Detecting Low-Slow-Small Unmanned Aerial Vehicles (LSS UAVs) through radio frequency (RF) signal analysis offers significant advantages in terms of operational concealment and cost-efficiency. However, the practical implementation of such detection systems faces substantial challenges due to the massive data volumes and complex processing requirements inherent in drone RF signal analysis. To address these limitations, this paper proposes an innovative framework that combines advanced signal processing with transfer learning-enhanced deep neural networks. Our methodology involves three key components: (1) RF signal acquisition using Universal Software Radio Peripheral (USRP) devices, (2) conversion of raw signals into time-frequency representations (TFRs) through short-time Fourier transform, and (3) development of a knowledge-transfer-based deep learning architecture specifically optimized for scarce training data scenarios. Experimental validation demonstrates that our approach achieves about 88% detection accuracy with only 2 training samples per class, significantly outperforming the baseline model without pre-trained weights in low-data regimes, making it particularly suitable for practical field deployments where training is conducted under computationally restricted conditions.