DEEP LEARNING-BASED CLASSIFICATION OF VERY SIMILAR FASTENERS

5TH INTERNATIONAL BLACK SEA MODERN SCIENTIFIC RESEARCH CONGRESS, 8 - 10 Kasım 2023, ss.879-894

Deep learning has recently developed in object classification and object recognition in many

areas. Classification of objects that are quite like each other in object classification was

carried out in this study. The cases where the similarity between the classes is low and the

difference within the class is high are the situations where object classification is very

difficult. For this reason, the classification of the examples with such problems was not made

using too many object types at the stage of classification. In this study, in which objects with

more similarity between classes are classified, unlike other object classification studies in the

literature, many similar object classes are classified with deep neural networks. In this study,

6 types of screws, 7 types of bolts, and 5 types of nuts are classified. The training accuracy

rate of the model with the developed neural network is 99.31% and the validation accuracy

rate is 96.02%. In this study, the performance of the developed model has been demonstrated

with experimental results.