Classification of fabric images by deep learning

Millions of data are produced every day from human and technology interaction. Processing these data and obtaining certain results is important for the existence of information. Inferences made by humans are now made by artificial intelligence-based systems, which success rate is increasing day by day and the margin of error is decreasing. The use of deep learning has also become widespread due to the successes in many fields such as image, sound, language processing and biomedical. In this study, it is aimed to classify fabric images with convolutional neural networks (CNN), which is one of the multilayered neural network structures of deep learning. In the first step, while designing the models to be classified, hyper-parameters affecting success were taken into account and trained in 10 different models. In the second stage, the data were trained using two different approaches with the transfer learning method in seven pre-trained models and all results were compared. In total, the data were trained to 24 different models. Each model was evaluated according to precision, recall, F1 score and accuracy metrics, and the results were shared. In the designed models, the highest accuracy rate and the lowest accuracy rate were obtained as 96.26% and 28.24%, respectively. While the highest accuracy rate in the trainings made through transfer learning was obtained with the fine-tunning VGG-16 model 84.51%, the lowest accuracy rate was obtained from the InceptionV3 model, which only used feature extraction approach, with a rate of 76.53%.