Flying Drone Classification based on Visualization of Acoustic Signals with Deep Neural Networks
Soonyong Song, Young-Sung Son, Youngil Kim · 2020
In this paper, we proposed acoustic signal visualization for flying drone classifications and provided performance benchmarks for backbone deep neural networks. To visualize acoustic signals, we transformed the signals to 3-channel images by spectrogram. We put the images into deep neural networks to train their weights by transfer learning. We evaluated our classifiers by accuracy, recall, precision, and F1 score as well as loss. In our experiments, we accomplished maximum 99% drone classification performance in terms of accuracy with our dataset.