基于物联网视觉的森林火灾监控系统设计

At present, the forest fire is seasonal boundary fuzzy and difficulty to predict. The research of forest fire prevention based on Internet of things is attracting more and more attention. Presently, proposed systems can not solve the permanent placement of sensor nodes, the coverage targeted detection area and problems of non-structural forest environment. This makes that such researches is not practicable. . This application focusing on the designing of the sensor node from sensing layer, target coverage to design forest fire monitoring system based on the vision of internet of things from four aspects: Firstly, design of the real-time task level energy management strategy for sensor node and energy self-collecting system; Secondly, multiple coverage of target area and the network protocol applications; Thirdly, solution for data fusion and transport of temperature, humidity data and acquired video. Based on transfer learning mechanism of forest fire image recognition technology to improve the fire assessment and identification; Finally by simulation and experiment forms an innovation-driven high precision forest fire prevention system.

当前森林火灾呈现出季节性界限模糊、可预测性难度增大等新特点，基于物联网的森林防火研究越来越得到重视。现有物联网防火方案的研究往往未能解决传感器节点永久布置，目标区域多重覆盖与探测，火点判别时面临的森林环境非结构性问题，使得其研究不具有实际应用意义。本课题着眼于设计感知层的可永置的无线传感器节点，基于目标覆盖与物联网视觉的森林火点图像识别技术，从四个方面对系统进行设计研究。首先研究匹配节点任务级别的实时能量管理策略，设计微能量自收集节点。其次结合网络覆盖控制技术进行目标区域与节点的多重覆盖，应用无线传感网的网络协议，解决温湿度、视頻复合无线传感器网络数据采集、融合、探测、处理和查询等一系列的数据反馈，运用基于迁移学习机制的森林火灾图像识别技术对目标区域火点评估与识别；最后通过仿真实验和应用示范对上述理论、技术、协议和算法进行验证，形成创新驱动、应用牵引的基于物联网视觉的高精度森林防火系统。

当前森林火灾呈现出难控性增强、季节性界限趋于模糊、火灾可预测性难度增大等新特点，基于物联网的森林防火研究越来越得到重视。现有物联网防火方案的研究往往未能解决传感器节点永久布置，目标区域多重覆盖与探测，火点判别时面临的森林环境非结构性问题，使得其研究不具有实际应用意义。本课题着眼于从感知层的可永置的无线传感器节点，基于目标覆盖与物联网视觉的森林火点图像识别技术，以系统能量能耗管理技术、优化能量管理算法、以传感器部署技术等传感器关键技术为理论基础，实现无线传感网的节点部署、探测判决到决策处理的过程。通过视频等复合传感器捕获反馈森林现场图像后，引入自动火灾图像特征选择机制，设计半监督的林火图像识别算法。通过本项目的研究，在国内外核心期刊和会议上发表多篇学术论文，部分成果形成自主知识产权(例如发明专利和软件著作权)，在物联网能耗管理理论、节点部署策略与图像识别研究中获得了相应的结果，设计的林火预警系统具有广泛的地区适用性。