摘要
近年来,因施工升降机标准节螺栓状态异常造成的事故频频发生,针对施工升降机螺栓状态异常问题,提出一种基于改进的YOLO v3算法的检测方法。首先,通过K-means++聚类算法优化先验框(anchor boxes)的尺寸,使其更加适合升降机的螺栓状态检测;其次,根据升降机螺栓体积较小的特点,将3个特征尺度改为2个特征尺度,再通过特征金字塔网络FPN(feature pyramid network)进行融合。经过实验证明,改进后算法的mAP(mean average precision,平均精度均值)由81.40%提升了4.54%,达到85.95%,检测速度由34帧/s,提升了6帧/s,达到40帧/s。通过利用无人机搭载摄像头,能够实时准确地检测出施工升降机的螺栓状态,进而减少因螺栓故障引起的安全事故,在建筑安全领域发挥重要作用。
In recent years, accidents frequently occur due to abnormal bolt status of standard joint of construction lifts. Aiming at solving the problem of abnormal bolt status of construction lifts, a detection method based on improved YOLO v3 algorithm was proposed. Firstly, the size of anchor boxes was optimized by K-means++ clustering algorithm to make them more suitable for bolt status detection of the lift. Secondly, according to the small volume of lift bolts, the three characteristic scales were changed into two characteristic scales, and then the fusion was carried out by feature pyramid network(FPN). Through experiments, the mAP of the improved algorithm was increased by 4.54% from 81.40% to 85.95%, and the detection speed was increased by 6 frames/s from 34 frames/s to 40 frames/s. By using a drone equipped with a camera, the bolt status of the construction lift could be accurately detected in real time, thereby reducing safety accidents caused by bolt failures, and playing an important role in construction safety.
作者
陈国栋
林愉翔
赵志峰
黄明炜
林进浔
CHEN Guodong;LIN Yuxiang;ZHAO Zhifeng;HUANG Mingwei;LIN Jinxun(School of Physics and Information Engineering,Fuzhou University,Fuzhou 350108,China;Fujian Shuboxun Information Technology Co.,Ltd.,Fuzhou 350002,China)
出处
《贵州大学学报（自然科学版）》
2022年第6期81-86,124,共7页
Journal of Guizhou University:Natural Sciences
基金
国家自然科学基金资助项目(61471124)
福建省科技计划引导性项目(2021H0013)
福建省科技型中小企业创新资金项目(2021C0019)。
