Research And Application Of Foreign Object Debris Detection In Airport Runway Based On Convolution Neural Network

With the rapid economic development,more and more people choose to travel by air,and the proportion of air transportation in the transportation field is gradually increasing.Because the aircraft is very fast in the process of taking off and landing,foreign object debris on the runway will pose a serious threat to the safety of the aircraft,it is very necessary to clean up the runway in time.The traditional method is manual inspection.With the development of technology,radar-based detection technology and optical-based detection technology have emerged.In recent years,the object detection technology based on deep learning has made great progress.Aiming at the problem of detecting foreign object debris in airport runways,this paper has carried out the research on the real-time detection algorithm of foreign object debris in airport runways based on deep learning.The main contents are as follows:Aiming at the practicality of the airport runway foreign object detection algorithm,this paper improves the algorithm on the basis of the current single-stage object detection algorithm research.Drawing lessons from general object detection,the feature aggregation module is designed to increase the amount of information in the feature map,and to predict on multi-level feature maps to improve the network’s detection effect on FOD data with many small objects;a hybrid detection head is designed,Improve the network’s recall rate of multi-scale object detection;with a suitable data enhancement strategy and a loss function that is more suitable for FOD data sets,it effectively improves the performance of the FOD detection network.A series of experiments were carried out on the FOD data set,which significantly improved the performance of the single-stage FOD detection network.Aiming at the lightweight problem of the FOD detection model,a pruning algorithm of the FOD detection network is designed to compress the parameters of the FOD detection model,which reduces the requirements of the FOD model on the processor and makes the model more convenient to deploy.Through the convolutional neural network pruning method,the FOD detection network is pruned at the convolutional channel and the convolutional layer;Andbased on the convolutional channel pruning,a variety of pruning strategies are designed to achieve In order to achieve better results.At the cost of very little performance,the amount of parameters is compressed by about 90%.Considering both the performance and lightweight of the FOD detection model,the model is further optimized on the basis of the pruned model to improve the performance of the model.On the basis of the original network structure,a cross-level local network module is introduced,which improves the basic components of the original network and improves the learning ability of the network.The focus module is designed to reduce the amount of calculation for the initial input of the convolutional neural network.The spatial pyramid pooling module is introduced to further improve the network’s detection performance for multi-scale objects.In the inference process of the network,the convolutional layer and the batch normalization layer of the convolutional neural network are merged,which improves the inference speed of the network.In the final experimental results,while maintaining a high inference speed,the performance is still improved by 3% compared to the unlightened network.