Named Entity Recognition (NER) aims to identify entities with specific meanings and their boundaries in natural language texts. Due to the differences between Chinese and English language families, Chinese NER faces challenges such as ambiguous word boundary delineation and semantic diversity. Previous studies on Chinese NER have focused on character and lexical information, neglecting the unique feature of Chinese—pinyin information. In this paper, we propose CPL-NER, which combines multiple feature information of Chinese characters as embedding to enhance the semantic representation by introducing pinyin and dictionary information. For Chinese named entity recognition, pinyin information of Chinese characters helps to resolve the polyphonic phenomenon, while dictionary information aids in addressing word segmentation ambiguities. Additionally, we innovatively designed the Pinyin-Lexicon Cross-Attention Mechanism (PLCA), which calculates attention scores between various embeddings. This mechanism deeply integrates character, pinyin, and lexicon embeddings, generating character sequences enriched with semantic information. Finally, BiLSTM-CRF is employed for sequence modeling. Through this design, we can more comprehensively capture semantic features in Chinese text, improving the model’s ability to handle polyphonic characters and word segmentation ambiguities, thereby enhancing the recognition performance of Chinese named entities. We conducted experiments on four standard Chinese NER benchmark datasets, and the results show that our method outperforms most baselines, demonstrating the effectiveness of our proposed model.

Introduction: Retinal vessel segmentation is difficult because fundus images contain branching structures, scale variation, uneven illumination, lesions, and weak vessel boundaries. Methods: We propose BGAFS-UNet, combining Boundary Perception Module (BPM), Lightweight Wide Gated Attention (LWGA) with Dynamic Channel Allocation (DCA) and Adaptive Branch Fusion (ABF), Cross-Stage Fusion (CSF), Contextual Mixing (ContMix), and Attention-Guided Dynamic Sampling (AGDS). We evaluate three prespecified image-disjoint runs on DRIVE, STARE, and CHASE and use image-level paired tests with Holm correction on DRIVE. Results: The full model obtains F1 scores of 80.61% on DRIVE, 76.15% on STARE, and 79.53% on CHASE. On DRIVE, corrected improvements are metric- and vessel-scale-specific: boundary F-score, clDice, and large-vessel Dice are significant, whereas Dice and small-vessel Dice are not. Discussion: BGAFS-UNet is parameter-efficient but not unconditionally more efficient, with higher FLOPs and latency than U-Net; the findings support structure-aware segmentation while motivating broader cross-device validation and optimization. The code and reproducibility materials remain available at https://gitee.com/Zhangguoce/bgafs-net.git , fixed at commit 477b217990f68d43d5c1c8f3a818d8cf2c8af384.