Undergraduate research: Cyclostationary plot classification using machine learning

Abstract

Undergraduate students at WVUTech collaborated with Greenbank Observatory, WV to develop models to automatically identify the astronomical data containing radio frequency inference and noise. Normally the identification of RFI signals would be done manually by observing the images built from the data and would take a lot of manual time for processing. A machine learning model could automate the process and save a lot of time in such identification as well as the model can be used in real-time identification. The project was done over the course of two semesters with students from Computer Science majors. Students developed a convolutional neural network model that can run on GPU and obtain an accuracy of more than 90% after training on the available data. This paper discusses in detail the research work done and the results obtained by the students. Students used concepts of image processing and machine learning to develop the model.