Contact-free wheat mildew detection with commodity wifi

Abstract

Mildew is recognized as one of the most critical causes of the damages in food storage. Due to the complex operation and high cost, many advanced detection instruments cannot be widely deployed, while the detection of grain mildew are mostly carried out manually (i.e., inspection by experts) nowadays with very low efficiency. To address this problem, we present a non-destructive, non-intrusive, and low-cost mildew detection system for stored wheat implemented with off-the-shelf WiFi devices, which represents a new application of the Internet of Things (IoT) for smart agriculture. In this paper, we introduce the impact of wheat mildew in stored food, and demonstrate that it is viable to utilize WiFi Channel State Information (CSI) amplitude for detection of mildew in stored wheat. Next, we propose the MiFi system, which comprises sensing of WiFi CSI data, data preprocessing, a radial basis function (RBF) neural network-based detection model, and mildew detection. We conduct extensive experiments to validate the performance of the proposed MiFi system using real grain samples. The results show that MiFi achieves an average detection accuracy of over 90% in both line-of-sight (LOS) and non-line-of-sign (NLOS) scenarios, as well as a comparable detection performance as manual detection by an expert.