In aerial images, numerous small objects with low resolution and sparse external features present significant challenges for object detection. Existing methods typically focus on single improvement strategies to enhance detection performance, often overlooking the potential for collaborative optimization within the overall network structure. Moreover, complex network architectures considerably increase resource consumption, making efficient deployment on edge devices difficult. To address these challenges, this study proposes a lightweight Aggregation and Recalibration Network (ARNet) that optimizes the model construction across feature extraction, feature aggregation, and feature alignment. Specifically, we design three key components: the Adaptive Cross Stage Partial Network (ACSPNet), the Multi-Scale Aggregation Reconstruction (MSAR) module, and the Feature Recalibration (FR) module. ACSPNet effectively prevents the loss of critical information during feature extraction; the MSAR module captures semantic information to complement the details of objects with weak feature representations; and the FR module aligns offsets generated by feature interactions. Experimental results show that ARNet achieves notable improvements in mAP50 on the VisDrone2019-DET, AI-TOD, and TinyPerson datasets, with increases of 5.0%, 1.5%, and 6.3%, respectively, compared to baseline algorithms. Furthermore, ARNet reduces the number of parameters by 73.4%, reaching a compact size of only 2.6M. These results highlight that ARNet effectively balances detection accuracy and model size, making it particularly suitable for UAV-based object detection tasks.

Graph Convolutional Networks (GCNs) have excelled in skeleton-based action and gesture recognition due to their strong feature extraction capabilities. However, existing methods often overlook the aggregation of motion information from low-level to high-level features. To address this, we propose an Adaptive Memory Multi-level Feature Graph Convolutional Network (AMF-GCN), which preserves low-level features during propagation and progressively passes them to the output. We also introduce a Motion Capture Module (MCM) to capture intrinsic relationships between motion and spatial features, and a Dynamic Fusion Graph Convolution (DFGC) algorithm to efficiently transmit multi-level features while preserving low-level information. Additionally, a Motion Feature Enhancement Mechanism (MFEM) uses attention to highlight important features. Experimental results indicate that AMF-GCN demonstrates competitive performance compared to mainstream models, achieving 93.5% accuracy on NTU RGB+D (X-sub) and 97.5% on X-view, while delivering favorable results across multiple benchmark datasets.

Some traditional action recognition methods are limited to using appearance features to identify human behaviors. However, in real-world environments, human activities often occur in complex and changeable scenes. Therefore, learning the action features in videos from complex environments is the key to action recognition. When performing action recognition tasks in complex spatio-temporal backgrounds, fusing action information from different granularity levels is often a necessary step. In response to this idea, we propose a Multi-Granularity Action Network (MGAN), considering the modeling and fusion of spatio-temporal information with regard to action granularity. The central component of this network is the Multi-Granularity Action Excitation Module (MGAE), which excites features in parallel along different channel groups through different granularities and is capable of modeling multi-scale spatio-temporal information. To enhance the motion information between frames, we propose a Local Motion Module (LMM) for extracting fine-grained motion features. We insert the MGAE and the LMM into ResNet-50 to form a simple yet effective Multi-Granularity Action Network. We have conducted a large number of experiments on three widely used datasets, namely Something-Something V1, Something-Something V2, and UCF-101. The experiments demonstrate that with only a small number of additional parameters and computational cost introduced, the proposed MGAN achieves competitive performance.

Deep learning has achieved state‐of‐the‐art performance in accuracy of many computer vision tasks. However, convolutional neural network is difficult to deploy on resource constrained devices due to their limited computation power and memory space. Thus, it is necessary to prune the redundant weights and filters rationally and effectively. Considering that the pruned model still exists, redundancy after weight pruning or filter pruning alone, a method of combining weight pruning and filter pruning is proposed. First, filter pruning is performed, which is to remove filters with least importance and using fine‐tuning to recover the model's accuracy. Then, all connection weights below a threshold are set to zero. Finally, the pruned model obtained by the first two steps is fine‐tuned to recover its predictive accuracy. Experiments on MNIST and CIFAR‐10 datasets demonstrate that the proposed approach is effective and feasible. Compared with only weight pruning or filter pruning, the mixed pruning can achieve higher compression ratio of the model parameters. For LeNet‐5, the proposed approach can achieve a compression rate of 13.01×, with 1% drop in accuracy. For VGG‐16, it can achieve a compression rate of 19.20×, incurring 1.56% accuracy loss.