文献类型：期刊文献

英文题名：An Improved YOLOv8 Algorithm Based on BiFPN Architecture for Detection of Aerospace Connector Pins

作者：Wang, Chaofeng[1]; Liu, Shijun[2]; Ni, Zongze[3]; Yi, Jianjun[1]

机构：[1] East China University of Science and Technology, School of Mechanical and Power Engineering, Shanghai, China; [2] Shanghai Aerospace Equipment Manufacturer Co, Shanghai, China; [3] School of Mechanical and Power Engineering, Shanghai, China

年份：2025

起止页码：189

外文期刊名：2025 4th International Symposium on Aerospace Engineering and Systems, ISAES 2025

收录：EI(收录号：20260920171196)

语种：英文

外文关键词：Aerospace engineering - Architecture - Connectors (structural) - Network architecture - Object recognition

摘要：To address the performance limitations of the YOLOv8 model in aerospace connector pin detection due to the small scale and dense distribution of pins, this paper proposes an improved YOLOv8 algorithm based on the bidirectional feature pyramid network (BiFPN) architecture. First, we integrate the BiFPN architecture into the model, effectively enhancing feature interaction and information transfer efficiency across different scales. Second, to improve the model's localization and recognition capabilities for small-scale targets, we introduce a P2 detection layer for small objects on top of the existing detection network. Finally, to mitigate the sensitivity of the IoU loss function to positional deviations in small targets, we optimize the regression loss function using the Normalized Wasserstein Distance (NWD) metric. Experimental results on a self-constructed aerospace connector dataset demonstrate that the proposed method achieves improvements of 2.3% in Precision and 16.3% in mAP50-95, respectively, compared to the baseline YOLOv8 model. Our approach significantly enhances performance in aerospace connector pin detection tasks while balancing computational requirements. ? 2025 IEEE.