基于深度学习的座舱开关状态识别研究

Cockpit Switch State Recognition Based on Deep Learning

飞机飞行前为保证飞行安全,需要机务人员对座舱开关、安全销等状态进行检查确认,由于检查对象多、状态多,人为检查常常出现错检、漏检。因此,以飞机座舱开关为具体研究对象,提出了一种基于深度学习的座舱开关状态识别方法。在典型的目标检测算法Faster R-CNN的基础上,提出基于特征融合的多分支Faster R-CNN改进算法,既提升了小目标开关的检测精度,又兼顾了一般大小目标物体的检测精度。多组对比实验表明,所提出的方法在座舱开关状态检测场景中的平均精度较原始的Faster R-CNN有明显提升。

In order to ensure flight safety,aircraft flight personnel need to check and confirm the status of cockpit switch and safety pin,etc.Due to the large number of objects and states under inspection,artificial inspection often occurs wrong inspection and missed inspection.Therefore,a new method of cabin switch state identification based on deep learning is proposed,which takes the cockpit switch as the research object.Based on the typical target detection algorithm Faster R-CNN,an improved multi-branch Faster R-CNN algorithm based on feature fusion is proposed,which not only improves the detection accuracy of small target switch,but also gives consideration to the detection accuracy of general object size.The comparison experiments show that the average accuracy of the proposed method in the cockpit switch state detection scene is significantly higher than that of the original Faster R-CNN.