Remote Sensing Estimation of Winter Wheat Residue Coverage Based on Sentinel-2 Images and Machine Learning Algorithm

ZHU Qilei,LIANG Dong,XU Xingang,AN Xiaofei,CHEN Liping,YANG Guijun,HUANG Linsheng,XU Sizhe
DOI: https://doi.org/10.7606/j.issn.1009-1041.2023.04.15
2023-01-01
Abstract:In order to explore the estimation method for crop residue coverage(CRC) in a large scale, winter wheat residue was taken as the research object.Based on the spectral index, band and texture features of Sentinel-2 remote sensing satellite image and their different feature combinations, the optimal model of wheat CRC estimation was studied by using the sensitive feature extraction method of grey correlation-random forest(GRA-RF),combined with a variety of machine learning algorithms, such as Gaussian process regression(GPR),LASSO,ridge regression(RR),and partial least squares regression(PLSR).The results showed that the machine learning model based on GRA-RF feature selection significantly improved the estimation accuracy of wheat CRC,and the LASSO algorithm had the best overall estimation effect on wheat CRC,and showed different results for different spectral feature combinations.The CRC remote sensing estimation model based on the combination feature set composed of spectral index, band and texture information had the best accuracy(r~2=0.65,RMSE= 9.25%).The CRC accuracy estimated by the combination characteristics of band and texture was the second(r~2=0.63,RMSE=9.31%).Only using a single spectral index, band or texture feature to estimate the CRC of winter wheat, the model accuracy was inferior to the results of the combined features.It indicates that the GRA-RF integrated method can effectively select spectral features related to CRC,and the combined features had the better capability of estimating CRC.
What problem does this paper attempt to address?