Abstract: This study aims to identify tobacco virus disease by integrating UAV-based multispectral remote sensing technology with an improved BP neural network (BPNN). Multispectral images of healthy and diseased tobacco plants with varying degrees of virus infection were captured using the DJI P4M UAV. A total of 19 vegetation indices were calculated to construct feature sets for correlation analysis. K-Nearest Neighbors (KNN), Random Forest (RF), Support Vector Machine (SVM), traditional BPNN, and improved BPNN were used to perform comparative tests on binary and ternary classification samples. The improved BPNN, with optimizations in network structure, imbalance data handling, activation function replacing, and optimizer enhancement, achieved 89% accuracy and an F1 score of 0.88 for binary classification, and 79% accuracy with an F1 score of 0.76 for ternary classification-both outperforming traditional algorithms. These results indicate that UAV multispectral data combined with an improved BPNN holds application potential for the detection of tobacco virus disease, providing technical support for early warning and prevention of agricultural diseases.