Fig. 3 | Scientific Reports

Fig. 3

From: Multiscale diffusion-enhanced attention network for steel surface defect detection in Polysilicon Production

Fig. 3

The overall flowchart of MSEOD-DDFusionNet. The diagram depicts the complete pipeline from image input to defect detection output. The process begins with multi-scale feature extraction via the MTECAAttention module, followed by shape adaptation using ODConv. Subsequent stages involve feature fusion and optimization through SPPF, C2PSA, and LMDP modules. Simultaneously, the DDFusion module enhances robustness via noise injection and denoising. The pipeline concludes with defect identification and localization in the detection head.

Back to article page