This study addresses the challenges of missed and false detections in server motherboard defect identification, which arise from factors such as small target size, positional rotation deviations, and uneven scale distribution. To tackle these issues, we propose an enhanced detection model, EAE-DETR, which is based on an improved version of RT-DETR. Initially, we developed the CSP-EfficientVIM-CGLU module to enhance feature extraction capabilities while simultaneously reducing the model's parameter count through the implementation of dynamic gated convolution and global context modeling. Subsequently, we introduced the AIFI-ASSA module, designed to mitigate background noise interference and improve sensitivity to minor defects by employing an adaptive sparse self-attention mechanism. Lastly, we constructed the EUCB-SC upsampling module, which integrates depth convolution and channel shuffling strategies to enhance feature reconstruction efficiency. Experimental results on the PCBA-DET dataset indicate that EAE-DETR achieves a mean Average Precision (mAP) of 78.5% at IoU = 0.5 and 32.6% across IoU thresholds of 0.5 to 0.95, surpassing the baseline RT-DETR-R18 by 3.6% and 6.5%, respectively. Furthermore, the model demonstrates a reduction in parameter count by 21.7% and a decrease in computational load by 12.0%. On the PKU-Market-PCB dataset, the mAP50 reached 96.1%, and the mAP50:95 reached 65.1%.This model effectively facilitates high-precision and high-efficiency defect detection for server motherboards in complex industrial environments, thereby offering a robust solution for the intelligent manufacturing sector.

Under adverse weather conditions, especially haze, small target detection in Unmanned Aerial Vehicle (UAV) images faces significant challenges due to low contrast, blurred features, and scale changes. To address these issues, this paper proposes DCM-DETR (Dynamic Contrast Module), an enhanced version of Real-Time Detection Transformer (RT-DETR) tailored for drone-based robust detection in complex environments. The framework integrates three new modules: a dynamic convolution fusion module, which uses adaptive convolution kernel and multi-scale feature mixing to improve angle and scale invariance; contrast wavelet feature aggregation, which is used to decouple high-frequency details from low-frequency structural features to enhance anti-haze capability; and frequency-spatial attention and multi-scale progressive channel attention combined with frequency domain filtering and spatial attention to reduce noise and improve feature resolution. Numerous experiments on the HazyDet and VisDrone-2019 datasets show that DCM-DETR outperforms the most advanced detectors, achieving an AP50 score of 74.3% and 40.8%, respectively, while maintaining a real-time inference speed of 128.8 FPS. The model’s efficiency (53.5 GFLOPs) and balance parameter design (22.1M) further validate its practicality for deployment on resource-constrained UAV platforms. This work advances the application of RT-DETR-based models in UAVs, providing a powerful solution for extreme weather scenarios.

Aiming at the phenomena of missing detection and false detection caused by multiple instances of small and medium-sized targets, occlusion between targets and rotation deviation of target positions, an improved YOLO11(You Only Look Once 11) based small target detection algorithm for UAV images is proposed. Firstly, the C3DC module is proposed by DynamicConv and C3k2 to enhance the expression ability of different features. Secondly, C2PDA is introduced into the neck by combining Dynamic Position Bias (DPB) module to reduce the sensitivity of the model to the target position deviation. Finally, SCO module is proposed to improve the feature pyramid structure to improve the detection ability of small targets. Through the experimental verification on the VisDrone2019 dataset, the experimental results show that compared with the original YOLO11 model, the mAP50 of the improved model is increased by 3.2% to 42.2%, and the MAP50:95 is increased by 2.2% to 25.6%.It achieves 33.7% mAP50 on the UAVDT dataset, which is also better than other SOTA models.Compared with the existing YOLO model and its improved version, the proposed model maintains high detection accuracy in complex scenarios. The research results show that the algorithm has significantly improved the detection ability of UAV images in complex scenes, and its comprehensive performance comprehensively exceeds that of YOLO11 model.