Cigarette Perforation Point Cloud Segmentation and Hole Depth Calculation Based on the Improved PointNet++ Network and DMCP Algorithm

Parallel perforation around the cigarette splicing paper has been used for sensory quality adjustment and dilution of cigarette smoke. Particularly, different sizes and depths of cigarette perforation can significantly affect the tar content in mainstream smoke. Therefore, the accurate detection of cigarette perforation parameters is crucial for ensuring product quality and reducing the damage to human health before cigarette processing completion. However, the tiny size and different depths and morphologies of cigarette perforations (e.g., tiny cigarette perforations are extremely similar to interference points) pose a great challenge to the identification of 3-D information on cigarette perforations and the subsequent task of perforation depth measurement. In view of that, this study focuses on two main challenges: identifying and locating tiny cigarette perforations of different brands and calculating each hole depth of a cigarette. A 3-D point cloud segmentation model for cigarette perforations based on an improved PointNet++ network is proposed. The contextual features downsampled from the original network SetAbstraction1 are merged, and the focal loss function is introduced. Furthermore, the K-nearest neighbors with an attention mechanism are used for group sampling. An accurate depth measurement algorithm (DMCP) for a cigarette-perforated point cloud is proposed for the hole depth calculation. The experimental results show that the final segmentation accuracy and recall of the proposed model are 91.7% and 94.7%, respectively, and the average absolute error of the hole DMCP for four brands of cigarettes is 3.59%. Thus, a high segmentation accuracy and hole depth measurement precision are achieved. This study introduces an innovative way for performing 3-D information identification and detection of cigarette perforation and accurate measurement of perforation depth. Finally, the results presented in this study can provide a useful reference for the implementation of other microhole identification segmentation and accurate measurement tasks in related fields.