Classification of UML Diagrams from Images in Software Engineering Using Deep Learning

Software design is a fundamental phase in software engineering, where Unified Modeling Language (UML) diagrams play a critical role in representing system structures and behavior for software engineers. However, UML diagrams are often available only as images, which limits their automated processing and integration into tools such as search engines, software documentation systems, and accessibility solutions for visually impaired users. In this study, we propose a deep learning approach for automatically classifying UML diagrams from images. A lightweight convolutional neural network (CNN) model is designed with a minimal number of layers and reduced trainable parameters, providing an efficient solution without requiring data augmentation or transfer learning. The model is trained to accurately classify six UML diagram types: Activity, Component, Class, Sequence, Deployment, and Use-Case. Experimental evaluations on a UML image dataset demonstrated a high classification accuracy of 94.61%, highlighting its strong predictive capability despite its compact design.