基于MODIS时序数据物候特征下的多源遥感玉米提取

Maize Extraction by Multi-source Remote Sensing Based on Phenological Characteristics of MODIS Time Series Data

为解决农作物遥感分类易混淆的问题,准确获取玉米种植结构信息,提出了一种能够充分利用高时间分辨率和中高空间分辨率数据优势的多源遥感数据玉米提取方法。以沈阳为研究区域,利用S-G滤波重构MODIS-EVI时序曲线提取物候特征,同时利用EVI转换模型平衡Landsat-8、Sentinel-2、GF-1与MODIS之间的EVI差异构建了30 m分辨率的时序数据,采用决策树方法,基于MODIS物候特征对多源时序数据分类,逐层掩膜水稻、大豆等易混淆地类获取玉米提取结果,并采用决策树与混合像元分解结合的方法进一步提高玉米提取结果的精度。结果表明:基于多源转化遥感数据的决策树分类总体精度与Kappa系数分别为92.27%和0.8825,相较于CART决策树、随机森林、最大似然法,其分类总体精度和Kappa系数均有较大幅度的提高,相较于数据未经模型转换的决策树分类的总体精度和Kappa系数分别提高4.59个百分点和0.0663。决策树分类后结合混合像元分解的玉米提取总精度提高至95.98%,玉米分类精度得到进一步提高。

In order to solve the problem of confusing the remote sensing classification of crops and accurately obtain the information of maize planting structure,we proposed a multi-source remote sensing data maize extraction method that can make full use of the advantages of high temporal resolution and medium and high spatial resolution data.In this paper,Shenyang was taken as the research area,and S-G filtering was used to reconstruct the MODIS-EVI time series curve to extract phenological characteristics.At the same time,the EVI conversion model was used to balance the EVI difference between Landsat-8,Sentinel-2,GF-1 and MODIS to construct a 30 m resolution time series data,and the decision tree method was used to classify multi-source time series data based on MODIS phenological characteristics.The corn extraction results were obtained layer by layer by masking rice,soybean and other confusing species.The method of decision tree combined with mixed cell decomposition was used to further improve the accuracy of maize extraction results.The results show that the overall accuracy and Kappa coefficient of decision tree classification based on multi-source conversion remote sensing data are 92.27%and 0.8825,respectively,which are greatly improved compared with the overall accuracy and Kappa coefficient of CART decision tree,random forest and maximum likelihood classification,and increase by 4.59 percentage points and 0.0663,respectively,compared with the overall accuracy and Kappa coefficient of decision tree classification without model transformation.After decision tree classification,the total accuracy of maize extraction combined with mixed cell decomposition was improved to 95.98%,and the maize classification accuracy was further improved.