ANALYSIS, DESIGN, AND REALIZATION OF INDUSTRIAL INTERNET OF THINGS NETWORKS

Abstract

Currently, the Industrial Internet of Things (IIoT) is reshaping into highly efficient, intelligent, and context-aware systems where remote sensing and aerial monitoring are concerned. Towards realizing Unmanned Aerial Vehicle (UAV) based surveillance for disaster response and urban planning, this thesis designs and implements two semantic communication pipelines for UAV-based surveillance that address practical challenges in real-time bandwidth efficiency and geospatial reasoning. Geometric and semantics information are jointly and automatically compressed using a Joint Source Channel Coding (JSCC) autoencoder to reduce bandwidth cost while preserving information for the downstream. The second approach is detecting objects onboard the UAV, augmenting those detections with UAV telemetry data, and estimating real-world coordinates in real time. The empirical results show that the JSCC model can compress aerial imagery with little loss of semantic features. At the same time, the onboard system has an average precision (AP) of 7.35%, with excellent localization in the region of a few meters. In particular, the second pipeline allows the UAV to send only object class labels and coordinates, which is ideal for long-distance or high-altitude operations. These contributions jointly enable designing the next generation of IIoT UAV systems that can efficiently communicate and autonomously act intelligently. Based on this, two topics are discussed in later research to explore tradeoffs, challenges, and further directions: multimodal fusion and federated semantic inference.