In Unmanned Aerial Vehicle (UAV) target detection tasks, issues such as missing and erroneous detections frequently occur owing to the small size of the targets and the complexity of the image background. To improve these issues, an improved target detection algorithm named RLRD-YOLO, based on You Only Look Once version 8 (YOLOv8), is proposed. First, the backbone network initially integrates the Receptive Field Attention Convolution (RFCBAMConv) Module, which combines the Convolutional Block Attention Module (CBAM) and Receptive Field Attention Convolution (RFAConv). This integration improves the issue of shared attention weights in receptive field features. It also combines attention mechanisms across both channel and spatial dimensions, enhancing the capability of feature extraction. Subsequently, Large-Scale Kernel Attention (LSKA) is integrated to further optimize the Spatial Pyramid Pooling Fast (SPPF) layer. This enhancement employs a large-scale convolutional kernel to improve the capture of intricate small target features and minimize background interference. To enhance feature fusion and effectively integrate low-level details with high-level semantic information, the Reparameterized Generalized Feature Pyramid Network (RepGFPN) replaces the original architecture in the neck network. Additionally, a small-target detection layer is added to enhance the model’s ability to perceive small targets. Finally, the detecting head is replaced with the Dynamic Head, designed to improve the localization accuracy of small targets in complex scenarios by optimizing for Scale Awareness, Spatial Awareness, and Task Awareness. The experimental results showed that RLRD-YOLO outperformed YOLOv8 on the VisDrone2019 dataset, achieving improvements of 12.2% in mAP@0.5 and 8.4% in mAP@0.5:0.95. It also surpassed other widely used object detection methods. Furthermore, experimental results on the HIT-HAV dataset demonstrate that RLRD-YOLO sustains excellent precision in infrared UAV imagery, validating its generalizability across diverse scenarios. Finally, RLRD-YOLO was deployed and validated on the typical airborne platform, Jetson Nano, providing reliable technical support for the improvement of detection algorithms in aerial scenarios and their practical applications.

Edge computing frameworks support the large-scale deployment of ground-based drone detection systems in the lowaltitude economy by addressing the complexity and latency issues of traditional cloud architectures. However, the small size of drone targets and complex backgrounds reduce detection accuracy, while high-precision models are challenging to deploy on resourceconstrained edge devices. To tackle these issues, this paper presents DACG-Net, a detection network comprising a Dynamic Alignment Network (DANet) and a Context-Guided Feature Pyramid Network (CG-FPN). DANet consists of two distinct backbones with a shared Stem to eliminate redundant shallow feature computation. It introduces a Dynamic Alignment Fusion Module (DAFM) to address spatial misalignment in dual-backbone fusion, enhancing adaptability to diverse features. In CG-FPN, the fusion approach based on the Concat module in traditional feature pyramids lacks effective feature selection and inter-layer interaction. To improve this, a Context-Guided Fusion Module (CGFM) is designed for refined feature aggregation. Experiments on multiple datasets demonstrate the superior performance of DACG-Net, achieving mAP50 improvements of 5.8%, 6.7%, 6.6%, and 8% over baseline models. Moreover, it reduces parameters by 10.3% with only a 16.9% increase in computational complexity. DACG-Net runs at 18 FPS on embedded devices, satisfying real-time detection requirements.

Machine vision-based anti-drone detection systems enable long-range, cost-effective target monitoring in complex environments. However, small drones typically occupy only a few pixels in captured images. Existing detectors suffer from semantic loss and insufficient fusion during feature extraction and cross-scale interaction, resulting in limited detection accuracy. To address these challenges, this paper proposes Diffusion Focusing Former (DFFormer), a detection framework specifically designed for small target identification. The framework employs a backbone network to extract multi-layer features, which are enhanced through an Advanced Feature Processing Layer (AFPL) to strengthen semantic representation. A Feature Scaling Layer (FSL) then organically fuses shallow and high-level information before encoder processing, preserving fine-grained cues while minimizing computational overhead. Subsequently, the Multi-Scale Focusing Diffusion Network (MSFDN) processes scaled features for cross-scale interaction and progressive fusion. The Focusing Fusion Module (FFM) injects comprehensive contextual information into each scale throughout this process. Experimental results on three anti-drone datasets (DUT-Anti-UAV, Bird-UAV, and Anti-UAV (Inf)) demonstrate that DFFormer consistently outperforms existing state-of-the-art methods across multiple evaluation metrics. Generalization validation on the VisDrone2019 aerial dataset further confirms the method’s applicability to diverse scenarios and configurations.