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

Foreign Object Debris(FOD)on airport runways pose a threat to aviation safety,and cause safety problems for passengers and economic losses for airports.The real-time detection and timely processing of FOD are of great significance to eliminate potential safety hazards on the ground and better protect the safety of aircraft take-off and landing,and the safety of passengers’ lives and property.At present,foreign object detection on runways at home and abroad mainly relies on manual detection,which has low efficiency and poor reliability.In response to this problem,this thesis studies the foreign object detection problem on the airport runway based on deep learning technology.The thesis starts from the difficulty of detecting small and medium objects by FOD,focuses on target detection,image classification and other technologies and methods,and conducts related theoretical and technical research.First,the research status of FOD detection system,detection technology and data set is analyzed,and the related methods and evaluation indicators of FOD detection are introduced;secondly,a FOD detection data set containing multi-semantic attribute labels is constructed;thirdly,a YOLOX-based The FOD detection model uses the inverted residual module for feature extraction,improves the Focal loss loss to increase the attention to difficult samples,and uses depth separable convolution to lighten the model;then,proposes FOD material recognition based on Mobile Net V2 The model is optimized with multi-level feature fusion and ECA attention mechanism,and uses transfer learning to save model training time and improve model performance;finally,a vehicle-mounted FOD detection and recognition system is designed to help airports manage.The thesis verifies the feasibility of the proposed method on the self-built data set.In terms of FOD detection,m AP@0.5 of the improved YOLOX model reached 85%,and m AP@0.5:0.95 reached 62.4%,which were respectively increased by 3.3% and 4.7%compared with the original model;in terms of FOD material identification,the improved MobileNetV2 The Top-1 Acc of the model is 94.75%,which is 1.13% higher than the original model;in terms of the FOD detection system,a vehicle-mounted solution from data collection to model deployment terminal equipment is designed,including target image collection,real-time video display,The functions of foreign object detection on the runway,data storage and management,and equipment management basically meet the needs of airport management and provide a new mode for the inspection and management of foreign objects on the runway.