全文下载:
202605017.pdf
文章编号:1672-6987(2026)05-0143-09;DOI:10. 16351/j. 1672-6987. 2026. 05. 017
陈凤魁 1 ,谢绍勤 2 ,杨 清 1* (1. 青岛科技大学 信息科学技术学院,山东 青岛 266061;2. 兖矿物流科技有限公司 铁路分公司,山东 济宁 273500)
摘 要:现有开放式障碍物检测算法面临着计算资源有限,检测速度慢,难以满足实时需求, 在保证检测准确性的同时,提高算法实时性成为了一项关键挑战。为了解决上述挑战,本文 提出一种改进的 WSM-DETR 网络,在保证模型准确性的同时,提升了算法实时性。首先,使 用 WTConv 对原网络中 RepC3 模块进行改进,使用小波变换结合小核卷积对输入图像进行处 理,增强模型特征提取能力并显著降低模型计算消耗。其次,添加 SCSA 注意力机制,结合空 间注意力和通道注意力,再不增加模型参数量和计算量的条件下,提升目标检测精度。然后, 引入 MPDIOU 损失函数,通过最小化预测框和真实框之间的关键点距离,提升模型预测精 度。实验结果表明,WSM-DETR 模型的参数量和计算量分别减少了 18. 1%、34. 5%,同时模 型 Precision、Recall、mAP50 和 FPS 分别提升了 0. 7%、0. 9%、1. 2%、8. 9%,验证了所提算法 的实时性与准确性。
关键词:开放式铁路; RT-DETR;轻量化;障碍物检测
中图分类号:TP 391. 4 文献标志码:A
引用格式:陈凤魁,谢绍勤,杨清 . 基于改进 RT-DETR 的开放式铁路障碍物检测算法[J]. 青岛科技大学学报(自然科学版),2026,47(5):143-151.
CHEN Fengkui, XIE Shaoqin, YANG Qing. Open railway obstacle detection algorithm based on improved RT-DETR[J]. Journal of Qingdao University of Science and Technology (Natural Science Edition),2026,47(5):143-151.
Open Railway Obstacle Detection Algorithm Based on Improved RT-DETR
CHEN Fengkui1 ,XIE Shaoqin2 ,YANG Qing1 (1. College of Information Science and Technology,Qingdao University of Science and Technology,Qingdao 266061,China; 2. Railway Branch Company,Yankuang Logistics Technology Co. , Ltd. ,Jining 273500, China)
Abstract:The existing obstacle detection algorithms for open railways face challenges such as limited computational resources, slow detection speeds, and difficulty in meeting real-time requirements. Improving the algorithm's real-time performance while ensuring detection accuracy has become a critical challenge. To address this challenge, this paper proposes an improved WSM-DETR network that enhances algorithm real-time performance while maintaining model accuracy. First, WTConv is used to modify the RepC3 module in the original network. By combining wavelet transform with small kernel convolutions to process the input image, the model's feature extraction capability is strengthened, and the computational cost is significantly reduced. Second, the SCSA attention mechanism is added, which combines spatial and channel attention to improve object detection accuracy without increasing the model's parameter count and computational load. Then, the MPDIOU loss function is introduced, which minimizes the key point distance between the predicted and ground truth bounding boxes to improve model prediction accuracy. Experimental results show that the WSM-DETR model reduces the parameter count and computational load by 18. 1% and 34. 5%, respectively, while Precision, Recall, mAP50, and FPS increase by 0. 7%, 0. 9%,1. 2%, and 8. 9%, respectively, validating the real-time performance and accuracy of the proposed algorithm.
Key words:open railway; RT-DETR; lightweight; obstacle detection
收稿日期:2025-10-04
基金项目:山东省自然科学基金项目(ZR2022ME194);山东能源集团科技项目(YK2023B47-R45).
作者简介:陈凤魁(2002—),男,硕士研究生 . * 通信联系人 .