Organising unstructured data using Double Net Self-Organising Map (DNSOM) model

Surface reconstruction is the process of representing surfaces using the data obtained from scanning devices. When the data obtained is unstructured during data collection, it can cause problems in presenting the surface due to the lack of connectivity information for the data. It shows that data collection is a crucial task in surface reconstruction. Previous works have proved that Self-Organising Map (SOM) models can be used to organise unstructured data. However, the 2-D SOM model will generate a surface with holes for the closed surfaces. Besides, the 3-D SOM model will generate an incorrect surface due to the connectivity of the internal neurons. In addition, the Cube Kohonen SOM (CKSOM) model is limited to the same grid size. Hence, a SOM model known as Double Net SOM (DNSOM) is proposed via the merging of two 2-D SOMs to overcome the issues. Three data sets (cube, sphere, and talus bone) with different grid sizes are applied to test the models. When the grid size of all the models increases, the surface becomes smoother. The DNSOM contains a lower quantisation error than the 2-D SOM and the lowest topographic error among the other SOM models. It performs faster than the 3-D SOM and CKSOM. It presents the correct closed surface with a smaller number of neurons and different grid sizes. Microsoft Visual Studio 2022 with the C++ programming language is used to develop the models, while GNUPlot is used to visualise the results. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2025.