PSI - Issue 70

Bagathi HarshaVardhan et al. / Procedia Structural Integrity 70 (2025) 447–452

451

Fig.4. Epoch Vs Accuracy for ResNet Model

Fig. 5. Confusion matrix for ResNet model

In VGGNet model the achieved accuracy is 67% indicating it is less efficient in handling the corrosion datasets (Fig.6) on, Recall and F1 score are evaluated as 0.6343,0.6333 and 0.6036 respectively. The epoch versus accuracy curves in both the models depict jagged pattern, which may a result of complex and imbalanced data from the images or overfitting of the data.The abrupt spikes in the accuracy curve may be due to data inconsistency.

Fig. 6. Epoch Vs Accuracy for VGGNet Model

Fig. 7. Confusion matrix for VGGNet model

3.1 Relevance of Confusion Matrix: The confusion matrix plays a crucial role in evaluating the classification performance of the deep learning models employed in this study. It provides a detailed summary of correct and incorrect predictions across each corrosion severity class — no corrosion, medium corrosion, and severe corrosion — thereby offering insight into the strengths and limitations of the model. The matrix highlights misclassification patterns, particularly between classes with similar visual features, helping to identify specific areas for model improvement. Performance metrics such as precision, recall, and F1-score are derived from the confusion matrix, complementing the overall accuracy and providing a more comprehensive assessment of the model’s reliability in practical corrosion detection tasks. In the present study, these performance metrics were evaluated as 0.865 (precision), 0.866 (recall), and 0.864 (F1-score) respectively, indicating a strong predictive capability of the ResNet model.

4. Conclusion: