Research On Detection Method Of Foreign Object Debris In Airport Runway Based On Deep Learning

The existence of foreign object debris(FOD)on airport runway may lead to aircraft damage and endanger the lives of crew and passengers.Various detection technologies are used in conjunction with intelligent detection systems to identify FOD targets.The target detection algorithm based on deep learning has good detection effect on the common data set.However,specific to FOD target detection,due to the particularity of its shape and size,the commonly used deep learning target detection methods have low detection accuracy and high missed detection rate.In addition,the amount of airport runway images at night or extreme weather conditions is too limited to meet the requirements of deep learning,results in poor detection effect.Therefore,it is necessary to further analyze and study the FOD target detection algorithm in the case of multiple small-size targets and low-illumination scenes.The convolutional neural networks(CNN)method and the generative adversarial networks(GAN)method are combined in the work of FOD target detection.YOLOv3 algorithm is improved to detect the small-size target.Cycle GAN algorithm is used to improve the detection effect in low-illumination scenes.Firstly,the FOD target detection data set is constructed by collecting sample data from actual airport pavement and simulated airport pavement of normal illumination scene and low-illumination scene.Secondly,the architecture of FOD target detection algorithm based on CNN is designed.Aiming at the small-size targets,the network structure,clustering method and loss function of YOLOv3 algorithm are improved.Finally,a method of FOD data set expansion based on GAN is proposed.The data of low illumination scene is augmented by Cycle GAN algorithm,so as to solve the problem of low detection accuracy caused by insufficient training samples in special scenes.The experimental results show that the improved YOLOv3 has an accuracy of 95.3%under normal conditions,effectively solves the problem of missed detection.After the Cycle GAN algorithm is applied to expand the data set,the average accuracy of the algorithm in low-illuminance scenes is improved from 65.6% to 83.4%,which is significantly better than the existing algorithm.Besides that,a mobile FOD detection experimental platform is designed to realize the complete work from algorithm research to practical application.