ABSTRACT
High-fidelity sensor data quality is the fundamental base of smart agriculture. Since crop information inference and cultivating strategy optimization mainly depend on data-driven methods; the data quality assessment is essential to ensure the reliability of the IoT systems for smart agriculture. The traditional data quality assessment methods focus on sensor data consistency with the costly reference truth, which is often not scalable for agricultural applications.
We present DaQual, a data-driven data quality assessment scheme for tree trunk relative water content sensors in a pomegranate orchard. The objective is to leverage the neural network architecture to learn the underlying relationship between sensors deployed on the same farm and utilize the quantified associated relationship between sensors to assess its reliability.
DaQual first builds a prediction model for each sensor with all other sensors’ data as input.
Then, the trained network’s parameter values (weights) are used to quantify the contribution of the sensor – the higher the contribution, the higher its data quality. We evaluate DaQual via a real-world tree trunk relative water content sensor dataset with nine sensors deployed in a pomegranate orchard, and our scheme demonstrates up to 1.8× improvements when used to select a subset of sensors with high data quality
