Deep Learning for SAR: Beyond Hype, Toward Sensor-Aware Models and Practical Challenges

On the FundaYao Podcast, we interview Michael Schmitt (Professor of Earth Observation, University of the Armed Forces, Germany) and Ronny Hänsch (DLR Microwave and Radar Institute) about their book “Deep Learning for Synthetic Aperture Remote Sensing.” They argue deep learning for SAR is not overhyped, enabling capabilities beyond classic signal processing while complementing, not replacing, traditional methods. The book aims to cover the full SAR processing chain and bridge gaps between data scientists lacking SAR expertise and SAR experts lacking modern AI skills. They discuss the niche nature of deep learning for SAR, warn against treating SAR as simple grayscale images, and advocate sensor-aware modeling that respects SAR physics and geometry. They highlight labeling and compute bottlenecks, as well as the industry–academia compute gap.

00:00 Welcome and Guests

00:30 Is Deep Learning Overhyped

02:21 Why This Book Exists

03:20 Who the Book Is For

06:14 Editing and Writing Challenges

10:13 Why SAR Deep Learning Is Niche

11:41 Doing SAR Deep Learning Right

14:47 Labels and Data Bottlenecks

18:46 Sensor Aware Deep Learning

21:02 Compute Limits and Access

27:26 What Counts as Foundation Model

32:02 Skepticism and SAR Backbones

35:13 How to Learn Deep Learning

36:19 Closing Thanks