基于GF-1与Sentinel-2融合数据的地膜识别方法研究

随着我国地膜使用面积的增加和人们对土壤微塑料污染问题的日益关注,大尺度的地膜遥感识别已成为农业生产管理、土壤污染防治的必要手段。针对地膜光谱反射特征的复杂性以及基于单一遥感影像光谱特征识别方法错分率高等问题,该文以河北省邯郸市邱县为试验区,利用GF-1数据的空间细节与Sentinel-2数据的光谱信息进行NN Diffuse Pan Sharpening融合,据此建立地膜识别的特征矩阵(NDVI、MNDWI、NDBI、IBI、PSI),基于该特征矩阵可实现自动阈值地膜分层分类识别。多种方法的地膜识别结果精度对比表明:多源光学遥感数据融合方法的总体精度为94.87%,Kappa系数达0.89,显著优于基于单一数据源的深度学习法的精度(93.14%)以及基于传统机器学习分类方法的支持向量机(85.91%)和随机森林分类法(86.78%)的精度;通过与Sentinel-2多光谱影像融合,弥补了GF-1数据光谱分辨率低的缺陷,实现了多源数据在地膜识别中的优势互补,可为相关部门农业规划与管理以及生态环境保护等研究提供大尺度、高精度的地膜分布参考数据。

Study on Plastic Mulch Identification Based on the Fusion of GF-1 and Sentinel-2 Images

With the increasing use of plastic mulch in China and more attention to the soil microplastic pollution,remote sensing identification of large-scale plastic film becomes an essential means for agricultural production management and soil pollution control.According to the complex spectral characteristics of plastic mulch and high misclassification rate resulted by single-source remotely sensed data,multi-source data fusion technology were adopted to balance spatial resolution and ground reflec-tion information.In this paper,Qiu County,Handan City,Hebei Province was selected as study site,and the spatial details of GF-1 data and spectral information of sentinel-2 data were used for NNDiffuse Pan Sharpening.The feature matrix(NDVI,MNDWI,NDBI,IBI,PSI)of plastic mulch identification was established based on the fusion data,and the automatic threshold hierarchical classification and recognition method of plastic mulch based on the matrix was realized.Then,the accuracy of this method was compared with that based on deep learning and traditional machine learning classification methods(SVM and RF).The results show that the accuracy of automatic threshold hierarchical classification and recognition method based on multi-source fusion data(the overall accuracy was 94.87%,Kappa coefficient was 0.89)was significantly better than that of the method based on single-source data(the overall accuracy of DL,SVM and RF classification methods were 93.14%,85.91%and 86.78%,respectively).The fusion combines the advantages of high spatial resolution of GF-1 panchromatic image and high spectral resolution of sentinel-2 multispectral image for plastic mulch identification.With the development of GF-1 data in China and sentinel-2 data of European Space Agency,the method proposed in this paper could provide plastic mulch mapping data with large scale and high precision for agricultural planning and management and ecological environment protection.