Answers are grounded in this paper and cite their sources.
This article presents a distributed thermal sensing and cloud–edge collaborative intelligent framework for real-time classification of indoor moisture and humidity. In this work, multiple FLIR Lepton 3.5 radiometric thermal sensors are deployed across humidity-prone locations, such as kitchens, showers, basements, and window areas in the home, to monitor moisture-induced temperature patterns continuously. Each thermal sensor node is connected to a Raspberry Pi 4B for data acquisition and wireless transmission to a GPU-enabled cloud server. Then, a lightweight, resource-efficient deep learning (DL) classifier is trained on the distributed thermal sensing data. In this work, we curated a novel dataset, thermal humidity extraction and recognition for maintenance in the indoor domain (THERMID), collected from multiple locations and comprising 22 classes of moisture, cold-spot, condensation, and temperature-state conditions. Furthermore, the proposed lightweight, resource-efficient DL model is centrally trained using the aggregated thermal sensing data and deployed on Rock Pi C+ edge devices for low-latency, on-device inference. The proposed lightweight DL model outperforms existing state-of-the-art pretrained models, providing 98.91% test accuracy with only a 2.6-MB model size and requiring only 0.542 GFLOPs for inference. In addition, the model is evaluated on Raspberry Pi 4B and Rock Pi C+ edge modules and is observed to provide 273- and 314-ms inference per image, respectively. The proposed system provides a scalable and energy-efficient solution for autonomous indoor moisture diagnostics, supporting proactive building health monitoring and integration with smart home control systems.
No takes yet. Share an insight, caveat, or question.
Reddy et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: