Aiming at the demand of cigarette enterprises for the inspection of the appearance quality of cigarette packets in the production process, a method for detecting surface defects of cigarette packs based on machine vision was designed. The EML-YOLO used in this paper is an industrial surface defect detection algorithm based on improved YOLOv8. It improves the feature extraction capability of the model by designing an efficient large convolution module (ELK) and provides multi-scale feature representation while preserving spatial information. The modules of image acquisition, image preprocessing and defect classification are built; the self-fitting brightness adjustment algorithm is proposed to complete the pixel value statistics and obtain a clear defect feature image. The experimental results show that the detection accuracy of the EML-YOLO algorithm is 98.84%, which has good application potential in industrial defect detection scenarios, and can accurately and efficiently realize the classification and recognition of defect information on the surface of cigarette packets.