ENHANCING EARLY FOREST FIRE DETECTION: A COMPREHENSIVE FRAMEWORK INTEGRATING WIRELESS SENSOR NETWORKS (WSNS), UNMANNED AERIAL VEHICLES (UAVS), AND ARTIFICIAL INTELLIGENCE (AI)

---

Tezin Türü: Yüksek Lisans

Tezin Yürütüldüğü Kurum: Orta Doğu Teknik Üniversitesi, Odtü Kuzey Kıbrıs Kampusu (Kktc-Güzelyurt), Bilgisayar Mühendisliği Bölümü, Türkiye

Tezin Onay Tarihi: 2025

Tezin Dili: İngilizce

Öğrenci: Hamzeh Abu Ali

Danışman: Enver Ever

Özet:

Forest fires pose significant environmental, ecological, economic, and social threats globally, necessitating rapid and accurate detection systems for timely intervention. Traditional detection methods, such as watchtowers, manual patrols, and satellite imagery, face challenges including limited coverage, delays, and accuracy constraints. In response, this thesis introduces a three-tier edge-centric framework integrating wireless sensor networks (WSNs), wireless multimedia sensor networks (WMSNs), unmanned aerial vehicles (UAVs), and lightweight artificial intelligence (AI) models for efficient early detection of forest fires.

Our proposed architecture improves accuracy, energy efficiency, and communication reliability by integrating scalar sensors for initial detection, smart sensors with a machine learning model for intermediate verification (achieving a 94% F1-score with a minimal subset, comparable to the 95% of state-of-the-art methods), and UAVs equipped with a lightweight convolutional neural network (CNN) for final confirmation. The CNN model achieves 100% accuracy and F1-score on the FireMan-UAV-RGBT dataset and 99.57% accuracy with a 99.5% F1-score on the UAV-FFDBs dataset. Despite its strong performance, the model remains compact at 1.6 MB, significantly smaller than the 98 MB state-of-the-art using the FireMan-UAV-RGBT dataset, and delivers an inference speed of 157 ms per image on edge devices, validating its practical deployment.

Extensive simulations reveal that the proposed framework significantly reduces end-to-end delay to 813.59 ms compared to traditional WSN-only methods (865.84 ms) and WSN combined with machine learning approaches (1066.18 ms). Additionally, it achieves a 100% packet delivery ratio and increased throughput (7.05 kbps versus 3.80 kbps and 3.06 kbps, respectively) against these methods. Real-world WSN testbed experiments further confirm these findings, showing a packet delivery ratio of 97%, latency of 144.39 ms (lower than simulation latency of 258.37 ms), and energy consumption of 0.0559 J/s compared to simulation results of 0.0442 J/s, closely aligning and validating the framework’s feasibility and effectiveness for real-time forest fire monitoring and rapid response.