Object detection is challenging in large-scale images captured by unmanned aerial vehicles (UAVs), especially when detecting small objects with significant scale variation. Most solutions employ the fusion of different scale features by building multi-scale feature pyramids to ensure that the detail and semantic information are abundant. Although feature fusion benefits object detection, it still requires the long-range dependencies information necessary for small objects with significant scale variation detection. We propose a simple yet effective scale enhancement pyramid network (SEPNet) to address these problems. A SEPNet consists of a context enhancement module (CEM) and feature alignment module (FAM). Technically, the CEM combines multi-scale atrous convolution and multi-branch grouped convolution to model global relationships. Additionally, it enhances object feature representation, preventing features with lost spatial information from flowing into the feature pyramid network (FPN). The FAM adaptively learns offsets of pixels to preserve feature consistency. The FAM aims to adjust the location of sampling points in the convolutional kernel, effectively alleviating information conflict caused by the fusion of adjacent features. Results indicate that the SEPNet achieves an AP score of 18.9% on VisDrone, which is 7.1% higher than the AP score of state-of-the-art detectors RetinaNet achieves an AP score of 81.5% on PASCAL VOC.

This paper studies motor structures and optimization methods for space robots, proposing an optimized stepped rotor bearingless switched reluctance motor (BLSRM) to solve the poor self-starting ability and significant torque fluctuation issues in traditional BLSRMs. Firstly, the advantages and disadvantages of the 12/14 hybrid stator pole type BLSRM were analyzed, and a stepped rotor BLSRM structure was designed. Secondly, the particle swarm optimization (PSO) algorithm was improved and combined with finite element analysis for motor structure parameter optimization. Subsequently, a performance analysis of the original and new motors was conducted using finite element analysis software, and the results showed that the stepped rotor BLSRM had an improved self-starting ability and significantly reduced torque fluctuation, verifying the effectiveness of the proposed motor structure and optimization method.

Recognizing video actions in the wild is challenging for visual control systems. In-the-wild videos show actions not seen in training data, recorded from various angles and scenes with the same labels. Most existing methods address this challenge by developing complex frameworks to extract spatiotemporal features. To achieve view robustness and scene generalization cost-effectively, we explore view consistency and scene joint understanding. Based on this, we propose a neural network (called Wild-VAR) to learn view and scene information jointly without any 3D pose ground truth labels, a new approach to recognizing video actions in the wild. Unlike most existing methods, first, we propose a Cubing module to self-learn body consistency between views instead of comprehensive image features, boosting the generalization performance of across-view settings. Specifically, we map 3D representations to multiple 2D features and then adopt a self-adaptive scheme to constrain 2D features from different perspectives. Moreover, we propose temporal neural networks (called T-Scene) to develop a recognizing framework, enabling Wild-VAR to flexibly learn scenes across time, including key interactors and context, in video sequences. Extensive experiments show that Wild-VAR consistently outperforms state-of-the-art methods on four benchmarks. Notably, with only half the computation costs, Wild-VAR improves accuracy by 2.2% and 1.3% on the Kinetics-400 and the Something-Somthing V2 datasets, respectively. Note to Practitioners —In human-robot interaction tasks, video action recognition technology is a prerequisite for visual control. In real applications, humans move freely in 3D space, which results in significant changes in the view of video capture and constantly changing scenes. Deep Neural Networks are limited by the perspectives and scenarios contained in the training data, resulting in most existing methods are only effective for identifying actions from 2-4 fixed views, and the background is single. Therefore, existing models are often difficult to generalize to unconstrained application environments. Human view and video scene understanding are often treated separately. Inspired by the human visual system, this paper proposes a view-to-scene video processing method in a cost-efficient way. In real-world applications, this lightweight method can be integrated into robots to help identify human behavior in complex environments. Fewer parameters indicate that the method can be easily migrated to different types of behaviors, and the reduced computational costs represent the ability to achieve real-time performance under limited hardware conditions.

Skeleton-based action recognition has achieved remarkable progress. However, in open-world scenarios, limited human visual labels, drifting skeletal structures, and novel action categories introduce complex visual disturbances that severely limit the robustness of pose representations. Herein, we propose UnicornPose, a Universal Compact Human Pose Representation, that learns robust skeleton correlations and recognizes action across various open-world scenarios. The core advantages include: 1) Continuously modeling human skeletal structures along the action timeline to construct a rich feature volume of human poses, ensuring sufficient information for universal representation. 2) Utilizing a multiview decoupling method to compress visual information further, obtaining robust pose representations that facilitate easier generalization across different open-world scenarios. 3) Coherence training and regularization constraint methods should be employed to enhance the generalization capability of noise-containing pose representations. These contributions enable UnicornPose to effectively counter noise interference and surpass the existing top results by 3-4%.

Video object detection is essential for human-interaction applications, including bimanual manipulation sensing (BMS). The effects of video detection in practical applications still need to be improved, as they are restricted by long-range spatiotemporal dependency analysis. How do humans sense bimanual manipulation in videos, especially for deteriorated clips? We argue that humans analyze the current clips based on earlier memory, namely, long-term spatial and temporal dependencies (LTSTD). However, most existing methods have yet to report significant results, as the limited exploration of these dependencies limits them. Developing an easy-to-integrate module is generally preferred for future applications rather than designing a complex end-to-end framework. Therefore, we propose a dynamic neighborhood feature multiplexing mechanism for online video object detection in this article, which is better at learning LTSTD in flexible and robust ways, boosting existing detection results, called DNFM. Specifically, we develop dynamic memory enhancement neural networks for better long-term feature aggregation with negligible additional computation costs. We multiplex each frame feature to aggregate key enhanced representations under the guidance of dynamic memory recall. The DNFM contributes to various famous detectors in BMS and other challenging detection tasks, and particular attention has been devoted to “low-quality” frame detection. Experimental results show that, while achieving state-of-the-art detection performance, DNFM clearly illustrates the easy-to-integrate operation for boosting the video object detection results.

As skeletal data can be collected non-invasively while preserving patient privacy, it is widely used in public medical datasets to document patient behavior. Autism Spectrum Disorder (ASD) is characterized by significant behavioral heterogeneity, reflected in the topological structure and dynamic evolution of skeletal movements. This complexity poses substantial challenges for skeleton-based behavioral analysis. Existing methods struggle to effectively utilize behavioral evolution for subject-specific reasoning, leading to suboptimal representations that lack diagnostic relevance for autism. To address this limitation, we propose a Behavioral Evolution-based Edge Reconstruction (BER) Strategy for learning autism-related behavioral representations. By reconstructing a high-granularity adjacency matrix that spans both spatial and temporal dimensions, utilizing dynamic evolution and spatial location information, BERGCN enhances behavioral reasoning. Specifically, we first compute channel-level spatial and temporal edge reconstruction parameters by performing feature compression and targeted convolution operations on the differences between neighboring frames. Based on these, the spatial edge reconstruction module is designed by combining a generic attention map with two personalized attention maps, while the temporal edge reconstruction module is implemented using flexible frame replace ment and weighted aggregation. Finally, we investigate both single-modal and multimodal network architectures under various fusion strategies. We evaluate BERGCN on three autism clinical tasks and a benchmark action recognition dataset. Experimental results demonstrate competitive performance, showing improved sensitivity to subject-specific behavioral patterns while maintaining computational efficiency.