Abstract: Lotus leaf diseases and pests have seriously threatened the yield and quality of lotus seeds, such as rhizome rot, leaf spot, virus disease, and Spodoptera litura. The detection of diseased leaves has been the most important measure to prevent and control the lotus leaf diseases and pests. However, manual detection cannot fully meet the large-scale production in recent years, due to the subjective and inefficient experience. It is still lacking in professional knowledge, easily leading to missed or false detection. Therefore, automatic detection can be expected to improve the planting quality in the actual environment of lotus fields. In this study, a lightweight detection model was proposed for the lotus leaf disease and pest using improved YOLOv8. The detection accuracy was improved to reduce the calculation scale for the better deployability of the model. At the same time, a new dataset of lotus leaf disease and pests was established to consider the different environmental conditions. Firstly, the convolution module (Conv) in the YOLOv8 neck network was replaced with GSConv. The C2f module was replaced with the VoV-GSCSP to form a slim-neck architecture, in order to reduce the computational complexity of the model for the high recognition accuracy. The C2f_EMA module was integrated with the EMA efficient multi-scale attention mechanism. The C2f module was then replaced in the backbone network, in order to extract the features of lotus leaf pests and diseases in the complex environments. The experimental results show that the improved YOLOv8 detection model effectively detected the lotus leaf pests and diseases. The mean average precision (mAP) was achieved at 89.3%, which was 1.6 percentages higher than the baseline model; The number of parameters of the model was reduced by 0.2M, compared with the baseline model. The model size was only 5.6 MB. A comparison was performed on several mainstream one-stage target detection models: Faster R-CNN, SSD, YOLOv3, YOLOv4, YOLOv5, YOLOv7, YOLOv8, and YOLOv9. The results show that the improved YOLOv8 model shared significant advantages in the detection accuracy, number of parameters, and model size, compared with the rest mainstream models. Finally, the improved YOLOv8 model was deployed on the Jetson Xavier NX and Raspberry Pi 4B edge computing devices, where the frame rates were 27 and 0.7 frames/s, respectively. Compared with the YOLOv5 model, the frame rates increased by 8.9 and 0.3 frames/s, respectively. In terms of actual deployment performance, the improved YOLOv8 model performed better than the classic YOLOv5 model, indicating a better prospect for the mobile terminal deployment. The accurate identification can provide support to the automatic prevention and control of lotus leaf diseases and pests.