Timely and accurate detection of leaf rust helps scientists to make better control deci-sions for poplar. Use of Machine Learning (ML) methods for analysis of Unmanned Aerial Vehi-cle (UAV)-based multispectral data can improve the detection accuracy. In this paper, using poplar trees as the research object, a six-rotor UAV with a multispectral camera was used to ac-quire 5-band spectral images, analyze the canopy images of asymptomatic and diseased poplar trees and calculate 25 common vegetation indices. Firstly, the Multilayer Perceptron (MLP) and Random Forest (RF) algorithms were used to select the first 5 classification vegetation indices of asymptomatic and low, asymptomatic and moderate, asymptomatic and severe infection, multi-class, then the detection models based on Multilayer Perceptron (MLP), Decision Trees (DTs), Naive Bayes (NB), Support Vector Machines (SVM), Random Forest (RF) algorithms were estab-lished. Based on the analysis of the detection accuracy and the reliability of models, for the bina-ry classification, the MLP model had the highest overall accuracy (OA=95%) and the highest value of Area Under the Curve (AUC=1.0), for the multi-class classification, the Naive Bayes (NB) model based on the RF selected binary classification VIs set had highest overall accuracy (OA=100%) and the highest F1-score (F1-score=1.0). The optimal classification vegetation indices for different disease levels were analyzed. This paper demonstrates the feasibility of rapid and accurate detection of poplar leaf rust by UAV multispectral imaging technology, which can pro-vide timely and accurate detection of different disease levels, improve the efficiency of field management and disease control.