Scholarlink

Unified Modeling Language (UML) diagrams are standard graphical tools to specify, visualize and understand the design and behavior of software systems. Accurate classification and identification of UML components are crucial for the quality and efficiency of software development processes. This work deals with the application of Convolutional Neural Networks (CNNs) in the classification of UML diagrams. The proposed approach focuses on deep learning techniques to accurately classify UML diagrams into different categories, including UML and non-UML classes, as well as identifying the components of UML diagrams, addressing the challenges of multi-labeling and multi-classification. Furthermore, transfer learning techniques are investigated to utilize pre-trained models and test their contribution to improving classification performance given the limited size of the dataset. The proposed approach is applied to a dataset of 4705 images of UML and non-UML diagrams. The results demonstrate the effectiveness of CNNs for multi-classification and multi-labeling of UML diagrams with an accuracy of 97.54%, outperforming the traditional models.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Multi-classification and Multi-labeling of UML Diagrams Using Transfer Learning and CNN

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Abderrahim Lakehal,
  • Adel Alti,
  • Boubakeur Annane,
  • Bensalah Imene,
  • Smaani Anissa Afaf,
  • mami meriem

摘要

Unified Modeling Language (UML) diagrams are standard graphical tools to specify, visualize and understand the design and behavior of software systems. Accurate classification and identification of UML components are crucial for the quality and efficiency of software development processes. This work deals with the application of Convolutional Neural Networks (CNNs) in the classification of UML diagrams. The proposed approach focuses on deep learning techniques to accurately classify UML diagrams into different categories, including UML and non-UML classes, as well as identifying the components of UML diagrams, addressing the challenges of multi-labeling and multi-classification. Furthermore, transfer learning techniques are investigated to utilize pre-trained models and test their contribution to improving classification performance given the limited size of the dataset. The proposed approach is applied to a dataset of 4705 images of UML and non-UML diagrams. The results demonstrate the effectiveness of CNNs for multi-classification and multi-labeling of UML diagrams with an accuracy of 97.54%, outperforming the traditional models.