Multi-scale Deep Learning-based Robust Surface Defect Inspection for Enhancing Industrial Quality Control

Abstract

Surface imperfections in metallic, ceramic, and electronic components pose significant threats to structural integrity and brand reputation. Traditional quality control methods largely depend on manual inspection, which suffers from subjectivity, inconsistency, and high operational costs. Earlier automated approaches based on handcrafted features, such as edge detection and histogram-based texture analysis, fail to deliver sufficient robustness in complex industrial settings characterized by varying illumination conditions and diverse defect orientations. To address these limitations, this study introduces a robust hybrid framework that combines deep transfer learning with optimized machine learning classifiers. The proposed system employs the Xception architecture as a deep feature extractor, leveraging depthwise separable convolutions to effectively capture fine-grained surface irregularities while maintaining computational efficiency. The extracted high-dimensional features are then analyzed using multiple classifiers, including Stochastic Gradient Descent (SGD), Passive Aggressive Classifier (PAC), Histogram-based Gradient Boosting (HGB), and Quadratic Discriminant Analysis (QDA). Experimental findings demonstrate that this hybrid approach surpasses Convolutional Neural Network (CNN) inspection techniques in both generalization capability and real-time performance. For practical deployment, the framework is implemented within a Graphical User Interface (GUI), enabling non-technical users to easily upload images and obtain instant defect classification results. Overall, this work offers a scalable and high-precision solution for predictive quality assurance, reducing human error and enhancing productivity in modern smart manufacturing environments