Research Of Deep Learning Based Airport Runway Foreign Object Debris Detection And Recognition

As a fast and efficient mode of transportation,air transportation plays an important role in the field of transportation.More and more airports are being built or renovated.The safety of air transport has always been a topic of intense concern.Because the speed of the aircraft’s take-off and landing process is very fast,any foreign objects on the runway may pose a serious threat to the aircraft’s take-off and landing safety.It is an important link to ensure the safety of the aircraft’s take-off and landing in time to detect and clean up the foreign objects on the airport runway.The traditional airport runway foreign object detection method relies on manual walking inspection.This method is not only inefficient,but also difficult to detect new foreign objects in time,resulting in a large safety hazard.Therefore,a highly intelligent automatic detection system is urgently needed.In response to this demand,this paper carries out research work on deep-learning detection and identification technology of airport runway foreign objects,including:An airport runway foreign object detection dataset with multi-attribute semantic tags was designed and constructed.In view of the current lack of public airport runoff detection dataset,the data for the airport runway foreign object detection was designed and constructed according to the data distribution of the actual scene of the airport runway and the technical characteristics of deep learning.The scene of the airport runway under real environment is simulated to the maximum extent.The multi-attribute semantic label is defined by the labeling criteria of the universal dataset.The validity of the dataset is tested and verified by a typical target detection network.The result shows that the data is The set can better achieve the training of the target detection network and achieve better detection results.Based on the established dataset,the algorithm of foreign object detection and multi-attribute semantic recognition based on feature fusion is designed and implemented.In view of the characteristics of different shapes and small targets of airport runways,the feature fusion module is designed to make full use of different receptive field features to achieve effective detection of multi-scale targets.At the same time,according to the identification requirements of target hazard level and material identification in the actual scene,a multi-attribute semantic recognition branch is designed for the network,and the multi-attribute identification of the target is realized while detecting and locating.The experimental results show that the algorithm can accurately detect and identify foreign objects in various types of airport runways,especially for targets with high risk levels,so it has more important significance in practical applications.In order to further improve the ability to identify foreign materials in airport runways,a material identification algorithm based on augmentation of network data was designed.The network adopts the neural network training idea that minimizes the experience.By countering the augmentation of sample data,the classification of the residual convolutional neural network based on the feature attention mechanism is improved,and the classification performance is improved.The effect of existing algorithms.