Fig. 5 | Scientific Reports

Fig. 5

From: Confidence–gradient reweighting and lightweight feature enhancement algorithm for steel surface defect detection

Fig. 5

DS-CBSM++ workflow. Solid lines denote the forward data flow; input images are grouped into a mini-batch by a weighted random sampler (orange) and fed into the YOLO11 detector; dashed lines indicate the feedback loop: statistics of the classification loss (as a gradient surrogate) and the detection head’s sigmoid confidence are sent to the DS-CBSM++ module (blue) to update \(g_c^{\textrm{ema}}\) and \(c_c^{\textrm{ema}}\).

Back to article page