The work aims to propose a paper cup defect detection method based on improved YOLOv5s model aiming at the problem that paper cup defects in small size and with insignificant features are easy to be missed and misdetected in the detection process. The CBAM attention mechanism module was introduced in the Backbone part of the original model to improve the feature extraction ability of the model. A YOLO detection head was added, and the three-scale detection was changed to four-scale detection to improve the detection ability of the model for small targets and targets with insignificant features. In the Neck part, the weighted bidirectional feature pyramid network BiFPN was used to partially improve the PANet in the original model to strengthen the feature fusion ability of the model. The results indicated that the precision of the improved model YOLOv5s-CXO was 89.1%, the recall was 90.4%, and the average precision mean was 89.5%. Compared with the original model, the precision was increased by 1.5%, the recall was increased by 1.3%, and the average precision mean was increased by 1.2%. The proposed improved method effectively improves the detection ability of the model, and significantly improves the detection effect of paper cup defects in small size and with insignificant features.