Adaptive classification of special events in agroecological monitoring systems for pest management

Abstract

With powerful feasible integration of distributed sensing capability, real-time data analysis, and remote surveillance, wireless sensor networks (WSNs) provide a new insight into agroecological observation in ways of extending spatial and temporal scales. Through WSNs, some unexpected phenomena can be found, and new paradigms can be developed. Although employing the WSN technology can facilitate agroecological observation, one of the major challenges that need to be overcome is the abnormal readings caused by sensor failure, energy depletion of sensor nodes, low durability of protective cases in a wild environment, unreliable wireless communication, etc. In this study, a WSN-based ecological monitoring system is presented and practically deployed in a field to monitor the number of the oriental fruit fly (Bactrocera dorsalis (Hendel)) and capture long-term and up-to-minute natural environmental fluctuations. Moreover, an adaptive classification approach, built upon self-organizing maps and support vector machines, is incorporated into the monitoring system to automatically identify special events of pest outbreaks and sensor faults. Once the events are detected, farmers and government officials can take precautionary action in time before pest outbreaks cause an extensive loss or schedule maintenance tasks to repair monitoring devices. The proposed classification approach is easily adopted in different monitored farms, and it can automatically identify special events based on machine learning techniques without requiring additional manpower.