23 July 2024 in IEEE Transactions on Pattern Analysis and Machine Intelligence
The challenge of semantic segmentation with scarce pixel-level annotations has induced many self-supervised works, however most of which essentially train an image encoder or a segmentation head that produces finer dense representations, and when performing segmentation inference they need to resort to supervised linear classifiers or traditional clustering. Segmentation by dataset-level clustering not only deviates the real-time and end-to-end inference practice, but also escalates the problem from segmenting per image to clustering all pixels at once, which results in downgraded performance. To remedy this issue, we propose a novel self-supervised semantic segmentation training and inferring paradigm where inferring is performed in an end-to-end manner. Specifically, based on our observations in probing dense representation by image-level self-supervised ViT, i.e. semantic inconsistency between patches and poor semantic quality in non-salient regions, we propose prototype-image alignment and global-local alignment with attention map constraint to train a tailored Transformer Decoder with learnable prototypes and utilize adaptive prototypes for segmentation inference per image. Extensive experiments under fully unsupervised semantic segmentation settings demonstrate the superior performance and the generalizability of our proposed method.
26 June 2026 in Measurement Science and Technology
Accurate and efficient PCB defect detection is a key component of automated quality inspection in electronic manufacturing. In recent years, with the rapid development of deep learning, surface defect detection technology has made significant progress. However, the small size, diverse morphology, and complex background of PCB defects still pose significant challenges for accurate detection. To address these issues, this paper proposes a PCB surface defect detection network (MPHF-Net) based on transformer multi-dimensional perception and high-low frequency fusion. First, this paper proposes a cross stage partial multi-scale edge enhancement module to enhance multi-scale feature extraction capabilities and improve the representation of defect edge details. Second, we propose a lightweight oriented spatial convolution method to capture directional defect features. Finally, to enhance information interaction between features at different scales, we introduce the CGAFusion feature fusion mechanism to effectively integrate high-frequency texture information with low-frequency semantic information. The experimental results show that the network proposed in this paper exhibits excellent performance on PKU-Market-PCB and DeepPCB datasets. Compared with the baseline model, mAP50 and mAP50–95 improved by 2.9% and 1.2% respectively, reduced computational complexity by 11.1%, and reduced parameter count by 26.6%. The overall performance is better than popular object detection networks. The results validated the detection performance of MPHF-Net, providing a new approach for industrial quality inspection.