Wireless Sensor Based Data Analytics for Precision Farming

With advances in the Internet of Things, the use of Wireless Sensor Networks (WSN) has been widely proposed for monitoring and automation of farm processes under the umbrella of Precision Farming. In conventional WSN systems, data gathered by sensors is transmitted to remote cloud servers for analysis. These systems, however, incur delay in getting insights into the processes due to the high volume of data generated on the farms coupled with the poor Internet connectivity. This negatively affects the delay-sensitive applications that require immediate response. The Fog Computing paradigm suggests a shift in intelligence from the cloud towards the network edges to cater to the requirements of delay-sensitive applications. It proposes the use of compute, memory and networking resources available at edge devices such as gateways, routers and sensors to reduce dependency on cloud and, thereby, improve the responsiveness of the system. In this work, we focus our attention on the development of on-board intelligence for sensor devices in the context of Precision Farming. Firstly, we identify gaps in the current WSN-based Precision Farming technologies and examine the suitability of Edge Mining, an instance of Fog Computing, for real-time event detection in farm processes. In addition, we propose an extension of the Edge Mining approach to allow for context-aware operation of sensor devices in farms. A WSN prototype consisting of a plug-n-play universal sensor device and gateway node has been designed to validate the performance of these algorithms. Next, we develop two cooperative frameworks - Collaborative Edge Mining and Iterative Edge Mining, to represent the analytic problems as a set of cooperative Edge Mining-based tasks for parallel and sequential analysis respectively within WSN. The cooperation between tasks allows for scaling of analysis within and across devices to improve computational capability of the network. Finally, we discuss resource management through cooperative computing within WSN. Cooperation between devices is considered to improve accuracy and timeliness of in-network analytics while optimizing the use of energy resources of sensor devices for improved network longevity.