Steel Plate Defect Detection and Classification using Light weight MobileNetV2 Model

In the steel manufacturing sector, automated surface flaw identification is critical to manufacturing quality monitoring. Unfortunately, due to its poor accuracy and sluggish operation, the manual approach is not suitable to be utilized on a manufacturing process. The prevalent AI-based technique now in use suffers from poor accuracy as well, while there remains much space for development. This paper presents a comparative analysis of various classifiers for steel plate defect detection, focusing on the performance of Random Forest, Xgboost, MobileNet, and MobileNetV2. The Random Forest classifier achieved an accuracy of 46.66%, while Xgboost performed slightly lower at 43.33%. A combination of MobileNet for feature extraction and Xgboost for classification resulted in a notable improvement, reaching an accuracy of 60.00%. However, the most impressive performance was observed with MobileNetV2, which achieved an accuracy of 83.00%, significantly outperforming the other models. This study demonstrates the superior efficacy of MobileNetV2 for detecting defects in steel plates, highlighting its potential for industrial applications. The paper also discusses the future scope of enhancing defect detection through advanced deep learning techniques and real-time implementation.