Range–Doppler Spectrum Anomaly Detection for Sky-Wave Over-the-Horizon Radar Based on Image Recognition and Data Augmentation

Sky-wave over-the-horizon radar (OTHR) receives various kinds of environmental signals, such as sea-land clutter (SLC), external interference, and ionospheric clutter. OTHR detects targets in the range-Doppler spectrum (RDS) map, where SLC should be present, while others are considered unwanted. If SLC is absent or if unwanted interference is severe enough to obscure targets, we refer to this as an RDS anomaly regarding target detection. This paper addresses automatic detection of RDS anomalies and proposes an RDS anomaly detector (RDSAD) based on image recognition techniques. To develop the image classifier, real OTHR data are utilized to construct the database. However, the available data are imbalanced and scarce, thereby necessitating data augmentation. Noise and interference data are expanded via signal modeling, while SLC and clutter are synthesized using a generative adversarial network (GAN) designed in this paper. A lightweight ResNet18 model is developed as the RDS image classifier. The RDSAD evaluates the classification results to determine RDS anomalies. It can serve as an auxiliary module in the OTHR system, with the ability to issue warnings and offer suggestions to operators, thereby alleviating the burden of manual monitoring. Experimental results of real data and literature images demonstrate the effectiveness of our design.