基于ConvNext的无人驾驶机车运行中的障碍物检测
Obstacle Detection in the Operation of Driverless Locomotive Based on ConvNext
刘莹莹 1李昱衡 2何江 3任宇昕 4董洋2
作者信息
- 1. 鞍钢集团关宝山矿业有限公司,辽宁 鞍山
- 2. 鞍钢集团矿业有限公司眼前山分公司,辽宁 鞍山
- 3. 北京科技大学,北京
- 4. 东北大学,辽宁 沈阳
- 折叠
摘要
矿用电机车是矿山运输的主要设备,由于井下运行环境复杂,照明条件有限,无人驾驶的电机车容易与前进方向上的障碍物碰撞导致脱轨,从而影响电机车的运行效率和矿山的正常生产.因此,能够精准地识别出机车前进方向上威胁机车正常行驶的障碍物,对提高电机车的运行效率、保障作业人员的人身安全具有重要意义.ConvNext模型特征提取效果好,模型的训练速度较快,目标检测的检出率高,能达到预期效果.故本文尝试采用ConvNext来对电机车行驶过程中的障碍物进行检测实验.实验结果与预期效果相同,ConvNext算法的检测精度符合要求,可精准检测机车运行各种复杂环境下的障碍物,mAP可达到 87.5%.
Abstract
Mine electric locomotive is the main equipment for mine transportation.Because of the complex underground operating environment and limited lighting conditions,unmanned electric locomotive is easy to collide with obstacles in the forward direction and cause derailment,thus affecting the operating efficiency of electric locomotive and the normal production of mine.Therefore,it is of great significance to accurately identify the obstacles that threaten the normal running of the locomotive,which is of great significance to improve the operating efficiency of the electric locomotive and ensure the personal safety of the operators.ConvNext model has good feature extraction effect,fast training speed and high detection rate of target detection,which can achieve the expected results.Therefore,this paper attempts to use ConvNext to detect obstacles in the process of electric locomotive driving.The experimental results are the same as exp ected,and the detection accuracy of ConvNext algorithm meets the requirements,which can accurately detect obstacles in various complex environments when locomotives are running,and the mAP can reach 87.5%.
关键词
矿用电机车/障碍物检测/ConvNext/深度学习Key words
mine electric locomotive/obstacle detection/ConvNext/dseep learning引用本文复制引用
出版年
2024
NETL
NSTL
万方数据