Research report: TriFocus-YOLO: a triple-focus mechanism for concurrent

A research team led by Fan Wu and colleagues has introduced TriFocus-YOLO, a real-time object detection framework aimed at improving defect identification in industrial textile quality control. Published in Cluster Computing, the study targets problems that commonly derail YOLO-based approaches in real manufacturing settings—particularly textured backgrounds and defects with widely varying sizes. TriFocus-YOLO adds three mechanisms: the Feature Pyramid Edge Injector (FPEI) to feed shallow edge information into deeper layers, the Separable Dynamic Kernel Fusion (SDKF) to better fuse multi-scale features using dynamic convolution and adaptive weighting, and the Multi-scale Context Gating (MSCG) to prioritize informative channels while suppressing noise. The model reportedly outperformed a YOLO11n baseline on metrics including Recall, F1-Score, mAP50, mAP75, and mAP50-95, with gains ranging from 2.17 to 4.82 percentage points, and showed robustness across textile textures using tests on the TILDA-V2 dataset.