An approach for classifying ceramic tile defects based on a two-dimensional Genetic CNN algorithm

Ceramic tile industry is facing significant challenges in the industrial revolution 4.0. Most of the ceramic tile factories are under backward technologies, especially in the developing countries. Due to their complicated process and current technologies, there are many surface defects occurring on the final products. Classifying and grading the product still relies on humans which has caused many problems. Hence, developing an optimal model to deal with this surface defect detection and classification automatically is necessary for these companies. This study aims to propose a two-dimensional genetic algorithm-based convolutional neural network (2DG-CNN) which can automatically generate an optimal convolutional neural network (CNN) for detecting and classifying the defect products. In particular, a general CNN structure is firstly determined including the number of convolution layers, pooling layers, and fully connected layers. A two-dimensional chromosome is designed to represent a CNN model efficiently. In addition, a novel matrix crossover is developed to create more diversified offspring. A database of ceramic tile surface images is constructed to validate the proposed approach. The 2DG-CNN was compared with the other well-known algorithms. The results have shown the efficiency of the proposed approach.