qPCR) (~4 days) and visual phenotyping (~7 days), highlighting the potential of integrating multimodal wearable sensing and online ML for precision agriculture.