Detecting Design Patterns in UML Class Diagram Images using Deep Learning

Detecting Design Patterns in UML Class Diagram Images using Deep Learning

연구 분야: Analysis

학회: 2022 IEEE/ACIS 23rd International Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD)

Detecting software design pattern is an important part of software reverse engineering because design patterns can provide the most intuitive design idea of software products, which can be useful for maintenance engineers. In past studies, a lot of approaches have been proposed to detect design patterns, and the machine learning-based approach is a new trend in recent years. In this paper, we propose a preliminary idea of a deep learning-based approach to detect design patterns from UML class diagrams of software products, which can be used in some cases that traditional approaches may not work. We propose an overall process, which is divided into preparation phase and application phase. In preparation phase, we train a deep learning-based classifier to do the image classification task. In application phase, users may input the UML class diagram of a micro-architecture into the model and get the pattern it belongs to. We conduct a preliminary experiment to show the effectiveness of our approach, we train a Convolutional Neural Network (CNN) as the classifier and test it on our image dataset, which is constructed with UML images we collected from the Internet. We also use Gradient-weighted Class Activation Mapping (Grad-CAM) to do the visualization and use it to explain why our approach works. Lastly, we analyze the potential advantages and disadvantages of our approach.