基于半监督学习的煤矿井下行人检测模型

Pedestrian Detection Model in Underground Coal Mine Based on Semisupervised Learning

在井工煤矿辅助运输车辆无人驾驶及安全监测领域,行人检测技术至关重要。目前已有许多工作进行了相应研究,但这些研究均需要使用大量的精确标注图片,而煤矿井下图片获取较为困难,标注难度也十分大,这些因素都极大地影响了相关模型的落地应用以及推广。为了解决这一问题,提出了一种基于半监督学习的煤矿井下行人检测模型。通过使用半监督学习框架可以有效降低煤矿井下行人检测模型对于高质量标注数据的大量需求。此外,针对煤矿井下设备运算能力较低的特点,还对YOLOv5模型进行了改进,在维持模型检测精度的条件下提升了模型的检测速度。实验表明基于半监督学习的煤矿井下行人检测模型可以使用仅相当于原数据集5%的数据训练得到较为有效的检测模型,大幅度较少了对于标注数据的依赖,对煤矿井下行人检测模型的快速应用和推广起到了帮助作用。

Pedestrian detection technology is crucial in the field of unmanned and safety monitoring of underground coal mine auxiliary transport vehicles.Many studies have been conducted,but these studies require the use of a large number of accurately labeled images,which are difficult to obtain and difficult to label in underground coal mines,and these factors greatly affect the application and promotion of relevant models.In order to solve this problem,proposes a semi-supervised learning-based pedestrian detection model for underground coal mines.By using a semi-supervised learning framework,the need for high quality labeled data for underground pedestrian detection models can be effectively reduced.In addition,the YOLOv5 model for the low computing power of underground equipment in coal mines has also been improved,which improves the detection speed of the model while maintaining the detection accuracy.The experiments show that the semi-supervised learning-based pedestrian detection model can be trained with only 5%of the original dataset to obtain a more effective detection model,which significantly reduces the reliance on labeled data and helps the rapid application and promotion of the pedestrian detection model in underground coal mines.