Skip to content

Yong-Ting He and Jin-Shan Zhang, EASE-DET: An efficient adaptive sparse enhancement algorithm for steel surface defect detection, Eur. J. Math. Appl. 6 (2026), Article ID 16.

Full Text: PDF

Volume 6 (2026), Article ID 16

https://doi.org/10.28919/ejma.2026.6.16

Published: 13/11/2026

Abstract:

Steel surface defects pose a persistent threat to product quality and structural integrity. Standard detection networks remain limited in three respects: the backbone struggles to discriminate defects from visually similar background textures; cross-scale fusion in the feature pyramid introduces spatial misalignment, degrading localization accuracy for small objects; and per-scale parameterization of the detection head causes redundancy while offering insufficient sensitivity to defect edge gradients. Taking YOLO11 as the baseline, this paper introduces an Adaptive Position-aware Feature Extraction Module to strengthen global context modeling in the backbone, an Adaptive Context-aware Feature Pyramid Network to improve cross-scale alignment, and an Efficient Shared Detection Head that combines parameter sharing with multi-directional differential convolution for lightweight edge enhancement. On the NEU-DET dataset, EASE-DET achieves 78.3% [email protected], a 3.7-point improvement over the baseline. Cross-domain evaluation on GC10-DET further validates the generalization capability of the proposed method.

How to Cite:

Yong-Ting He and Jin-Shan Zhang, EASE-DET: An efficient adaptive sparse enhancement algorithm for steel surface defect detection, Eur. J. Math. Appl. 6 (2026), Article ID 16. https://doi.org/10.28919/ejma.2026.6.16