铁路物流2025，Vol.43Issue(3)：15-24,10.DOI:10.16669/j.cnki.issn.2097-5899.202410120001
基于场景约束的铁路货车装载状态图像样本生成技术研究
Scene-Constrained Image Sample Generation Technology for Railway Wagon Loading Status
摘要
Abstract
Railway wagon loading status images are critical carriers of information reflecting the operational status of freight wagons,and the intelligence of these images directly impacts both train operation and cargo safety.Due to the impacts of cargo types,wagon types,and methods for loading safety,certain scenarios suffer from a scarcity of wagon loading status images,hindering the construction of models for object detection at inspection items in these contexts.Considering the characteristics of degraded railway wagon loading status images,this study proposed a multi-technology fusion model for generating railway wagon loading status images to satisfy the requirements of railway freight inspection,integrating several methods with a generative adversarial network(GAN)and instance segmentation.Specifically,the model incorporated generative and adversarial-based constraints,an object extraction method of foreground items,and a heuristic method for object placement at inspection items.This approach enriched the railway wagon loading status image database for special scenarios and provided technical support for constructing reliable and intelligent detection models.关键词
铁路货运/货检/货车装载状态/图像生成/目标检测/生成对抗网络Key words
Railway Freight/Cargo Inspection/Wagon Loading Status/Image Generation/Object Detection/Generative Adversarial Network分类
交通运输引用本文复制引用
葛悦,王志敬,柯向喜,孙文桥,王小卫..基于场景约束的铁路货车装载状态图像样本生成技术研究[J].铁路物流,2025,43(3):15-24,10.基金项目
中国国家铁路集团有限公司科技研究开发计划课题(J2024X004) （J2024X004）
中国铁道科学研究院集团有限公司科研项目(2024YJ152) （2024YJ152）
中国铁道科学研究院集团有限公司运输及经济研究所自主创新研究基金项目(2023YJS06) （2023YJS06）
