A DTW-Based Feature Extraction Method for Frequency-Hopping Signal Classification with Data-Optimized CNN

The increasing demand for unmanned aerial vehicle(UAV) and the growing prominence of safety supervision issues highlight the need for effective monitoring methods, which can be achieved through frequency hopping signal monitoring. Focusing on the characteristics of UAV frequency hopping signals, this paper proposes a neural network classification system based on cycle matching dynamic time warping (DTW) and employs the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm for feature enhancement and noise removal. The oversampling algorithm is utilized to address class imbalance and achieve feature capture, recognition, and classification of UAV frequency hopping signals. Through multiple sets of comparative experiments, the results demonstrate a 10.54% improvement in feature extraction performance and a 2.04% enhancement in classification accuracy compared to the baseline model. The system achieves an accuracy rate of 93% at signal-to-noise ratios (SNR) above 0 dB, demonstrating its effectiveness across multiple aspects.