落叶松毛虫害的大规模爆发导致了森林生态系统和经济的严重损失,快速、准确地对落叶松毛虫害发生区域进行识别具有重要的意义.本研究采用知识图谱技术,对与森林虫害相关的敏感特征进行筛选,并综合利用多源数据,分别构建3种不同数据组合的随机森林模型,对研究区内落叶松毛虫害发生区进行遥感识别.结果表明:1)依据图谱构建流程可以实现知识图谱构建及筛选,本研究实现了构建森林虫害遥感识别特征图谱,并筛选了中国东北地区落叶松毛虫害的遥感识别特征;2)知识图谱与遥感技术手段相结合,能够为构建虫害识别模型提供有效特征;3)与单一数据源相比,基于多源数据的落叶松毛虫害识别效果更好,本研究综合使用Sentinel-1A、Sentinel-2A和地形数据的总精度和Kappa系数分别为92.78%,0.876 6.
Large-scale outbreaks of Dendrolimus superans infestation have led to severe losses of forest ecosystems and economies.It is of great significance to identify the area of Dendrolimus superans infestation quickly and accurately.In this study,knowledge graph technology was used to screen the sensitive features related to forest pests,and a random forest model with three different data combinations was constructed by comprehensive use of multi-source data to identify the Dendrolimus superans infestation occurrence area in the study area by remote sensing.The results are as follows:1)The knowledge graph can be constructed and screened according to the graph construction process.In this study,the characteristics of forest insect pests were established by remote sensing,and the characteristics of Dendrolimus superans infestation in Northeast China were screened by remote sensing.2)The combination of knowledge graph and remote sensing technology can provide practical features for constructing the pest identification model;3)Compared with a single data source,the identification effect of Dendrolimus superans infestation based on multi-source data is better.In this study,the overall accuracy and Kappa coefficient of Sentinel-1A,Sentinel-2A and topographic data were 92.78%and 0.876 6,respectively.