Low-altitude drone detection method based on environmental antiinterference

Abstract:
In order to achieve real-time detection and rapid warning of low, small and slow UAV targets in severe weather, in this paper we combine multiple cascaded autoencoders, with the attention mechanism, to remove rain streaks in the input images by using multi-stage information exchange and feature fusion. We use ResNet as feature extraction network, to quickly obtain the category and position information of the UAV by predicting the target center point, so as to realize the real-time early warning for small drone targets. The experimental results show that the average recognition accuracy of the proposed algorithm can reach 82% under the heavy rain weather, and the calculation speed can reach 24 frames per second, which can effectively prevent the penetration of non-cooperative UAVs in the airspace.