PhytoNodes for Environmental Monitoring: Stimulus Classification based on Natural Plant Signals in an Interactive Energy-efficient Bio-hybrid System

Cities worldwide are growing, putting bigger populations at risk due to urban pollution. Environmental monitoring is essential and requires a major paradigm shift. We need green and inexpensive means of measuring at high sensor densities and with high user acceptance. We propose using phytosensing: using natural living plants as sensors. In plant experiments we gather electrophysiological data with sensor nodes. We expose the plant Zamioculcas zamiifolia to five different stimuli: wind, temperature, blue light, red light, or no stimulus. Using that data we train ten different types of artificial neural networks to classify measured time series according to the respective stimulus. We achieve good accuracy and succeed in running trained classifying artificial neural networks online on the microcontroller of our small energy-efficient sensor node. To indicate later possible use cases, we showcase the system by sending a notification to a smartphone application once our continuous signal analysis detects a given stimulus. Data repository for our paper "PhytoNodes for Environmental Monitoring: Stimulus Classification based on Natural Plant Signals in an Interactive Energy-efficient Bio-hybrid System", submitted to the GoodIT conference. Please refer to the paper for more information. Contents of this repository mu_interface: Code for our data collection plant experiments, based on Raspberry Pis and the Cybertronica phytosensing and phytoactuating system. raw_data: The datasets from our plant experiments for the stimuli wind, temperature, red light, blue light, and no stimulus. dl-4-tsc: Deep learning framework developed by Fawaz et. al (Deep learning for time series classification: a review) and adapted to our use case. Find the training and testing datasets in the archives folder as well as the trained classifiers in the results folder. classification_results.ods: Overview of the results from the deep learning framework (accuracy, precision, recall, training time). TFLite_Models: The trained classifiers in TensorFlow Lite Format. 00_AI_BLE_MeasuringOnlyWind: Source code for classification on STM-based PhytoNodes (using MCDCNN two-class classifier) and Bluetooth communication. The code is written for the STM32WB55 Nucleo board and can be transferred to the dongle. zavrsniProjekt_iOS: Source code of the iOS app used to receive data from the STM-based PhytoNodes. Watchplant_application_documentation.pdf: Instructions to build and use the iOS app.