An end-to-end Structure-strategy Defect Detection Network (SSDDNet) is proposed in this research to solve the problems in surface defect detection of metallic workpieces, such as complex defect morphologies, diverse scales, ambiguous boundaries, unstable label quality, and the high cost of pixel-level annotation. A multi-scale contextual aggregation module is structurally established to fuse dynamic convolution and multi-dilation information, while a boundary enhancement module is introduced to strengthen the modeling of ambiguous boundaries. A spatial label uncertainty modeling approach is strategically introduced to enable stable training under weakly annotated conditions. Experimental results show that: (1) SSDDNet achieves a 1.1 percentage points improvement in mean precision over the state-of-the-art MixSup model on the public KolektorSDD2 dataset without pixel-level annotations. (2) SSDDNet outperforms MixSup by 17.6 percentage points in mean accuracy and achieves approximately 15 percentage points improvement in SSH on the self-constructed industrial dataset BatteryBase, indicating the strong generalization capability of the proposed model. This research provides a novel approach for surface defect detection of metal workpieces.

The circumferential fibers are the key areas of meniscus stress. The construction of three-dimensional microstructure of circumferential fibers is of great significance for the treatment of meniscus injury and the development of artificial meniscus. At present, the circumferential fibers of meniscus in micro computed tomography (MicroCT) images are segmented manually. Because of the complex microstructure of meniscus, manual segmentation has some problems such as low efficiency and inconsistent segmentation standards. To solve the problem of few sample images, an image amplification method is proposed based on the characteristics of MicroCT images. To solve the problem of large edge segmentation errors, an improved model based on TransUNet algorithm is proposed, image Relative Position Encoding (iRPE) is introduced, and the loss function is improved. The experimental results show that: (1) the improved model can accurately and completely segment the meniscus tissue, and the segmentation results can successfully complete the three-dimensional reconstruction of circumferential fibers. (2) The introduced iRPE algorithm improves the segmentation effect of model edge details, the improved loss function enables the model to better adapt to the situation of sample imbalance, and the proposed image amplification method solves the problem of insufficient data sets and comprehensively improves the performance of the model.The results show that the average precision of circumferential fiber segmentation is 98.66%. (3) In the three-dimensional model of circumferential fibers, it is found that fibers are divided into two parts, and a small amount of fibers are divided into three parts. The proposed method can segment the meniscus circumferential fibers in MicroCT images with high accuracy and efficiency, and can pave the way for the study of the force analysis of meniscus in three-dimensional space.