首页 > 分享 > 基于改进YOLOv8的轻量化荷叶病虫害检测模型

基于改进YOLOv8的轻量化荷叶病虫害检测模型

花匠小妙招
2024-09-15 11:13

Abstract:

Lotus leaf diseases and pests such as rhizome rot, leaf spot, virus disease, and Spodoptera litura seriously affect the yield and quality of lotus seeds. Implementing lotus leaf disease and pest control is an effective way to solve this problem, and detecting diseased leaves is an important measure to prevent and control lotus leaf diseases and pests. Currently, the detection of lotus leaf diseases and pests is mainly based on manual experience, which is subjective and inefficient. If there is a lack of professional knowledge, it is easy to cause missed detection or false detection. Therefore, the study of automatic detection technology of lotus leaf diseases and pests that adapt to the actual environment of lotus fields is of great significance to improving the planting quality of lotus seeds, reducing economic losses, and promoting the development of the lotus seed industry. With the goal of improving the detection accuracy of lotus leaf diseases and pests, reducing the calculation scale of the model, and improving the deployability, this study proposed a lightweight lotus leaf disease and pest detection model based on improved YOLOv8. At the same time, a lotus leaf disease and pest dataset considering different environmental conditions was established. First, the convolution module (Conv) in the YOLOv8 neck network is replaced with GSConv, and the C2f module is replaced with VoV-GSCSP to form a slim-neck architecture, which reduces the computational complexity of the model while maintaining high recognition accuracy. At the same time, the C2f_EMA module, which integrates the EMA efficient multi-scale attention mechanism, is used to replace the C2f module in the backbone network to improve the model's ability to extract features of lotus leaf pests and diseases in complex environments. The experimental results show that the improved YOLOv8 lotus leaf pest and disease detection model can effectively detect lotus leaf pests and diseases, and the mean average precision (mAP) achieved is 89.3%, which is 1.6% higher than the baseline model; the number of parameters of the model is reduced by 0.2M compared with the baseline model, and the model size is only 5.6 MB. Several mainstream one-stage target detection models: Faster R-CNN, SSD, YOLOv3, YOLOv4, YOLOv5, YOLOv7, YOLOv8 and YOLOv9 were selected for performance comparison experiments. The results show that compared with other mainstream detection models, the improved YOLOv8 model has significant advantages in detection accuracy, number of parameters, and model size. Finally, the improved YOLOv8 model was deployed on the Jetson Xavier NX and Raspberry Pi 4B edge computing devices, and the detection frame rates achieved by the model were 27 and 0.7 frames/s, respectively. Compared with the YOLOv5 model, the frame rates were increased by 8.9 and 0.3 frames/s, respectively. It can be seen that in terms of actual deployment performance, the established improved YOLOv8 model performs better than the classic YOLOv5 model, showing a good prospect for mobile terminal deployment. The proposed model realizes the accurate identification of lotus leaf diseases and pests and can provide support for the automatic prevention and control of lotus leaf diseases and pests.