CFS-DETR: cross-scale invariant feature learning for fiber ultrastructure detection

Agglomeration detection in high-performance fiber ultrastructure is crucial for material quality control. However, traditional automatic detection methods suffer from insufficient accuracy when confronted with cross-scale feature shift and complex textures, resulting in numerous missed detections and false detections. This study proposes CFS-DETR (Cross-scale Fiber SEM DEtection TRansformer), a specialized detection framework that is designed for high-performance fiber cross-scale ultrastructure. The framework effectively addresses key challenges for ultrastructural image detection by designing a Cross-scale Invariant Module (CIM) and a Fiber SEM Edge Enhance Network(FSE-Net). Specifically, the CIM employs adversarial training mechanisms to guide the backbone network in learning cross-scale invariant feature representations, thereby mitigating scale-related feature shift issues. The FSE-Net achieves selective enhancement of edge features, improving the model’s capability for representing ultrastructural edge information. Experiment results on the DHU-OD-696 dataset demonstrate that CFS-DETR achieves improvements of 5.8 %, 1.4 % in precision and average precision metrics respectively, while reducing parameters by 1.4 million, fully validating the effectiveness and advancement of the proposed method.