Radio Frequency Interference Detection in Microwave Radiometry Using Multi–Dimensional Semi Supervised Learning

Radio frequency interference (RFI) in the measurements of Earth-observing radiometers is increasing over time with the increase in frequency spectrum demand for active services [1] . RFI contamination may lead to erroneous retrieval of the critical geophysical parameters in passive remote sensing missions. To overcome this problem, several detection and mitigation algorithms have been proposed and implemented with only limited success in cases of weak or noise–like interference [2] . The existing algorithms can be divided into two categories based on the number of dimensions used in the detection process, i.e., single (e.g., pulse blanking [3] ) and multiple dimensions (e.g., SMAP radiometer [4] ) algorithms. It should be noted that the cumulative information from the multiple dimensions is proven to be more effective than the algorithms that rely on a single dimension [5] , [6] . Furthermore, the occurrence of RFI is diverse. Therefore, currently, there is no labeled baseline dataset is available for RFI–contaminated data. Taking these into consideration, in this work, we have proposed the one class support vector machine algorithm (OCSVM) to use the information from RFI–free measurements alone. Furthermore, we have performed an extensive feature analysis to differentiate between the RFI–contaminated and RFI–free measurements.