Multi-Frequency and Multi-Component Sparse Near Field Scanning Based on Active Machine Learning

This letter presents an active machine learning method for multi-frequency and multi-component sparse near-field scanning (NFS), which significantly accelerates the NFS process with multiple frequencies and multiple field components. The proposed active learning algorithm iteratively evaluates the uncertainties of unvisited positions using the query-by-committee method and adaptively selects the most uncertain samples. The uncertainties are calculated by strategically considering and balancing the contributions of different frequencies and field components. Compared to a uniform-like sampling strategy called greedy sampling, the proposed approach demonstrates a more advantageous performance in achieving a higher accuracy using the same amount of sparse samples across almost all the frequencies and field components. Compared to conventional full scanning, the NFS efficiency is remarkably improved while high-quality near-field imaging is maintained.