HyADS: A Hybrid Lightweight Anomaly Detection Framework for Edge-Based Industrial Systems with Limited Data

Industrial defect detection in edge computing environments faces critical challenges in balancing accuracy, efficiency, and adaptability under data scarcity. To address these limitations, we propose the Hybrid Anomaly Detection System (HyADS), a novel lightweight framework for edge-based industrial defect detection. HyADS integrates three synergistic modules: (1) a feature extractor that integrates Histogram of Oriented Gradients (HOG) and Local Binary Patterns (LBP) to capture robust texture features, (2) a lightweight U-net autoencoder that reconstructs normal patterns while preserving spatial details to highlight small-scale defects, and (3) an adaptive patch matching module inspired by memory bank retrieval principles to accurately localize local outliers. These components are synergistically fused and then fed into a segmentation head that unifies global reconstruction errors and local anomaly maps into pixel-accurate defect masks. Extensive experiments on the MVTec AD, NEU, and Severstal datasets demonstrate state-of-the-art performance. Notably, HyADS achieves state-of-the-art F1 scores (94.1% on MVTec) in anomaly detection and IoU scores (85.5% on NEU/82.8% on Seversta) in segmentation. Designed for edge deployment, this framework achieves real-time inference (40–45 FPS on an RTX 4080 GPU) with minimal computational overheads, providing a practical solution for industrial quality control in resource-constrained environments.