• ISSN: 1674-7461
  • CN: 11-5823/TU
  • 主管:中国科学技术协会
  • 主办:中国图学学会
  • 承办:中国建筑科学研究院有限公司

基于深度学习的桥梁病害检测研究与应用

Research and Application of Bridge Disease Detection Based on Deep Learning

  • 摘要: 本文研发基于YOLOv5算法的桥梁病害检测系统,用于无人机桥梁病害检测。首先,通过无人机巡航系统进行病害采集;其次,通过K-means聚类算法对YOLOv5算法中的先验框比例进行改进,获取桥梁病害的尺寸;再次,利用图像切割算法将无人机输入图像切割为640*640尺寸大小,减小模型训练与推理难度,同时利用APAP算法对切割图像识别结果进行拼接,达到完整无人机图像识别结果;最后,采用Python Flask搭建开放式桥梁病害检测Web前端,可根据实际需求调用不同桥梁部位病害权重,实现13种桥梁病害实时检测。结果表明,该模型可有效地降低硬件需求,可直接对无人机图像进行识别处理,实现快速病害检测,有望为桥梁病害检测提供一个高效、安全且具有前景的检测方法。

     

    Abstract: In this paper, a bridge disease detection system based on YOLOv5 algorithm is proposed for UAV bridge disease detection. Firstly, diseases are collected through UAV cruise system. Secondly, K-means clustering algorithm is applied to improve the prior box proportion in YOLOv5 algorithm to obtain the size of bridge disease. Thirdly, the UAV input image is cut to 640 * 640 size by using the image cutting algorithm to reduce the difficulty of model training and reasoning. What's more, this paper uses APAP algorithm to splice the recognition results of cutting images to achieve a complete UAV image recognition result. Finally, the paper employs Python Flask to build an open bridge disease detection Web front-end, which can call the disease weights of different bridge parts according to actual needs, and further realize real-time detection of 13 bridge diseases. The results show that the model can effectively reduce the hardware requirements, directly identify and process UAV images, and achieve rapid disease detection, which is expected to provide an efficient, safe and promising detection method for bridge disease detection.

     

/

返回文章
返回