结合变异系数法和机器学习模型的棉花长势监测

Cotton growth monitoring combined with coefficient of variation method and machine learning model

为了更加准确地获取棉花关键物候期的长势信息,本文首先通过棉花制图指数提取棉花种植区域;然后利用变异系数法将反映棉花长势的株高、SPAD值、叶片湿重、叶片干重与叶面积5种指标构建为一个综合长势指标,即棉花长势指数(FBCGI);最后选取最优特征变量,结合随机森林模型构建棉花长势反演模型。结果表明:①棉花总体分类精度达到81.65%;②与5种单一长势指标相比,构建的FBCGI与植被指数的相关性更高;③基于最优特征变量和随机森林模型构建的棉花长势监测模型,在建模集和验证集中的R 2和RMSE分别为0.74、0.07和0.51、0.10。研究结果可为棉花长势监测提供重要参考。

In order to obtain the growth information of the key phenological period of cotton more accurately,the cotton planting area is extracted through the cotton mapping index.Secondly,five indexes,including plant height,SPAD value,leaf wet weight,leaf dry weight and leaf area,reflecting cotton growth,are constructed into a comprehensive growth Index,namely Flowering and boll cotton growth index(FBCGI),using the coefficient of variation method.Finally,the optimal characteristic variables are selected and the inverse model of cotton growth is constructed by combining with random forest model.The results showed that:①The overall classification accuracy of cotton reached 81.65%.②Compared with the five single growth indicators,the constructed FBCGI had a higher correlation with vegetation index.③The R 2 and RMSE of the cotton growth monitoring model based on the optimal characteristic variables and random forest model in the modeling set and validation set are 0.74,0.07 and 0.51,0.10,respectively.The results can provide important reference for cotton growth monitoring.