Research on Prediction Models for Nitrogen Content in Ramie Canopy Leaves Using UAV-Based Multispectral Remote Sensing and Machine Learning
Nitrogen is one of the most significant nutrients affecting the growth and development of ramie.Accurately and timely mastering the nitrogen nutrition in ramie is crucial for improving management efficiency,precisely controlling growth dynamics,and implementing scientific management practices.This study conduc-ted the field trials with four nitrogen fertilizer levels and two top-dressing periods which were no nitrogen(N0),273 kg/hm2 of pure nitrogen(N1),332 kg/hm2 of pure nitrogen(N2),390 kg/hm2 of pure nitrogen(N3),as well as top-dressing during the row closure period(a)and vigorous growth period(b).Spectral im-ages were collected using DJI Phantom 4 drones at four growth stages(seedling stage,line-closing stage,full growth stage and mature stage)of three seasons of ramie.Combined with measured nitrogen data,four ma-chine learning algorithms including support vector machine(SVM),random forest regression(RFR),back propagation neural network(BPNN),convolutional neural network(CNN)were used to establish nitrogen nutrition estimation models for each growth stage.The results showed that the nitrogen content estimation mod-els based on the four machine learning algorithms all had some predictive capacity,but there were obvious differences in prediction accuracy.Among them,the CNN model had the highest prediction accuracy at seed-ling stage(R2 of 0.761,RMSE of 0.743 and MAE of 0.548 on the validation set),which could accurately predicte the nitrogen nutrition status of ramie,although its generalization ability at mature stage needed to be optimized.The RFR model demonstrated stable performance in nitrogen prediction across growth stages(R2 of 0.607 to 0.755,RMSE of 0.819 to 1.156 and MAE of 0.680 to 0.930 on the validation set),making it suitable as a general model for long-term monitoring.In conclusion,the machine learning model based on UAV multi-spectral remote sensing could achieve better prediction of nitrogen in ramie crown leaves,which could provide theoretical support for rapid monitoring of ramie growth and precise fertilization management.
NETL
NSTL
万方数据
