目前遥感林火监测主要侧重极轨卫星火点探测精度, 而基于多源遥感影像进行火点、 烟雾特征等综合火场信息遥感监测识别研究较少。 以云南省安宁市2020年5月9日森林火灾为研究对象, 基于高分六号卫星宽幅(GF-6 WFV)数据和风云三号D极轨气象卫星中分辨率光谱仪(FY-3D MERSI)数据进行火场烟雾、 火烧迹地提取及火点判识。 首先根据GF-6 WFV影像, 选取6种光谱特征指数, 以最大似然法、 支持向量机分类法及随机森林分类法, 识别火场烟雾和火烧迹地, 并进行精度评价; 然后, 基于1 km的FY-3D MERSI中红外通道数据, 对潜在火点识别算法进行改进, 根据FY-3C VIRR和MODIS火点探测基本原理, 利用动态阈值和上下文检测法识别火点, 再结合250m分辨率的远红外通道优化识别结果。 最后结合两种数据提取的烟雾、 火点及火烧迹地信息, 探讨分析GF-6 WFV与FY-3D MERSI对于林火的监测能力。 结果表明: 通过5种特征指数及GF-6 WVF数据的8个波段, 能有效识别出烟雾及火烧迹地, 3种分类方法中随机森林分类效果最佳, 总体分类精度和Kappa系数为97.20%和0.955; 改进后的FY-3D MERSI数据火点识别算法, 能有效提高火点识别的准确率; 将中红外通道与远红外通道相结合探测火点, 能使火点识别能力由千米级提高至百米级; 综合GF-6 WFV及FY-3D MERSI数据可有效提取火场的烟雾、 火烧迹地及火点信息。 利用多源数据, 可多方位进行林火监测预警, 对于提高卫星遥感林火监测能力具有重要意义。

At present, remote sensing forest fire monitoring mainly focuses on the accuracy of fire point detection by polar-orbiting satellites. At the same time, there is less research on remote sensing monitoring and identification of fire points, smoke characteristics and other comprehensive fire information based on multi-source remote sensing images. The forest fire of May 9, 2020, in Anning City, Yunnan Province, was studied based on the Gaofen-6 wide-field (GF-6 WFV) data and the FY-3D polar-orbiting meteorological satellite medium-resolution spectrometer (FY-3D MERSI) data for smoke, burned areas extraction and fire point identification. Firstly, Based on GF-6 WFV data, six spectral feature indices were selected to identify fire smoke and fire trails by maximum likelihood, support vector machine, and random forest classification methods, and evaluated for accuracy. Then, Based on the 1 km mid-infrared channel data of FY-3D MERSI, the potential fire point identification algorithm is improved, and the basic principles of FY-3C VIRR and MODIS fire point detection are combined with dynamic threshold and context detection method to identify fire points. Then the identification results are optimized by combining the far-infrared channel with 250 m resolution. Finally, the information on smoke, fire points and fire trails extracted from the two kinds of data were combined to explore and analyze the monitoring capability of GF-6 WFV and FY-3D MERSI for forest fires. The results show that the smoke and burned areas can be effectively identified by five feature indices and eight bands of GF-6 WVF data, and the random forest classification is the most effective among the three classification methods, with an overall classification accuracy and Kappa coefficient of 97.20% and 0.955. The improved fire point recognition algorithm for FY-3D MERSI data can effectively improve the recognition accuracy of fire points. Combining the mid-infrared -and far-infrared channels to detect fires can improve the fire detection accuracy from kilometer to 100 meter level. The combined GF-6 and FY-3D MERSI data can effectively extract smoke, burned areas and fire point information from the fire site, and the use of multi-source data can carry out forest fire monitoring and early warning in multiple directions, which is of great significance to improve the capacity of satellite remote sensing forest fire monitoring.