CNN Based Image Classification of Malicious UAVs

Unmanned Aerial Vehicles (UAVs) or drones have found a wide range of useful applications in society over the past few years, but there has also been growth use UAVs for malicious purposes. One way to manage this issue is allow reporting (e.g., through smartphone application) with report including photo UAV. It would be able automatically identify type UAV within image terms manufacturer and specific product identification using trained classification model. In paper, we discuss collection images three popular at different elevations distances from observer, camera zoom levels. We then train 4 models based upon Convolutional Neural Networks (CNNs) dataset concept transfer learning well-known ImageNet database. The can classify contained unseen test up approximately 81% accuracy (for Resnet-18 model), even though 2 represented are visually similar, fact that contains significant distance observer. This provides motivation expand study future include more types other usage scenarios carrying loads).