摘要
文章以列车行驶前方障碍物检测为例,介绍了根据所搭建的轨道目标智能检测实验平台设计和改进深度神经网络模型,并将其应用于实际场景的做法。选择了ResNeXt主干特征提取网络,使模型的特征提取能力更强;采用了自适应特征融合优化方法和注意力机制,大幅度提升了算法在铁路环境中的检测性能。
Taking the obstacle detection in front of the train as a case, this paper introduces the design and improvement of the deep neural network model based on the intelligent test platform of the railway objects, and applies it to the actual scene. The backbone of ResNeXt is selected to make the model that owns a more powerful feature extraction ability, and the adaptive feature fusion method and attention mechanism are adopted in this model, which can greatly improve detection performance in railway environments.
作者
叶涛
赵宗扬
张晞
YE Tao;ZHAO Zongyang;ZHANG Xi(School of Mechanical Electronic&Information Engineering,China University of Mining&Technology(Beijing),Beijing 100083,China)
出处
《实验技术与管理》
CAS
北大核心
2021年第11期237-242,共6页
Experimental Technology and Management
基金
中央高校基本科研业务费专项资金项目(8000150A073)
机械工程测试技术课程教学方法研究(J210410)。
