In the process of tobacco planting, timely and accurate detection of pests and diseases is crucial. However, traditional pest and disease detection algorithms have problems such as poor detection accuracy, large number of model parameters, and difficult to be conveniently deployed on the mobile side when facing complex tobacco images. In order to effectively solve these problems, this paper proposes a tobacco leaf pest and disease detection algorithm based on YOLOv10n optimization and improvement. In the improvement process, the Swin Transformer is added into backbone, the GSConv convolution module is integrated into Neck, and this innovative combination significantly strengthens the model’s ability to capture subtle features, significantly improves the overall detection performance, and further enhances the model’s ability to extract the characteristic information of tobacco leaf pests and diseases. Tests were carried out based on the tobacco leaf pest and disease dataset in this paper. The results show that the mean average precision (mAP) of the improved algorithm is significantly increased from 63.4% to 79.4%, and the improved algorithm realizes a qualitative leap in detection accuracy, which greatly meets the actual needs of real-time and accurate detection of tobacco leaf pests and diseases, effectively reduces the leakage of pests and diseases and misdetection, and demonstrates a high practical value in practical applications.