融合遥感和空间关系特征的森林火灾遥感探测研究  

Forest fire detection integrating remote sensing and spatial relationship features

作　　者：高世莹 张锦水 周佳乐 GAO Shiying;ZHANG Jinshui;ZHOU Jiale(State Key Laboratory of Remote Sensing Science,Beijing Normal University,Beijing,China;Institute of Remote Sensing Science and Engineering,Faculty of Geographical Science,Beijing Normal University,Beijing,China;Beijing Engineering Research Center for Global Land Remote Sensing Products,Faculty of Geographical Science,Beijing Normal University,Beijing,China)

机构地区：[1]北京师范大学遥感科学国家重点实验室,北京 [2]北京师范大学地理科学学部遥感科学与工程研究院,北京 [3]北京师范大学地理科学学部北京市陆表遥感数据产品工程技术研究中心,北京

出　　处：《北京师范大学学报(自然科学版)》2024年第6期852-860,共9页Journal of Beijing Normal University(Natural Science)

基　　金：国家自然科学基金重大资助项目(42192580,42192584)。

摘　　要：综合分析短波红外波段、地形校正、空间关系特征对过火面识别的影响,提出了融合多源特征的森林火灾分类方法.本文采用地形校正和增加短波红外波段的方法,以降低山体阴影和烟雾对过火面识别的影响,并引入空间语义关系特征进一步修正过火面的识别结果.以2022年重庆市的森林火灾为例,基于哨兵2号影像和支持向量机对森林火灾过火面进行遥感识别.研究表明:1)短波红外波段可降低烟雾对过火面识别的干扰,识别精度提高3%~8%;2)地形校正可在一定程度上改善山体阴影的影响,但对过火面识别精度提高不大;3)引入空间关系特征能明显降低裸地、硬质铺装地等因光谱混淆造成的干扰,精度会提升3%~4%.该方法物理机制明确,对森林火灾(尤其是燃烧的森林火灾)可有效地提取.A forest fire classification method integrating multi-source features is proposed after analyzing effects of short-wave infrared bands,terrain correction,and spatial relationship features on burnt area recognition.Terrain correction and short-wave infrared band features are used to reduce effects of mountain shadows and smoke on burnt area recognition,and spatial relationship features are introduced to correct burning area recognitions.The forest fire in Chongqing Municipality in 2022 is studied from Sentinel 2 images and support vector machine model.Application of short-wave infrared bands could reduce interference from smoke,with the recognition accuracy improved by 3%-8%.Terrain correction can reduce interference from mountain shadows,with little effect on recognition accuracy.Introduction of spatial relationship features can significantly reduce interferences from bare ground,hard paved ground,with the accuracy being improved by 3%-4%.The physical mechanism of this method is clear and can effectively extract forest fires(especially burning forest fires).

关 键 词：森林火灾 遥感提取 烟雾 山体阴影 光谱混淆 

分 类 号：P407.8[天文地球—大气科学及气象学] S762[农业科学—森林保护学]