Extraction of spring farmland plastic mulching in Qianguo County based on Google Earth Engine
Based on the Google Earth Engine(GEE)cloud platform,this paper comprehensively considers the band reflectivity and spectral index characteristics of optical images and the polarization and texture character-istics of radar images,and constructs three feature input combinations:Using only optical features,only radar features,and a combination of optical and radar features.After determining the best input features based on ac-curacy,this paper combines five machine learning methods,namely classification and regression tree,support vector machine,minimum distance,gradient boosting decision tree,and random forest,to establish a plastic mulching extraction model.The performance of different methods is evaluated based on the accuracy of the.results,and the final plastic mulching area is extracted based on the optimization model.The results show that:1)The combined optical and radar image characteristics have the highest accuracy in extracting plastic mulch-ing coverage,and the optimal input features are band reflectivity features+spectral index features+polariza-tion features+texture features;2)The model established using the random forest method has the highest accur-acy.The overall accuracy of study area I reached 95.84％,the Kappa coefficient was 0.95,and the ground ob-ject misclassification rate was 1.2％,which was significantly better than the other four methods(the ground ob-ject misclassification rate was 0.8％,7.3％,38.0％and 0.3％lower than that of classification and regression tree,support vector machine,minimum distance and gradient boosting decision tree method),and the verifica-tion accuracy of study area Ⅱ reached 87.84％,proving that the model can obtain more accurate results in plastic mulching extraction;3)Using the method in this paper,the area of plastic mulching farmland in study area Ⅰ in 2022 is 1302.48 km2,and the estimated amount of film used is about 7585.62 tons.This article com-prehensively considers the characteristics of optical and radar images in ground object recognition to establish a model,which can accurately and efficiently identify farmland mulching and grasp the area of mulch,which is of great significance for environmental management and prevention.
plastic mulchingGoogle Earth Enginefeature extractionrandom forestsupport vector ma-chineclassification and regression treeminimum distancegradient boosting decision tree
邓韵谣、李晓洁、任建华
展开 >
哈尔滨师范大学地理科学学院,黑龙江哈尔滨 150025
中国科学院东北地理与农业生态研究所,吉林长春 130102
覆膜 Google Earth Engine 特征提取 随机森林 支持向量机 分类与回归树 最小距离 梯度提升树
万方数据
维普
