Cigarette packaging is critical to enhancing consumer experience and brand image, but the complexity of the production process and the instability of the equipment can easily lead to defects such as empty packs, abnormal drawing lines, and abnormal printing, which affect quality and brand image. However, the traditional manual inspection has problems such as information lag, incomplete detection, etc., which makes it difficult to detect defects in time, thus affecting production efficiency and quality control. Therefore, this paper proposes to develop an efficient classification method for cigarette packaging defect detection by combining deep learning and traditional computer vision techniques. The method performs multi-label classification of cigarette packaging image dataset by constructing a deep learning network model based on ResNet and YOLOv8, and evaluates its classification effect. Based on the superior classification results of YOLOv8, for images with poor classification results, the detection results are further optimized by traditional CV methods such as color filtering and Canny edge detection. Through the two-phase detection process, the method is able to detect all defects (18 types) that occur in the production process of cigarette packages after experimentation and research, and can complete efficient and accurate defect detection in a short period of time, which greatly improves the efficiency of quality control, and not only provides a solution for automated inspection of cigarette packages, but also provides a reference value for defect inspection in other industries.