FAST: A Ubiquitous Inference Computation Model for Temperature and Humidity Sensing

Temperature and humidity sensing are crucial for Internet of Things (IoT) applications. However, traditional wide-coverage sensing methods require deploying massive sensor nodes, which incur high costs that limit large-scale applications. In this article, we propose a novel ubiquitous inference computation model for accurate and cost-efficient sensing of temperature and humidity. Instead of deploying massive sensor nodes at a high cost, our model can generate reliable sensing results through inference and computation, which incurs no extra deployment cost and thus facilitates large-scale applications. To achieve this, we first construct a universal inference model for the transmission characteristics of temperature and humidity. Then, to address data reliability issues and ensure dynamic adaptability, we divide the sensing space into regions with different calculation parameters and develop a dynamic adjustment algorithm to improve reliability. Finally, we propose an adaptive algorithm for rapid deployment in various scenarios. We evaluate our system under diverse scenarios, and the results demonstrate that the errors of inference sensing results for temperature are less than 1 °C and humidity errors are less than 5%, which is comparable to the performance of deep learning models but without requiring training or manual customization, thereby effectively enhancing the sensing capabilities for IoT applications.