Traditional cloud pattern classification algorithm based on semi-supervision with Random Line Augment

The classification of traditional patterns is of great significance for their digital protection. Most studies focus on classifying patterns of different categories, while this study addresses the difficulty of classifying patterns of different types within the same category by selecting traditional cloud patterns (TCP) with complex structures and numerous types for classification. Due to the large number of label annotations required by deep learning algorithms relying on supervised learning, this paper proposes a traditional cloud pattern classification algorithm based on semi-supervised learning, which achieves high-precision classification with a small number of label annotations. Meanwhile, this paper proposes a novel data augmentation strategy called Random Line Augment (RLA) based on the line features of cloud patterns and edge detection algorithms. The algorithm also introduces WideResNet as the backbone network, which comprehensively captures local detail features in cloud pattern images by increasing the number of feature channels. The experimental results show that the algorithm has a significant effect on cloud pattern classification with obvious line features, with an accuracy of 97.41%, reaching a high level in pattern classification work.