The use of convolutional neural networks in the detection and classification of foreign object debris on runways causing damage to aircraft

In this research, the use of the RetinaNet model, with ResNet-50 as its backbone convolutional neural network, is proposed for the detection and classification of foreign object debris (FOD) on runways. This approach aims to complement human factors in combating FOD, which can cause damage to aircraft and threaten flight safety. The FOD-A dataset, providing labeled environmental conditions, was used to train the model. Test findings indicate that the accuracy and discriminability of detection and classification increased for larger objects, achieving high success. However, as the object size decreased, the accuracy and discriminability in detection and classification decreased, leading to a lower success rate.