Treffer: Research on the Application of YOLOv13 Multimodal Object Detection Framework based on Ant Colony Algorithm and Transformer in Steel Surface Defect Detection.

Aiming at the problems of low contrast defects, complex textures, high missed detection rate of small targets, and difficult edge deployment in steel strip surface defect detection, this paper proposes ACA-TRANSFORMER-YOLOv13. The model uses cross-modal Transformer to fuse the complementary features of visible light, infrared and laser point cloud, and uses Ant Colony Optimization (ACO) to adaptively optimize the anchor box. At the same time, the Adaptive Channels Attention (ACA) is embedded to realize the dynamic adjustment of attention, which significantly enhances the discrimination of weak defects and suppresses background interference. On NEU-DET benchmark, mAP@0.5 reaches 80.1 % and F1 reaches 0.72. Compared with YOLOv13n, mAP@0.5 is increased by 3.7 % and mAP@0.5:0.95 is increased by 2.3%. Compared with YOLOv13s, the calculation amount is reduced by 68% and mAP@0.5 is increased by 2.1%. Experiments show that ACA-TRANSFORMER-YOLOV13 achieves the best balance between accuracy and lightweight, and provides a landing technical benchmark for real-time defect detection in industrial sites. [ABSTRACT FROM AUTHOR]