Work-in-Progress: Towards detection and syntactical analysis in UML class diagrams for software engineering education

Learning how to create UML class diagrams from requirements specifications in textual format is one of the fundamental competences of students in Information Technologies. However, students seem to struggle creating those with all the requested elements. This process is not only challenging for students, but for teachers as well. Teachers, who correct the students solutions, have to take a deeper look at each created diagram to verify whether it is correct or not. Created diagrams can differ from each other or a given solution, but can nevertheless be correct. Also we observed, that the manner of creating class diagrams differ from student to student. Some create them using tools like the Enterprise Architect, others draw them by hand. To support students in the progress of learning how to create UML class diagrams and support teachers, we began realizing a prototype which can visually detect UML class diagram elements and compare them to a given solution. Visual recognition enables universal support no matter how a student created the diagram. Deep learning technologies have lead to major achievements in visual image processing in the recent years. Therefore this paper investigates that they are of use in UML element detection and syntactical analysis.