Examining The Effect of Different Networks on Foreign Object Debris Detection

Abstract

Foreign Object Debris (FOD) at airports poses a risk to aircraft and passenger safety. FOD can seriously harm aircraft engines and injure personnel. Accurate and careful FOD detection is of great importance for a safe flight. According to the FAA's report, FOD types are aircraft fasteners such as nut, safety; aircraft parts such as fuel blast, landing gear parts, rubber parts; construction materials such as wooden pieces, stones; plastic materials, natural plant and animal parts. For this purpose, in this study, the effect of different networks and optimizer on object detection and accuracy analysis were examined by using a data set of possible materials at the airport. AlexNet, Resnet18 and Squeezenet networks were used. Application is applied two stages. The first one, 3000 data were divided into two parts, 70% to 30%, training and test data, and the results were obtained. The second one, 3000 data were used for training, except for the training data, 440 data were used for validation. Also, for each application, both SGDM and ADAM optimizer are used. The best result is obtained from ADAM optimizer with Resnet18, accuracy rate is %99,56.