Atıkların Transfer Öğrenme Yöntemi ile Sınıflandırılmasında Veri Artırımının Etkisi

Effective classification of waste is one of the precautions to be taken against global warming. Carrying out this process with more practical methods instead of traditional methods, namely through artificial intelligence, will contribute to the process. In this study, classification process was conducted via transfer learning while using RealWaste dataset which contains 9 waste classes. Four different scenarios were considered to test the classification performance of the deep learning networks. Besides, the effect of data augmentation on the classification process was investigated. AlexNet, GoogLeNet, MobileNet, ResNet versions, VGG versions and Inception_v3 networks were preferred in this study. While the last 2 layers of the networks were trained, the other layers were frozen. Geometric transformations were preferred for data augmentation. The highest performance acquired without doing any additional process on the data set was 91.57% with ResNet50 network. When data augmentation was made, 99.912% classification performance was achieved with the Inception_v3 network. It was observed that data augmentation affected the classification performance in a positive way.