| Abstract [eng] |
This paper presents a tool for digitising UML use case diagrams from a sketch. An analysis of the UML language and its application in systems design has shown that first sketches of diagrams are often drawn by hand without the CASE tool. The solution analysis showed that there is no comprehensive tool to digitise use case diagram sketch and to allow the modelling of the initial sketch to be extended in the CASE tool. To implement the tool, a new set of use case diagrams was created from 346 annotated sketches and machine learning capabilities for image and text recognition were analysed. For the diagram elements recognition model, the Faster R-CNN ResNet152 V1 800x1333 neural network architecture in the TensorFlow 2.5 ecosystem was chosen, and for the text recognition, an external service of the Google Cloud Vision API was used. The designed tool is implemented and composed of 5 main components with Python programming language: a diagram symbol recognition component, a keypoint detection component, an OCR component, a symbol linking component and a diagram to XMI file conversion component. Finally, an experiment on diagram symbol recognition and diagram digitisation was carried out with 32 new diagrams. Diagram symbol recognition experiment showed a recall of 0.98, precision of 0.92 and F1-score of 0.94. Diagram digitisation experiment showed that diagram elements are digitised with a recall of 0.96, precision of 0.99 and F1 score of 0.95. Symbols connection was achieved with a recall of 0.89, a precision of 0.99 and F1- score of 0.89. |
