Existing traffic sign detection algorithms, when deployed on edge computing devices, often face challenges such as high computational overhead and insufficient detection capabilities for small or deformed targets. To address these limitations, this paper proposes LNSE-YOLO, a lightweight traffic sign detection algorithm based on YOLOv11, which incorporates novel noise suppression and semantic enhancement techniques. Initially, building upon the YOLOv11n architecture, we design a Feature Fusion Module named NSSE (Noise Suppression and Semantic Enhancement) to optimize multi-scale feature representation and effectively mitigate background interference. Subsequently, an Edge-Driven Feature Enhancement Network (ED-FEN) is integrated into the backbone, and a Local Deformable Attention (LDA) module is incorporated into the detection head. These additions are specifically engineered to bolster the model’s perception of small target edges and enhance its robustness to geometric distortions, culminating in the development of a high-precision model, NSE-YOLO. Finally, to achieve model lightweighting, a channel pruning strategy based on BatchNorm scaling factors is employed to compress the NSE-YOLO, yielding the ultimate LNSE-YOLO model. Experimental results on the TT100K and CCTSDB datasets demonstrate that, compared to the baseline YOLOv11n, our high-precision NSE-YOLO model achieves significant improvements in mAP@50 by 5.5 and 2.5 percentage points, respectively. The pruned LNSE-YOLO model, while maintaining its accuracy advantage, exhibits a substantial reduction in parameters and comparable computational cost to the baseline. This unequivocally validates the effectiveness and practical utility of our proposed methodology.

Knowledge graphs have become core infrastructure for intelligent applications, yet entity alignment across heterogeneous knowledge graphs remains a significant challenge. Traditional approaches primarily rely on structured information, overlooking the rich multi-modal data associated with entities in real-world knowledge bases. This paper proposes a novel Cross-modal Hierarchical Graph Reasoning Network for multi-modal entity alignment. Our framework introduces a hierarchical attention mechanism that first performs intra-modal graph convolution to capture structural patterns within individual knowledge graphs, followed by cross-modal graph attention to dynamically fuse visual, textual, and structural representations. To address the challenges of modal heterogeneity and data sparsity, we design a semantic-gated adaptive fusion module that learns optimal weights for different modalities based on entity context and modality confidence. Furthermore, we develop an iterative collaborative reasoning process that enables mutual enhancement between multi-modal semantics and graph structures, with updated embeddings in turn optimizing intra-graph and inter-graph alignment relations. We also incorporate a cross-modal contrastive learning strategy specifically designed for sparse entities to enhance robustness in long-tail distribution scenarios. Comprehensive experiments on public multi-modal entity alignment benchmark datasets demonstrate that our approach achieves state-of-the-art performance. This work provides a robust and effective solution for multi-modal knowledge graph fusion, advancing the field toward more practical application scenarios where complete and clean data cannot be guaranteed.

Accurate detection of tobacco plants in complex field environments is critical for precision agriculture, crop monitoring, and yield estimation. Traditional manual counting methods are time-consuming, labor-intensive, and susceptible to environmental and subjective factors. In this study, we propose an improved YOLO11-based framework for automated tobacco plant detection, specifically designed to address challenges such as scale variation, dense distribution, and background interference. The framework integrates four key modules: the Edge-Enhanced Feature Stem (EEFS) to strengthen low-level feature extraction, the Multi-Scale Kernel Interaction (MSKI) to capture multi-scale contextual information, the Adaptive Weighted Feature Fusion (AWFF) to optimize feature aggregation, and the Global–Local Synergistic Attention (GLSA) to enhance feature discrimination by jointly modeling local details and global context. A comprehensive UAV-based tobacco dataset was constructed, encompassing multiple lighting conditions, collection heights, and observation angles. Experimental results demonstrate that the proposed method significantly outperforms the YOLO11 baseline and achieves superior performance compared to mainstream YOLO variants. Ablation studies and heatmap visualizations confirm the effectiveness of each module. Furthermore, the model exhibits robust performance under multi-dimensional environmental perturbations, including varying illumination, scale, and camera angles. The proposed framework provides a practical and efficient solution for automated tobacco plant counting, offering potential applications in UAV-based precision agriculture and large-scale crop monitoring.