A method for classifying tube structures based on shape descriptors and a random forest classifier

Machine-vision-based tube measurement is characterized by its accuracy, level of automation, noncontact nature and reliability. However, it cannot classify tube structures automatically. Commercial systems and previous algorithms cannot measure branch tubes due to difficulties of classifying tube structures. Therefore, this paper proposes a method for classifying tube structures. Multiple shape descriptors are used to extract tube structure features. Furthermore, RF classifier is used to distinguish among tube structures after tube features extraction. For efficient and accurate classification, the relative importance of each feature is calculated. Compared to results of ResNet-18 training on tube structures dataset, the precision of proposed method achieves 94% while the other is only 88%; experiments shows good performance on Recall and F-score. We developed a software to verify the method on the basis of the multi-view vision system built by our group, which can rapidly and automatically classify numerous complex tube structures used in engineering field.