An automatic data cleaning method for radar emitter signal recognition

With the rapid advancement of modern electronic warfare, the volume of signals intercepted by electronic reconnaissance equipment is increasing exponentially, yet the quality of these signals remains inconsistent. Low-quality signals hinder radar signal recognition tasks, particularly those employing deep learning algorithms. Despite this challenge, there has been limited research focused on cleaning radar emitter signal data to enhance the quality of radar signal datasets. To address this issue, we propose a data cleaning method that automatically detects and removes low-quality signal samples from radar emitter signal datasets. By analyzing the characteristics of the received radar signals, we extract resemblance coefficients from their envelopes, effectively distinguishing noise signals from pulse signals in the feature space. We then apply the Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm to cluster different types of signals. This approach enables the automatic detection and removal of noise data from the dataset. Extensive experiments on both simulated and experimental data demonstrate that the proposed method significantly improves data quality and enhances the recognition rate of deep learning networks for emitter recognition.