Detection of Cigar Defect Based on the Improved YOLOv5 Algorithm

To achieve the automatic detection of blue spots, plaques, and desquamation defects of manufactured cigars, an improved YOLOv5 model is proposed for the high-precision detection of manufactured cigar defects in the production process. The EPSA attention mechanism is added to the YOLOv5 model to make the network focused on the defect location. The PAN structure is replaced by the BiFPN structure in the Neck part of the model, which enhances the multi-scale fusion of features. Also, with the introduction of BiFPN in YOLOv5, the performances of the network with different attention mechanisms are compared. The experimental results show that the YOLOv5BE improves by 2.69 % at the mAP@0.5 compared with YOLOv5, reaching 94.15%. Therefore, the improved YOLOv5 model can effectively detect blue spots, disease spots, and desquamation defects of manufactured cigars, and provide technical support for the intelligent detection of manufactured cigars.