Deep Learning-Based Automated Defect Detection in Images

Image defect detection has become an essential process in manufacturing, quality control, and safety assurance, ensuring that defective products are identified before reaching consumers. However, traditional manual inspection methods are time-consuming, prone to human error, and ineffective in detecting small or complex defects, while conventional image processing techniques often fail in real-world scenarios with varying lighting, noise, and defect patterns. To address these challenges, this research implements RetinaNet, An object detection method based on deep learning algorithm known for its high accuracy in detecting small-scale defects using a Feature Pyramid Network (FPN) and the Focal Loss function to handle extreme class imbalance between defect and non-defect regions. The system is trained and evaluated on benchmark datasets such as NEU Surface Defect Database and DAGM Dataset, covering defects like scratches, pits, inclusions, and surface cracks. Images are pre-processed, augmented, and fed into the RetinaNet architecture, The ability to precisely locate and categorize flaws. Experimental results demonstrate that RetinaNet achieves superior performance compared to traditional CNN-based detectors, with accuracy of 97.5 %. This study shows that RetinaNet provides a robust, scalable, and automated solution for industrial defect detection, significantly reducing inspection time while improving detection accuracy, making it highly applicable for real-world manufacturing environments.