摘要
为了快速准确地识别飞行员在驾驶舱内的异常行为,以保证航空安全,设计了一种基于改进YOLOv5算法的驾驶舱内飞行员异常行为识别方法。在YOLOv5的骨干网络中加入坐标注意力机制,获取在位置和方向上的特征信息,增强对注意力信息的敏感程度;改良交并比作为损失函数,提高模型计算速度和精度。训练自制飞行员异常行为原始数据集,实验结果表明,在模拟飞行驾驶舱中进行测试,能够准确快速识别飞行员的3种异常行为,平均精度达到98.3%,满足了识别要求。
In order to identify anomalous behavior of pilots in the cockpit accurately and quickly to ensure aviation safety,a method for anomalous behavior recognition of pilots in the cockpit based on improved YOLOv5 algorithm was designed.The coordinate attention mechanism was added to the backbone network of YOLOv5 to obtain feature information in position and direction,enhancing the sensitivity to attention information.The improved intersection over union was used as the loss function to improve the model′s calculation speed and accuracy.The self-made dataset of abnormal behaviors of pilots was trained,the experimental results show that the model can accurately and quickly recognize three types of anomalous behavior of pilots in simulated flight cockpits,with a mean average precision of 98.3%.This achieved the requirements for identifying anomalous behavior of pilots accurately and quickly.
作者
魏麟
谭任翔
何峻毅
彭俊榕
WEI Lin;TAN Ren-xiang;HE Jun-yi;PENG Jun-rong(Civil Aviation Flight University of China,Guanghan 618000,China)
出处
《航空计算技术》
2023年第6期20-24,共5页
Aeronautical Computing Technique
基金
民航飞行技术与飞行安全重点实验室自主研究项目资助(FZ2021ZZ05)。
