The appraisal of stress in plants is of great relevance in agriculture and any time the transport of living plants is involved. Wireless sensor networks (WSNs) are an optimal solution to simultaneously monitor a large number of plants in a mostly automatic way. A number of sensors are readily available to monitor indicators that are likely related to stress. The most common of them include the levels of total volatile compounds and CO2 together with common physical parameters such as temperature, relative humidity, and illumination, which are known to affect plants’ behavior. Recent progress in microsensors and communication technologies, such as the LoRa protocol, makes it possible to design sensor nodes of high sensitivity where power consumption, transmitting distances, and costs are optimized. In this paper, the design of a WSN dedicated to plant stress monitoring is described. The nodes have been tested on European privet (Ligustrum Jonandrum) kept in completely different conditions in order to induce opposite level of stress. The results confirmed the relationship between the release of total Volatile Organic Compounds (VOCs) and the environmental conditions. A machine learning model based on recursive neural networks demonstrates that total VOCs can be estimated from the measure of the environmental parameters.

Indoor locations with limited air exchange can easily be contaminated by harmful volatile compounds. Thus, is of great interest to monitor the distribution of chemicals indoors to reduce associated risks. To this end, we introduce a monitoring system based on a Machine Learning approach that processes the information delivered by a low-cost wearable VOC sensor incorporated in a Wireless Sensor Network (WSN). The WSN includes fixed anchor nodes necessary for the localization of mobile devices. The localization of mobile sensor units is the main challenge for indoor applications. Yes. The localization of mobile devices was performed by analyzing the RSSIs with machine learning algorithms aimed at localizing the emitting source in a predefined map. Tests performed on a 120 m2 meandered indoor location showed a localization accuracy greater than 99%. The WSN, equipped with a commercial metal oxide semiconductor gas sensor, was used to map the distribution of ethanol from a point-like source. The sensor signal correlated with the actual ethanol concentration as measured by a PhotoIonization Detector (PID), demonstrating the simultaneous detection and localization of the VOC source.

No abstract