🤖 AI Summary
This work addresses the challenge of effectively segmenting slender, anisotropic defects—such as cracks and scratches—on steel surfaces, which existing methods struggle to handle accurately. To this end, the authors propose the SPDCN network, which incorporates a Fuzzy-enhanced Multi-scale Context Module (FMCM) to adaptively fuse multi-scale contextual information. Additionally, an Adaptive Direction-Aware Deformable Convolution (ADADC) is introduced, leveraging decoupled horizontal and vertical strip convolutions within a grouped multi-branch architecture enhanced by an intuitionistic fuzzy channel attention mechanism. This design enables precise modeling of defect morphology and dominant orientation. Evaluated on benchmark datasets including NEU-Seg, the proposed method achieves a state-of-the-art mIoU of 89.60% with only 3.54 million parameters, outperforming current advanced approaches.
📝 Abstract
Steel surface defect segmentation is critical for industrial quality inspection, yet existing methods struggle with elongated, anisotropic defects such as cracks and scratches due to the isotropic receptive fields of standard convolutions and rigid sampling grids that cannot adapt to irregular defect boundaries. To address these limitations, we propose Strip-based Predictor for Deformable Convolutional Networks (SPDCN) with two key innovations. The \textbf{Fuzzy-enhanced Multi-scale Context Module (FMCM)} employs group-wise multi-branch convolutions with an intuitionistic fuzzy channel attention mechanism to adaptively capture multi-scale contextual information across varying defect sizes. The \textbf{Adaptive Direction-Aware Deformable Convolution (ADADC)} replaces the conventional offset predictor with decoupled horizontal and vertical strip convolutions, enabling the deformable sampling grid to anisotropically align with the principal orientation of elongated defects. Extensive experiments on public steel surface defect benchmarks demonstrate that SPDCN consistently outperforms state-of-the-art methods, achieving 89.60\% mIoU on NEU-Seg with only 3.54M parameters. The source code is publicly available at https://github.com/DWlzm .