MSDC-FRN: Multi-Scale Dense Feature Reconstruction for Few-Shot Metal Surface Classification

Automated surface defect classification plays a pivotal role in industrial quality control. However, its efficacy is often constrained by the scarcity of annotated samples in real-world scenarios. This few-shot learning challenge is particularly challenging for metal surfaces, which are characterized by high intra-class variance and subtle inter-class differences among defect types. To address these issues, we propose a novel multi-scale densely connected feature reconstruction network (MSDC-FRN). Our model integrates a multi-scale dense connection backbone to extract rich and robust features across diverse receptive fields. In addition, we introduce a feature reconstruction network (FRN) classifier that measures the sample similarity by framing it as a ridge regression problem. This unique formulation markedly enhances feature discriminability and model generalization. Extensive experiments on two benchmark metal defect datasets demonstrate that MSDC-FRN achieves a performance that is superior or comparable to state-of-the-art methods in both intra-domain and challenging cross-domain settings. These results validate the effectiveness of our model and highlight its potential as a robust solution for intelligent quality inspection in modern manufacturing.