基于改进YOLOX的机场场面飞机目标检测

Airport Surface Aircraft Object Detection Based on Improved YOLOX

机场场面飞机实时监控是远程塔台系统的基础。为实现对机场场面飞机目标快速而准确的检测,提出一种基于YOLOX融合注意力机制的机场场面飞机目标检测方法。在加强特征提取网络中引入卷积块注意力模块,增大对飞机目标空间位置和特征的关注度,同时利用CIoU方法计算目标框回归损失函数,并基于Tensorflow深度学习框架对YOLOX及改进模型开展对比实验。结果表明,YOLOX模型具有较高的检测精度与速度,提出的YOLOX-CT与YOLOX-CS模型的mAP0.5分别达到97.34%及97.28%,FPS值达到46及35。基于YOLOX的改进模型对飞机目标具有较高的检测效率,可保障机场运行安全、提升运行效率。

The real-time monitoring of aircraft on the airport surface is the basis of the remote tower system.In order to achieve fast and accurate detection of aircraft on the airport surface,a method for aircraft object detection on the airport surface based on YOLOX fusion attention mechanism is proposed.The Convolutional Block Attention Module was introduced into the enhanced feature extraction network to increase the attention to the spatial position and features of the aircraft target.At the same time,the Complete Intersection over Union method was used to calculate the regression loss function of the detection frame,and a comparative experiment was carried out on YOLOX and the improved model based on the Tensorflow.The results show that the YOLOX model has high detection accuracy and speed.The mAP0.5 of the proposed YOLOX-CT and YOLOX-CS models reach 97.34%and 97.28%,respectively,and the FPS reach 46 and 35.The improved model based on YOLOX has high efficiency for aircraft object detection,which can ensure the safety of airport operation and improve operation efficiency.