Fig 1.
Comparative visualization of defect detection versus conventional object detection.
Fig 2.
The structure of YOLOv11.
Fig 3.
The structure of YOLOv11-WBD.
Fig 4.
Wavelet-Attentive Multiband Fusion (WAMF) module.
Fig 5.
Bottleneck-Enhanced Dilated U-Conv (BEDU) module.
Fig 6.
Bidirectional Depthwise Cross-Attention (BDCA) module.
Fig 7.
An example of the NEU-DET steel strip surface defect dataset.
Fig 8.
An example of the GC10-DET steel strip surface defect dataset.
Table 1.
Experimental basic environment configuration.
Fig 9.
Training and validation losses and metric progression on NEU-DET dataset.
Fig 10.
Training and validation losses and metric progression on GC10-DET dataset.
Fig 11.
Precision-recall curve.
Fig 12.
Heatmap comparison on NEU-DET datasets.
Fig 13.
Heatmap comparison on GC10-DET dataset.
Table 2.
Results of comparison experiments on dataset NEU-DET.
Table 3.
Results of comparison experiments on dataset GC10-DET.
Fig 14.
Comparative Analysis of Noisy Image Predictions on the NEU-DET dataset.
Fig 15.
Comparative Analysis of Noisy Image Predictions on the GC10-DET dataset.
Table 4.
The comparison of model missed detection rates under different noise intensities.
Table 5.
Results of ablation experiments on dataset NEU-DET.
Table 6.
Results of ablation experiments on dataset GC10-DET.