This project focuses on the classification of surface defects using the ResNet18 architecture as the backbone. The goal is to accurately identify various types of surface defects in materials.
ResNet18 Backbone: Utilizes the powerful ResNet18 deep learning architecture for feature extraction and classification. Seven-Class Classification: Capable of identifying seven distinct classes of surface defects. Confusion Matrix Results: Evaluation of the model's performance is depicted through a detailed confusion matrix, highlighting its accuracy across different defect types.
The performance of the classification network on the training set using confusion matrix:
Fig. 1: Results on the training set.
The performance of the classification network on the test set using confusion matrix:
Fig. 2: Results on the test set.