SPDCN: Strip-based Deformable Convolutional Network for Steel Surface Defect Segmentation

2026-07-23 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

SPDCN, a Strip-based Deformable Convolutional Network, is proposed to improve steel surface defect segmentation, specifically addressing challenges with elongated, anisotropic defects like cracks and scratches. Traditional methods fail due to isotropic receptive fields and rigid sampling grids. SPDCN introduces two key innovations: the Fuzzy-enhanced Multi-scale Context Module (FMCM), which uses group-wise multi-branch convolutions and intuitionistic fuzzy channel attention for adaptive multi-scale context capture, and the Adaptive Direction-Aware Deformable Convolution (ADADC), which employs decoupled horizontal and vertical strip convolutions to anisotropically align sampling grids with defect orientations. Experiments on public steel surface defect benchmarks show SPDCN achieves 89.60% mIoU on NEU-Seg with only 3.54M parameters, surpassing state-of-the-art methods. The source code is publicly available.

Key takeaway

For Machine Learning Engineers developing industrial quality inspection systems, if your current segmentation models struggle with anisotropic steel surface defects like cracks, you should evaluate SPDCN. Its novel Adaptive Direction-Aware Deformable Convolution and Fuzzy-enhanced Multi-scale Context Module offer superior performance, achieving 89.60% mIoU on NEU-Seg with only 3.54M parameters. Consider integrating this approach or its core components to significantly improve defect detection accuracy and efficiency in your applications.

Key insights

SPDCN uses strip-based deformable convolutions and fuzzy attention to segment anisotropic steel defects, outperforming existing methods.

Principles

• Adapting receptive fields to defect anisotropy improves segmentation.
• Multi-scale context with fuzzy attention enhances defect capture.
• Decoupled strip convolutions enable directional grid alignment.

Method

SPDCN integrates FMCM for multi-scale context via group-wise multi-branch convolutions and intuitionistic fuzzy channel attention, and ADADC for anisotropic grid alignment using decoupled horizontal/vertical strip convolutions.

In practice

• Implement strip-based deformable convolutions for elongated features.
• Apply fuzzy channel attention in multi-scale context modules.
• Utilize the provided source code for steel defect segmentation.

Topics

• Steel Surface Defect Segmentation
• Deformable Convolutional Networks
• Computer Vision
• Industrial Quality Inspection
• Fuzzy Channel Attention
• NEU-Seg Benchmark

Best for: Research Scientist, AI Scientist, Machine Learning Engineer, Computer Vision Engineer

Related on AIssential

• XCT-SAM: Sequential Parameter-Efficient Domain Adaptation of SAM for Industrial XCT Defect Segmentation — Sequential parameter-efficient domain adaptation effectively bridges large domain gaps for specialized image segmentation.
• Rail Track Extraction from Rasterized Classified Point Clouds Using a Full-Resolution, Fully Convolutional Recurrent Neural Network — The FCRNN method extracts rail tracks from classified 3D point clouds using synthetic data, achieving high-quality centerlines.
• RefDiffNet: Learning to Expose Subtle PCB Defects Before Detection — RefDiffNet enhances PCB defect detection by comparing inspected images against a defect-free reference to highlight subtle flaws.
• Balancing Real and Synthetic Data for CNN-based Masonry Crack Detection — Synthetic data can significantly reduce real data collection needs for CNN-based crack detection while maintaining or improving accuracy.
• DeltaSeg: Tiered Attention and Deep Delta Learning for Multi-Class Structural Defect Segmentation — Tiered attention and refined skip connections significantly improve structural defect segmentation accuracy and generalization.
• Hybrid Deep Learning for Traceability and Classification of Industrial Slate Tiles — A hybrid deep learning model combining feature matching and classification significantly improves industrial slate tile traceability and quality control.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.