An Application of Convolutional Neural Networks to Aircraft Ejection Seat Testing

The testing of aircraft ejection seats has been an important aspect of testing in the aerospace industry for several decades. Identification and tracking of aircraft ejection seat systems, during field testing, has practical benefits in post-test analysis and real-time tracking applications. This paper proposes a methodology to identify and locate ejection seat systems in a video stream as they are propelled from a dynamic aircraft test bed. The field testing environment is cluttered with natural and man-made artifacts that effect the background appearance and make the video stream data visually noisy. The identification problem involves the extraction of critical information from the video stream and feeding it into a deep learning convolutional neural network that determines the location of the ejection system within the field of view. The results achieved, using the convolutional neural network to identify features of the ejection seat system, were promising. In particular, the convolutional neural network demonstrated an ability to identify objects that were in a different spatial orientation (rotation) than were used to train the network.