A Tobacco Pest Detection System Based on the Improved YOLO v5 Model

During the storage process, tobacco leaves are vulnerable to infestation by Lasioderma serricorne. To achieve precise detection of this pest within stored tobacco leaves and provide real-time updates on pest activity, we propose a detection system based on an improved YOLOv5 model. Initially, an open-source hardware system is established to capture images of the pests, which are then transmitted to a computer terminal via a Wi-Fi network. Random data augmentation techniques, including translation, rotation, and the addition of Gaussian noise, are applied to expand the training set of Lasioderma serricorne images. Subsequently, the BiFormer attention mechanism is integrated to enhance the model's accuracy in detecting small targets such as Lasioderma serricorne. By combining the improved YOLOv5 model with hardware facilities and the detection system, a comprehensive device is formed. The results demonstrate that the improved YOLOv5 network model achieves an accuracy rate of 93.26% and an average detection precision of 92.40% for Lasioderma serricorne. Compared to models such as YOLO v3, Retinanet, Faster R-CNN, the original YOLOv5, and SSD, the improved YOLOv5 model exhibits superior detection performance, with efficiency improvements of 2.76% and 2.46% over the existing YOLOv5 algorithm, respectively. The integration of the improved YOLOv5 model with hardware and the detection system into a comprehensive device fully harnesses the latest advancements in modern computer vision and artificial intelligence technologies. This device not only offers efficient object detection capabilities but also adapts well to various environments and scenarios, opening up a wide range of application possibilities. In this study, the device accurately identifies and counts Lasioderma serricorne, providing technical support for early warning and pest control in tobacco storage systems.