The detection and classification of unmanned aerial vehicles (UAVs) is a strategic task for several military and civilian applications.The use of UAV RF signatures along with convolutional neural networks (CNNs) has proven to be an accurate approach for classification in positive signal-to-noise ratio (SNR) conditions.This work investigates the RF-CNN approach in negative SNR environments.It presents, first, a noise-resilient detection method based on spectral entropy drop.It then investigates the role of temporal resolution in preserving classification accuracy without explicitly training the CNN for any specific noise levels.Tests on a public data set showed an increased classification accuracy from ~32% to ~60 for an SNR of À10 dB by just prioritizing time over frequency resolution.