机电工程技术2025，Vol.54Issue(6)：34-39,6.DOI:10.3969/j.issn.1009-9492.2025.06.007
基于改进MobileNetV3的钢轨表面伤损识别模型
Improved Model for Rail Surface Damage Recognition Based on MobileNetV3
摘要
Abstract
Aiming to address the issues of insufficient accuracy and slow model convergence in the detection of rail surface damage,a high-performance lightweight model for recognizing surface damage on steel rails is proposed.By introducing a channel attention(CA)module containing spatial coordinate information,the precision of feature extraction and the generalization ability of the model are improved.The improved MobileNet V3 network is utilized as the backbone network to achieve model lightweight and efficiency.To verify the effectiveness,a dataset for rail surface damage is created.Experimental results show that on the rail surface damage dataset constructed,the initial recognition accuracy of MobileNet V3 is only 91.8%with an F1 score of 91.5%;the improved model increases the recognition accuracy and F1 score to 93.8%and 93.6%respectively;with a parameter quantity and time consumption of only 7.01×106,significantly less than other models.The improved MobileNet V3 model can effectively recognize rail surface damage,greatly reduce model parameters,improve detection speed,and provide an efficient means for detecting rail surface damage.关键词
改进MobileNetV3/CA模块/钢轨表面伤损/轻量化Key words
improved MobileNetV3/CA module/steel rail surface damage/lightweight分类
计算机与自动化引用本文复制引用
郭睿,姜云龙,宁善平..基于改进MobileNetV3的钢轨表面伤损识别模型[J].机电工程技术,2025,54(6):34-39,6.基金项目
2024年广东省科技创新战略专项资金(pdjh2024b573) （pdjh2024b573）
