Real-Time Auto Calibration for Heterogeneous Wireless Sensor Networks

Abstract

The constant technological advances bring new devices to the market every day. Due to this, heterogeneous Wireless Sensor Networks (WSN) are common scenarios in many applications. Neural Network (NN) based models may implement particular features provided by sensor hardware to collect surrounding environment information. Thus, a sensor can provide a specific group of features while others do not. In this perspective, it may be required, for some sensors in a WSN, to be trained and have their data manually categorized, which does not scale, particularly for large WSN setups. In light of this problem, this paper proposes a Real-Time (RT) auto-calibration framework to allow WSN devices to collaborate in the training process to enable new uncalibrated devices to join the network. The method does not need previous knowledge about sensor features. Also, the proposal is validated by practical experiments evaluating its accuracy in image classification. The provided experimental results demonstrate the feasibility of the proposed method.