基于Google Earth Engine的黄土高原覆膜农田遥感识别

Remote Sensing Recognition of Plastic-film-mulched Farmlands on Loess Plateau Based on Google Earth Engine

为了建立覆膜农田遥感识别技术体系,本研究选取甘肃省定西市安定区团结镇作为黄土高原地膜覆盖旱作农业代表性区域,基于Google Earth Engine云平台和Landsat-8反射率数据,采用特征重要性分析优选纹理特征,利用参数优化后的随机森林算法提取覆膜农田区域并选出最佳特征组合方案,最后通过对比随机森林、支持向量机、决策树和最小距离分类4种算法的分类结果来评价不同分类算法的性能。结果表明:优化关键参数后的随机森林算法能够显著提高遥感影像的分类精度;单一特征方案中,基于光谱特征的分类精度最高,且加入指数和纹理特征可提高总体识别精度;利用随机森林特征重要性分析选取的优选纹理特征分类性能优于全部纹理特征,基于“光谱+指数+优选纹理”特征方案的识别结果最佳,总体精度和Kappa系数达95.05%和0.94;与支持向量机、决策树和最小距离分类相比,随机森林优势明显,总体精度分别高3.10、7.74、50.78个百分点。本研究实现了对地形复杂地区覆膜农田空间分布较为精准的识别。

Plastic-film-mulching has made an outstanding contribution to agricultural production and food security in China,but also caused many serious environmental problems.It is very important to quickly and accurately obtain the spatial distribution information of plastic-mulched farmlands.In order to establish a framework for remote sensing recognition of plastic-film-mulched farmland,the Tuanjie Town of Dingxi City in Gansu Province was chosen as the research area,which was a typical dry farming agricultural area with heavy plastic film application on the Loess Plateau.Based on the Google Earth Engine,Landsat-8 reflectance data was used to analyze the importance of different features and select the optimal textural features.Then,the random forest(RF)algorithm with optimized parameters was used to extract the plastic-film-mulched farmland area and select the best feature combination.Finally,based on the best feature combination,the performance of RF algorithm was evaluated through comparison between the classification results based on the other algorithms of support vector machines(SVM),decision tree(DT),and minimum distance classifier(MDC),respectively.The results showed that the optimized parameters of RF algorithm could greatly improve the classification accuracy.Among the schemes based on single kind of features,the accuracy of scheme based on spectral features was the highest.The addition of index and textural features could also improve the overall identification accuracy to some extent.The performance of the selected optimal texture features was better than that of all texture features.The recognition result based on the combination of‘spectral+index+optimal textural features’was the best,whose overall accuracy and Kappa coefficient were 95.05%and 0.94,respectively.The overall accuracy of RF algorithm was 3.10 percentage points,7.74 percentage points and 50.78 percentage points higher than the algorithms of SVM,DT and MDC,respectively,which proved the RF algorithm had some obvious advantages in recognition of plastic-film-mulched farmlands.The research realized an accurate identification of plastic-film-mulched farmlands in areas with complex terrain features in China.The results can provide theory and technology supports for the studies related to spatial variations and sustainability production with plastic film mulching in the near future.