A Fast Cross-Band Spectrum Anomaly Detection Algorithm Based on Meta-Learning

Spectrum anomaly detection is an important research topic in cognitive radio. It can detect anomalies by predicting differences between the actual data. Existing deep learning-based spectral anomaly detection algorithms use a lot of training data. Due to the difference in frequency band, the detection model cannot be used directly across frequency bands. In order to solve this problem, a cross-band spectral anomaly detection method based on meta-learning is studied in this paper. Firstly, the data of different frequency bands are analyzed by using the pre-training of InceptionV3 to clarify the differences between different frequency bands. Secondly, a meta-learning data set is constructed and the optimal distribution of model parameters is found through the meta-learning training model. Finally, a small amount of target band data is used to fine-tune the model to detect anomalies in the target band. The experimental findings suggest that the proposed method is more stable than transfer learning and can detect cross-band anomalies with less target band data.