SDE-YOLOv5:一种用于自动摘钩机器人的实时检测算法OA
SDE-YOLOv5:A Real-time Detection Algorithm for Automatic Uncoupling Robots
在自动摘钩机器人对火车车钩进行检测和识别过程中,由于受到局部遮挡、强光照、灰尘和雨雪等环境因素干扰,会出现车钩关键特征减少甚至识别错误的情况.针对这一问题,提出了一种 SDE-YOLOv5 模型用于复杂环境下的车钩检测.首先,在 YOLO 的基础网络模型中引入 SwinTransformer 网络,增强对车钩信息的全局感知能力;然后,在颈部网络中引入重参数化模块 DBB,使模型更好地适应不同的输入和环境条件;最后,采用解耦头检测方法,可以更好地区分不同类型的车钩.实验结果表明,在自建的车钩数据集下,SDE-YOLOv5 模型在保持模型空间复杂度的同时,mAP 值达到了 96.8%.消融实验表明,改进后的网络模型在检测精度方面提高了 5.4%,具有出色的识别性能.
In the process of detecting and recognizing train couplers using automatic uncoupling ro-bots,environmental interferences such as partial occlusion,intense illumination,dust,rain and snow,can lead to the loss of critical coupler features or even misrecognition.To address this issue,we propose an SDE-YOLOv5 model for coupler detection under complex conditions.First,the Swin Transformer module is in-corporated into the backbone network of YOLOv5 to enhance global perception of coupler information.Second,a re-parameterization module DBB(diverse branch block)is introduced into the neck network,en-abling the model to better adapt to varying inputs and environmental conditions.Finally,a decoupled detec-tion head is adopted to improve differentiation among different coupler types.Experimental results on our self constructed coupler dataset demonstrate that the improved SDE-YOLOv5 algorithm achieves a mAP value of 96.8%while maintaining model spatial complexity.Ablation studies show that the improved net-work increases detection accuracy by 5.4%,exhibiting superior recognition performance.
陈文飞;韩进杰;丁佳毅;万书亭
甘肃电投常乐发电有限责任公司发电部,甘肃 酒泉 736100甘肃电投常乐发电有限责任公司发电部,甘肃 酒泉 736100华北电力大学河北省电力机械装备健康维护与失效预防重点实验室,河北 保定 071003华北电力大学河北省电力机械装备健康维护与失效预防重点实验室,河北 保定 071003
信息技术与安全科学
自动摘钩机器人车钩目标检测SDE-YOLOv5
automatic uncoupling robotcouplerobject detectionSDE-YOLOv5
《机械与电子》 2026 (9)
74-80,7
评论