Abstract: Aiming at the difficulty of obtaining high-throughput plant height data of tobacco plants, a method of plant height detection based on UAV images and improved Yolov8 was proposed. In this paper, a lightweight DSW-Yolov8n algorithm was proposed to obtain orthophoto image of tobacco plant in field by UAV oblique photography and extracted elevation information. The algorithm took Yolov8n as the baseline model, and replaced the trunk C2f convolutional module with DualConv lightweight convolutional module combining group convolution and heterogeneous convolution (HetConv) to reduce training parameters. A SV-neck constructed by Spatial depth transformation conv (SPD-Conv) and VoV-GSCSP was proposed to replace neck, which can integrate features of different levels more effectively and improve detection accuracy. Finally, WIOU(Wise-IOU) loss function was introduced to accelerate the convergence of the model, so as to detect the center of the plant in the ortho image and obtain the corresponding plant height. The experimental results show that compared with the original model, the parameters of the improved algorithm are reduced by 18.1% and the size of the model is reduced by 15.9%. The improved tobacco center recognition model mAP50 is 98.4% and mAP(50-95) is 63.1%, respectively, which are 2.1% and 1.6% higher than the original model. The slope of the fitting line between the estimated plant height and the measured value was 1.09, and the was 0.88, which showed strong correlation, and realized the high-throughput detection of the plant height of tobacco plants in the field.