A Generative Adversarial Network-Based System for Food Appearance Enhancement and Automatic Grading

Abstract

The study presents an automated system based on Generative Adversarial Networks (GANs) for food appearance refinement and defect classification. A modified StyleGAN model was trained to generate 20,000 high-resolution food images, thereby expanding the dataset and improving the classification accuracy of rare defect categories by 22%. Using this extended dataset, a two-stage inspection framework was developed. YOLOv8 is applied to detect candidate defect regions, followed by EfficientNet—integrated with an attention mechanism—for classifying defect types with improved sensitivity, particularly for subtle and small-scale flaws. Experiments conducted on a public food image dataset demonstrated that the proposed method achieved a classification accuracy of 94.3%, which is 15.8% higher than conventional CNN-based models. The system also maintained a processing speed of 40 frames per second (FPS), supporting real-time industrial applications. Compared with existing approaches, the method provides more reliable data augmentation, improved model integration, and better adaptability to diverse inspection scenarios, indicating its potential for practical deployment in automated food quality assessment.