Modeling the Operating Characteristics of IoT for Underwater Sound Classification

Abstract

In remote sensing applications, constraints of power, processing, and communications limit information acquisition. Pre-training the IoT improves the performance in information acquisition tasks such as detection, classification, and estimation. However, light and inexpensive IoT hardware still need to operate with strict resource constraints. In this paper, we provide a method for modeling the IoT operating characteristics that link information acquisition performance to resource use. The goal of modeling is to improve understanding of how to optimally utilize constrained resources to improve information acquisition performance. The proposed method is demonstrated using field, simulation, and lab-based experiments with real data and practical hardware for an underwater sound classification application utilizing deep learning.