Research On Airport Runway FOD Detection Algorithm Based On Convolutional Neural Network

With the development of aviation,more and more attention has been paid to aviation safety,and the detection of Foreign Object Debris(FOD)on airport runways has become a key research issue in recent years.Although the typical airport runway FOD detection system use an optical camera to obtain the image of the suspicious area,it still needs to be identified by the staff.Therefore,further research on FOD detection algorithms that can automatically identify targets is of great significance in reducing labor costs and improving the detection stability of the system.Based on the existing research results in the field of computer vision,in this thesis we improves the general object detection model YOLOv3 based on the convolutional neural network according to the characteristics of FOD,to study more effective FOD detection algorithms for airport runways.The main research contents and innovations of this thesis are as follows:1.Due to the small size of the airport runway FOD,the object itself contains relatively few salient features,and the abstract features corresponding to the object in the middle layer of convolutional neural network are easily affected by the background information,which increases the difficulty of feature extraction and utilization in object detection model.Therefore,in the third chapter,an airport FOD detection algorithm based on attention convolutional neural network is proposed.First,the spatial attention sub-module and the channel attention sub-module are designed to generate attention maps along different dimensions,the sub-modules are merged and combined with shortcut connection of the residual network to form a feature extraction attention module,which is embedded in the downsampling position of the YOLOv3 backbone network to make it focus more on the object features in the feature extraction process.Afterwards,a feature fusion attention module is constructed using two channel attention branches,and this module is added to the feature fusion part of YOLOv3,and the fusion feature map is weighted and adjusted according to the channel characteristics between the two feature maps in different scale.Experiments show that the proposed method improves the detection performance of FOD by using attention mechanism.2.Because the airport runway FOD objects are foreign object in the scene,there is no fixed rule in the location and angle of the objects,which leads to a larger difference between the objects of the same category and limits the generalization of the objects detection model in the application process.For this reason,in the fourth chapter,an airport FOD detection algorithm based on siamese convolutional neural network is proposed.Considering that the detection area of fixed-mounted optical camera is also relatively fixed,first,we use the structure of YOLOv3 as a basic branch to construct a siamese features extract and fuse network,and use the image without a object as the reference image and put it into network with the image to be detected for feature extraction and feature fusion,and predict the category of objects based on the features of the image to be detected.Then we construct a similarity-based region proposal network,put the feature maps at the same scale of the two images into it,and predict coordinates of objects and the confidence of the corresponding regions based on the characteristics of the image to be detected and the similarity between the two images.Experiments show that the proposed method improves the detection performance of the model for FOD by comprehensively using the information of two images.