Small target detection is an important research direction in computer vision, widely applied in scenarios such as drone monitoring, remote sensing image analysis, and autonomous driving. However, as small targets occupy fewer pixels, contain limited feature information, and often appear in complex backgrounds, existing detection algorithms face shortcomings in accuracy and robustness. To address this, this paper proposes a novel small target detection algorithm that integrates hierarchical feature fusion with a spatial dynamic collaboration mechanism. The hierarchical feature fusion module (HFA) effectively combines shallow detail features with deep semantic features, greatly enhancing the feature representation capability for small targets. Meanwhile, the dynamic collaboration mechanism (DCCA) dynamically adjusts feature fusion weights and detection strategies based on target scale and density distribution, thereby further improving detection accuracy and robustness. Extensive experiments are conducted on datasets such as VisDrone, TinyPerson, and NWPU VHR-10. Results demonstrate that, compared to state-of-the-art models like YOLOv8 and YOLOv10, the proposed algorithm achieves significant improvements in precision, recall, and mAP, with mAP increasing by 2.1% to 3.2% and mAP-95 by 1.2% to 1.8%. Ablation studies further validate the complementarity of HFA and DCCA in optimizing model performance, confirming the algorithm’s superiority and robustness in complex scenarios. This research provides a novel technical route for small target detection and offers valuable references for practical applications.

For the intricate task of multi-label image classification, this paper introduces an innovative approach: an attention-guided dual-branch dynamic graph convolutional network. This methodology is designed to address the difficulties faced by current models when handling multiple labels within images. By integrating multi-scale features, it enhances the retention of original category information and boosts the robustness of feature learning. Utilizing a semantic attention module, the study dynamically reweights feature categories in the training dataset, enhancing the network's capability to identify smaller objects and generate context-sensitive category representations. The effectiveness of the proposed model was evaluated using the MS-COCO2014 imagery dataset, demonstrating superior performance in critical metrics such as classification precision (CP), recall (CR), and F1 score (CF1), outperforming other state-of-the-art models. Furthermore, a cascaded classification structure was implemented to leverage the prior information from static images to inform the processing of dynamic ones, and to utilize original image category data to augment label correlations, thereby enhancing overall classification accuracy.

This paper addresses the problem of joint modeling for multi-source heterogeneous graph data in distributed environments by proposing a federated graph neural network classification framework driven by structural alignment and consistency regularization. The method preserves data locality by enabling each participant to learn node features and topological information through a local graph neural network encoder. A cross-source structural alignment module maps embeddings from different graphs into a shared representation space, mitigating semantic inconsistencies caused by structural differences. Additionally, a consistency regularization mechanism is introduced to enhance the robustness of node representations through multi-view perturbations, improving the model's generalization ability during training. At the global level, a federated averaging strategy is adopted to periodically aggregate local models, enabling collaborative optimization and enhancing the consistency and discriminative capacity of the global representation. To validate the effectiveness of the proposed approach, experiments are conducted in a multi-source heterogeneous graph environment using various node distribution strategies. The results show that the method outperforms existing federated graph learning models in terms of accuracy, clustering consistency, and structural expressiveness, achieving efficient multi-source graph classification while preserving data privacy.

This paper proposes a subgraph-aware classification framework that integrates efficient frequent subgraph mining with graph neural networks (GNNs) to address the limitations of existing GNNs in capturing explicit local structures. Specifically,we first introduce an improved mining strategy to extract representative and low-redundancy structural patterns from large-scale graph data, significantly reducing computational complexity and pattern duplication. These patterns are then dynamically injected into a subgraph-aware GNN module via a subgraph-corrected attention mechanism and gated fusion strategy, enabling deep collaborative modeling of local and global semantics. An end-to-end joint training mechanism is also introduced to coordinate subgraph selection with representation learning. Experimental results on the IMDB-MULTI and REDDIT-MULTI-12K datasets demonstrate that our method outperforms several state-of-the-art baselines. On IMDB-MULTI, our model achieves an Accuracy of 78.6% and a Macro-F1 score of 77.4%, showing notable gains over competing approaches. Further ablation and visualization analyses confirm the importance of both the frequent subgraph mining component and the subgraph-aware fusion mechanism in enhancing representation quality and model interpretability. This work presents a unified and interpretable framework for graph classification, offering new insights into structural perception and feature fusion in GNN-based learning.