YOLOv4-tiny及其改进算法在航空机务维修照相管理中的应用

Application of YOLOv4-tiny and Its Improved Algorithm in Photographic Management of Aviation Maintenance

针对飞机机务维修照相管理存在工作量大、不精确等问题,提出一种利用深度学习YOLOv4-tiny算法来执行照片对比检测的方法。利用一个自制的数据集来训练网络模型,为解决开口销螺母及其他背景干扰,引入注意力机制模块以改进YOLOv4-tiny。测试结果表明:准确率(precision,P)相较原YOLOv4-tiny提高了5%,召回率(recall,R)提高约8%,平均准确率均值(mean average precision,mAP)提高了4.9%,照片识别精度和定位精准性方面都有较优表现,满足照相管理中对目标精准识别与比对的要求。

Aiming at the problems of heavy workload and inaccuracy in the photographic management of aircraft maintenance, this paper proposes a method of using deep learning YOLOv4-tiny algorithm to perform photo comparison detection. A self-made data set is used to train the network model. In order to solve the problem of cotter nut and other background interference, the attention mechanism module is introduced to improve YOLOv4-tiny. Test results show that the precision(P) is improved by 5%, the recall(R) is improved by about 8%, and the mean average precision(mAP) is increased by 4.9%. It has excellent performance in photo recognition accuracy and positioning accuracy, and meets the requirements of accurate target recognition and comparison in photo management.