文献类型：期刊文献

英文题名：Deep learning-enabled forest fire detection: Yolov8 optimization via multiscale feature alignment and false alarm suppression

作者：Xiao, Yundan[1] Li, Fan[1,2,3] Guan, Li[4] Cheng, Quanying[5] Feng, Pengfei[6]

第一作者：肖云丹

机构：[1] Institute of Forest Resource Information Technique, Chinese Academy of Forestry, Beijing, 100091, China; [2] National Mobile Communications Research Laboratory, School of Information Science and Engineering, Southeast University, Nanjing, 211189, China; [3] Key Laboratory of Forest Protection of National Forestry and Grassland Administration, Ecology and Nature Conservation Institute, Chinese Academy of Forestry, Beijing, 100091, China; [4] Faculty of Sciences, Beijing University of Technology, Beijing, 100124, China; [5] China Fire and Rescue Institute, Beijing, 102202, China; [6] Research Institute of Forestry Policy and Information, Chinese Academy of Forestry, Beijing, 100091, China

年份：2025

卷号：14005

外文期刊名：Proceedings of SPIE - The International Society for Optical Engineering

收录：EI(收录号：20260219902889)

语种：英文

外文关键词：Alarm systems - Deforestation - Disaster prevention - Disasters - Ecosystems - Feature extraction - Fire detectors - Fire hazards - Fireproofing - Fires - Image resolution - Information management - Learning systems - Object detection - Object recognition - Semantics

摘要：Forest fires pose an increasingly severe global threat to ecosystems, biodiversity, and human lives, yet traditional detection methods - such as manual patrols and rule-based image processing - are plagued by limitations including slow response speeds, high false alarm rates, and poor adaptability to complex environments; to address these issues, this study has developed an automated deep learning-based forest fire detection system that integrates high-level semantic features, low-level spatial features, and false alarm suppression by optimizing the YOLOv8 model, while also focusing on constructing a forest fire dataset obtained through video surveillance, adopting data preprocessing methods like adaptive histogram equalization and mosaic augmentation, and employing model training techniques such as the focal loss function (which is designed to tackle the class imbalance problem). This study further proposes an optimization scheme and framework for the YOLOv8 model tailored specifically to forest fire detection scenarios, a framework that can be further extended to edge computing platforms to enable real-time monitoring, thereby providing support for proactive fire prevention, rapid response, and efficient disaster management and ultimately mitigating the global impact of forest fires. ? 2025 SPIE. Downloading of the abstract is permitted for personal use only.