Neural Network Structure with Alternating Input Training Sets for Recognition of Marble Surfaces

摘要

The automated recognition of marble slab surface textures is an important task in the contemporary marble tiles production. The simplicity of the applied methods corresponds with fast processing, which is important for real-time applications. In this research a supervised learning of a multi-layered neural network is proposed and tested. Aiming high recognition accuracy, combined with simple preprocessing, the neural network is trained with different alternating input training sets including combination of high correlated and de-correlated input data. The obtained good results in the recognition stage are represented and discussed, further research is proposed.