Deep Learning-Based Method for Detecting Surface Defects in Air Filters

To prevent defective air filters from reaching the market, it is essential to conduct defect inspections on their appearance. This paper proposes a deep learning-based method for detecting surface defects in air filters. We utilize a linear array camera to capture images of air filters, constructing a dataset and designing a defect detection algorithm. For the outer metal mesh and inner filter material, we propose a multistage detection approach: (1) segmenting the inner and outer structures using a style transfer model; (2) performing defect-free reconstruction of the metal mesh and localizing defects through residual analysis; (3) classifying and pixel-level localizing defects in the connected regions of the filter material. Experimental results demonstrate that the proposed method effectively achieves pixel-level defect localization even in the presence of an imbalance between defective and non-defective data.