Guest Lecture on Utilizing Transfer Learning in SAR Imagery Target Classification

 

Speaker:   Mr. Richard Al-Bayaty, EE Ph.D. Candidate (University of Florida) and DoD SMART

 
Sponsored by:   SPAWAR-Atlantic



Abstract:

 

In performing classification tasks within the SAR imagery medium, properly training classification algorithms with ample amounts of data is paramount. When faced with situations where little (or no) training data is available for a particular target, the use of transfer learning can mitigate the need for supplemental training data. Incorporating this method allows the use of corpuses of synthetic data in lieu of real SAR imagery of the target. This talk will discuss the advantages of using SAR for remote sensing, introduce image manifolds and why subspace projections are important, highlight relevant image processing applications, and explain the concepts of Transfer Learning used to aid in classification tasks within this research. (NOTE: Have no fear, everything will be covered at the highest level possible, so no math knowledge is required nor expected. To get your hands dirty with the mathematical framework, feel free to email ralbayaty@ufl.edu or ask questions after the talk.)