Papers Steel Defect Detection Model Based on State Space Model

Automatic detection of steel defects improves production efficiency and quality, ensuring the safety and reliability of structures. However, traditional detection methods struggle to accurately identify defects in steel images with complex background noise and low contrast. To address this, we propose a steel defect detection model based on the state space model (SSM). First, SSM extracts defect features with a global receptive field while maintaining linear complexity, and a dynamic routing mechanism aggregates results from different spatial directions. Next, we design a ghost convolution-based self-calibration block to adjust defect boundaries and enhance feature representation. Then, a decoupled detection head based on heterogeneous representations is constructed to improve detection results by enhancing the relationship between classification and localization. Finally, experimental results validate the effectiveness and advantages of the model.