Adaptive Sampling for Efficient Acoustic Noise Monitoring: an Incremental Learning Approach

Sensor nodes deployed for environmental monitoring usually have access to limited resources. They need a way to prioritize observations in real time in order to maximize the utility of its resource constrained system within its limited budget. Whereas static sampling policies fail to adapt in dynamic environments, we propose online sampling policies that autonomously adapt the sampling rate to the rate of change in the environment. In this paper, we study the use of incremental machine learning models to develop adaptive sensing policies based on incremental prediction models. We applied the proposed approach in the context of noise monitoring. Our numerical results show that we can extend the battery lifetime of such a sensing system from 9 days to over 2 years (790 days) with energy saving of 97%, while keeping the accuracy of the monitoring task at 94% of the accuracy obtained using the maximum sampling rate.