An Architecture for Information Fusion and for Detection, Identification and Treatment of Outliers in Wireless Sensor Networks

Abstract

The fields of precision agriculture, environmental engineering, among others, often have applications that use sensors to monitor the environment. Examples of such applications include pest control, irrigation process, soil fertility mapping, monitoring of forest areas and of urban rivers etc. Wireless sensor networks (WSNs) have been proposed as distributed infrastructures for these applications. These networks produce a large volume of data and use low-cost sensors. However, these sensors usually have low-reliability, generating anomalous data (outliers), affecting the final quality of the monitoring. These conditions imply the need to use methods for outlier detection and treatment, allowing the correct operation of the network and increasing the confidence in the monitored data. This article proposes an architecture for information fusion focusing on low-reliability sensors. The architecture is integrated with techniques for detection and treatment of outliers, and it was evaluated through two case studies. The first one involving low-cost barometric pressure sensors, whose monitored data were processed by outlier detection techniques. The second one involves the LUCE (Lausanne Urban Canopy Experiment) large-scale scenario, whose database is fed by 84 sensors for monitoring weather conditions. The results show that some of the low-level fusion methods and outliers detection techniques, when combined and organized according to the proposed architecture, can replace a single centralized, high-cost sensor, maintaining the confidence of the monitored data.