Natural or Synthetic Dye Classification in Textiles using Residual Neural Network and Transfer Learning

Natural and synthetic dyes helped to achieve a wide range of hues in clothing, carpets, blankets, tapestries, shrouds, and paintings on materials, among other things. In the Philippines, there are still no existing studies that propose different methods in dye classification. The only way available is to classify dyes according to the type of chromophores present in their structures through chemical analysis. However, chemical classification can be destructive to the historical object and takes time for the analysis results. The researcher created a cost-effective capturing device using a Raspberry Pi 4 and a Raspberry Pi NoIR camera V2 in this study, which gathered all images of natural-dyed and synthetic-dyed textiles. The researcher used Residual Neural Networks and transfer learning to classify dyes in textiles as natural or synthetic. The researcher built a Resnet model as a feature extractor. The extracted train image features to train the SVM classifier. The chosen polynomial kernel had the best accuracy of 91.67 percent from the 12 sample test images, compared to 75 percent for the linear kernel and 66.67 percent for the radial basis function kernel.