马铃薯产量的精准预测对于保障粮食安全具有重要作用.为实现快速精准获取马铃薯产量信息,研究利用无人机搭载可见光相机和多光谱相机,分别于马铃薯现蕾期、初花期、盛花期和终花期进行无人机RGB和多光谱影像的采集,并于收获期测定马铃薯产量.基于RGB和多光谱影像特征,对马铃薯产量与RGB指数和植被指数分别进行相关性分析,结合样本数量,筛选出相关性最高的前五个光谱指数作为模型输入变量,采用多元线性回归(Multiple linear regression,MLR)和随机森林(Random forest,RF)2种方法构建不同生育期和全生育期的估算模型,并进行对比,筛选出马铃薯产量估算的最优模型.通过RGB指数与植被指数结合可以提高马铃薯产量的估算效果,并且多元线性回归模型的反演效果要优于随机森林模型.最终产量的估算效果从高到低分别是盛花期>全生育期>终花期>初花期>现蕾期,马铃薯产量最优估算模型为盛花期以RGB指数与植被指数结合作为模型输入变量的MLR模型.测试集R2和RMSE分别为0.77、0.64 kg/m2;验证集R2和RMSE分别为0.68、0.56 kg/m2.研究结果可为精准农业定量化研究提供技术支持.
Potato Yield Estimation Based on UAV Remote Sensing Images
The accurate prediction of potato yield plays an important role in ensuring food security.In order to obtain potato yield information quickly and accurately,an unmanned aerial vehicle(UAV)with a visible light camera and a multi-spectral camera were used to collect RGB and multi-spectral images of potatoes at the bud flower,early flowering,full flowering and late flowering stages,and potato yields were recorded at harvest time.Based on the RGB and multi-spectral image features,correlation analysis was conducted of potato yield with RGB index and vegetation index,respectively,and the top five spectral indices with the highest correlation were selected as model input variables by combining the sample size.Finally,multiple linear regression(MLR)and random forest(RF)were used to construct estimation models for different growth stages and the whole growth period,and were compared to select the optimal model for potato yield estimation.The combination of RGB index and vegetation index improved the estimation effect of potato yield,and the inverse effect of the multiple linear regression model was better than that of the random forest model.The estimated effect of final yield from high to low was:full flowering>whole growth period>late flowering>early flowering>bud flower.The optimal model for estimating potato yield was the MLR model with the combination of RGB index and vegetation index as model input variables at full flowering.The test set R2 and RMSE were 0.77 and 0.64 kg/m2,respectively,whereas the validation set R2 and RMSE were 0.68 and 0.56 kg/m2,respectively.The results would provide technical support for accurate agriculture quantitative research.
potatoUAV remote sensingyieldRGB indexvegetation index
NETL
NSTL
万方数据
维普
