Classification of high-similarity objects based machine learning

The classification of objects using computer vision technologies is useful in a variety of situations. In recent years, the deep learning approach to object classification has grown in popularity. The visual similarity of the objects to be classified, the insufficient number of images in the data set, the presence of different derivatives of the objects in the same class, and the high similarity between the objects in different classes all make classification difficult. The goal of this thesis is to use deep learning-based approaches to classify object groups that have high visual similarity to each other. Many deep learning-based approaches for categorizing various types and numbers of data have been proposed within the scope of the thesis. The content of our data sets created for use in the thesis; industrial screw, bolt and nut images, pantograph images and textile machine spare parts images. Assistance has been received from NİT Örme Textile Ind.Trade.Co.Ltd. in obtaining images of textile machine spare parts to be used in the thesis. Furthermore, data sets containing similar object groups have been prepared for use in studies within the scope of the thesis. The proposed methods have been thoroughly compared to the literature, and the results obtained revealed that the methods we proposed are quite successful and effective.