Skip to content
Conference

Lightweight defect detection algorithms for industrial quality inspection with few samples and their implementation on OpenCV DNN

Nov 2026 · International Conference on Algorithms, Image Processing, and Deep Learning (AIPDL 2026) · 0 citations

Abstract

A lightweight real-time deep learning framework is proposed for industrial defect inspection, centering on an improved RT-DETR model that incorporates a large kernel attention backbone and a generalized layer aggregation network. This architecture addresses two issues: few-shot learning scenarios and stringent industrial hardware constraints. To ensure compatibility with various embedded and workstation platforms, the network is further optimized thru quantization-aware training and post-training INT8 conversion to accommodate edge deployment. To identify subtle abnormal features and structural changes, this solution is evaluated using a specially developed automotive filter defect dataset. Each image has been annotated at the pixel level. Thru comparative and ablation experiments, the enhanced system demonstrated advanced multi-scale contextual representation, robust feature integration, and significantly outperformed the baseline lightweight model in terms of computational efficiency and accuracy. In-depth analysis of the impact of each module indicates that the large receptive field mechanism and quantization strategy ensure accurate and reliable defect localization, even in cases of insufficient data. The results indicate that the framework can maintain stable inference speed and accuracy across various hardware, while achieving this by reducing resource consumption and energy costs. The method based on RT-DETR significantly enhances the practical feasibility of implementing real-time automated quality inspection in complex manufacturing environments. This method also provides a scalable foundation for the broader adoption of intelligent visual industrial inspection solutions.

View source

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.