Low-orbit satellite signal interference detection based on fast regional convolutional network and its multidimensional evaluation method

Aiming at the two main sources of threats faced by low-orbit communication satellites, namely, hacker network attacks and spectrum interference, the study builds an adversarial sample detection model based on variational self-encoder and a fast region-based convolutional network based spectrum interference detection model. The outcomes indicated that the detection accuracy of the proposed adversarial sample detection model in the intrusion traffic detection task was 97.68% with an F1 value of 96.86%. The AUC values of the proposed adversarial sample detection model in different typical network attacks were above 95%. The proposed spectrum interference detection model had the best detection effect in single tone interference detection. Its detection accuracy, recall and evaluation precision were 98.65%, 96.21% and 93.14%, respectively. It converged at about 200 iterations with an average recognition accuracy of 95.47%. The experimental results demonstrate the antisample and spectrum interference detection performance of the proposed model. The findings support the continued provision of satellite communication services as well as the safe operation of low-orbit communication satellites.