Research On Satellite Spectrum Signal Recognition Based On Computer Vision

As a technology to obtain satellite spectrum usage,satellite spectrum's recognition was divided into two parts: signal detection and anomaly recognition.In order to improve the performance of spectrum detection,a method based on computer vision and machine learning was proposed,proving the possibility of spectrum detection from the perspective of the image.A method based on digital image processing and density clustering algorithm was applied to satellite spectrum sensing.Firstly,the spectrogram is subjected to image filtering and pixel binarization.Secondly,the spectrogram is extracted with the coordinates of the effective pixel set.Finally,the pixel coordinates extracted from the spectrogram are median processed and then clustered by DBSCAN.The parameters of the signal in the spectrum can be obtained from the clustering results through parameter estimation.Simulation shows that the new method has got rid of the shortcomings of threshold sensitivity compared with the energy detection method,improved the accuracy of signal detection.In this paper,the method of target recognition was applied to the problem of anomaly detection,with the spectral anomalies in the satellite spectrum map were used to detect image anomalies and target detection.As the most mature target detection framework at present,YOLOV3 has been correspondingly adapted to perform spectrum abnormality identification.Through data amplification,algorithm modification and overfitting prevention,the detection accuracy of the target detection network is constantly improving anomalies.Finally,in the actual anomaly detection problem,the trained target detection network achieved 95.3% accuracy with a recall rate of 65.9%.