基于地形地物作用机制重构超分正立体卫星林火监测图像研究

FY3, EOS/MODIS and NOAA/AVHRR are the main data source for forest fire monitoring based on satellite image in our country. Due to information of forest and non-forest, mountain-shaped landforms, river，and other land cover is hard to identify in thermal infrared monitoring image with 1km resolution, local administration can only use hot-spot coordinate data to carry out emergency response decisions and ground verification, which seriously restricts the application efficiency and accuracy of satellite monitoring image. Aiming at above, a method will be present in reconstructing super-resolution image with true topographic perception based on remote sensing imaging mechanism of interaction between terrain and ground features, and computer image processing technology, which can enhance and highlight presentation of forest fire, improve understandability of topography, achieve identification between forest fire and non-forest fire to support application of monitoring image based on positioning service in field verification and aid decision-making. Concentrating on mechanism of interaction between topographic geometry unit and surface reflectance, relationship between parameters such as terrain tiles, slope, aspect, solar elevation and azimuth in kilometer scale will be revealed,then refine terrain tiles and parameters, reflectance rate,illumination parameters will be modeled, thus a super-resolution image with true topographic perception will be reconstructed by computer digital image processing.Preliminary exploration shows that it is feasible to reconstruct super-resolution image with true topographic perception in tens of meters from original 1km data, which will show the equivalent visual effect of land satellite. Research will show great significance in breaking through technology bottlenecks of forest fire monitoring, and promoting scientific and technological progress in the industry.

FY3、EOS/MODIS和NOAA/AVHRR是当前我国卫星林火监测的主要数据源，其1km分辨率的热红外监测图像难以区别森林与非森林、地貌、河流及其它地物，基层单位只能用热点坐标开展应用，严重制约卫星监测的效率及精准性。针对该问题，提出利用遥感成像作用机理及计算机图像处理技术，重构超分正立体监测图像方法，以增强和突出森林火情信息表达，提高地形地貌及地物覆盖的可辨识度及可读性，实现林火与非森林火的区别，支持监测图像在野外核查和辅助决策中的定位服务应用。研究拟聚焦于地形地物作用的遥感成像机理，揭示公里级地形瓦片、坡度、坡向、太阳高度角、方位角等参量的成像关系，建立精细化地形瓦片、地物反射率及光照参量,重构超分正立体影像。前期探索表明，有望将1km原始数据重构达几十米分辨率、与陆地卫星图像视觉相当的超分正立体图像。研究对突破林火监测的技术瓶颈、推动行业科技进步具有重要意义。

针对我国森林防火行业部门采用FY、EOS/MODIS和NOAA/AVHRR监测森林火灾影像质量提高的需求，通过项目实施，完成基于云南近年发生的林火，对行业常用的具有1km粗分辨率且为反立体，不清晰的FY、MODIS卫星热点影像数据进行超分正立体处理研究。构建的超分正立体影像与原发布的监测图像相比，可视化效果有显著提高，原1km影像的可视化效果可达到12.5m、30m、90m清晰程度，且火环境中地形为正立体，地形地貌、植被及土地覆盖信息突出，小火得到增强，对漏报的小火进行了恢复。并根据火环境，增加了火点是否为林火与非林火的判识信息，项目完成中明显提高了发布监测图像的质量，使热点的地面核实反馈效率得到极大提高。项目执行期间发表论文8篇，其中SCI论文1篇，EI检索1篇，CSSCI论文1篇，CSCD 2篇，中文核心3篇；获得授权发明专利3项；省部级科技进步奖等3项；培养硕士研究生16名，毕业7名。1名教师学位获得博士学位。