Polarization Aware Deep Learning Framework for SAR Despeckling
Organizations: Dipartimento di Ingegneria, Universit`a degli Studi di Napoli “Parthenope,” 80143 Naples, Italy · Dipartimento di Ingegneria, Universit`a degli Studi di Napoli “Parthenope,” 80143 Naples, Italy, and also with the National Inter-University Consortium for Telecommunications (CNIT), 80126 Naples, Italy · Department of Earth Observation Science, Faculty of Geo-Information Science and Earth Observation,University of Twente, Enschede 7514AE, The Netherlands · Dipartimento di Scienze e Tecnologie, Universit`a degli Studi di Napoli “Parthenope,” 80143 Naples, Italy, and also with the National Inter-University Consortium for Telecommunications (CNIT), 80126 Naples, Italy
Abstract
Speckle is an inherent characteristic of synthetic aperture radar (SAR) imagery that complicates image interpretation and the extraction of reliable information. Although deep-learning-based despeckling methods have demonstrated promising performance, their applicability across different polarimetric configurations is often limited by the need for specialized architectures or separate training procedures. This paper presents a unified deep-learning framework for despeckling single-, dual-, and full-polarization SAR data using a single trained convolutional neural network. The proposed approach adopts a common nine-channel representation of the polarimetric covariance matrix and a polarization-aware masking strategy to accommodate different configurations without modifying the network architecture. Qualitative and quantitative comparison on three different scenes from different sensors demonstrate competitive performance across the considered experiments and support the feasibility of a unified, polarization-flexible approach to SAR despeckling across different polarimetric configurations.
Figures & tables
| Sensor | Band | Center frequency | Area |
|---|---|---|---|
| ALOS PALSAR | L | 1.270 GHz | San Francisco (USA) |
| RADARSAT-2 | C | 5.405 GHz | Flevoland (NL) |
| BIOMASS | P | 0.435 GHz | Paranà Lake (PY) |
| Method | Single-pol | Dual-pol | Full-pol |
|---|---|---|---|
| MuLog [ 10 ] | |||
| MEET-SAR [ 17 ] | |||
| CL-SAR [ 22 ] | – | – | |
| CNN-PolSAR [ 23 ] | – | – | |
| DRUNet [ 24 ] | – | – | |
| PANDA |
| Method | KL-Tr | OVL-Tr | KL- max | OVL- max | |
| RADARSAT-2 | MEET-SAR | 1.3545 | 0.7364 | 11.3901 | 0.6079 |
| MuLog | 0.0293 | 0 .9282 | 0 .0836 | 0 .8761 | |
| DRUNet | 0.2073 | 0.7333 | 0.1152 | 0.7357 | |
| PANDA | 0 .0297 | 0.9539 | 0.0615 | 0.8824 | |
| BIOMASS | MEET-SAR | 0.4693 | 0.7332 | 0.9943 | 0.5769 |
| MuLog | 0.0766 | 0.8883 | 0.1131 | 0.8398 |
| Method | KL-Tr | OVL-Tr | KL- max | OVL- max | |
|---|---|---|---|---|---|
| RADARSAT-2 | MEET-SAR | 0.9571 | 0.8344 | 0.8611 | 0.8044 |
| MuLog | 0.0160 | 0.9348 | 0.0145 | 0.9254 | |
| CNN-PolSAR | 0.0090 | 0.9614 | 0.0115 | 0.9441 | |
| PANDA | 0.0079 | 0.9621 | 0.0079 | 0.9454 | |
| BIOMASS | MEET-SAR | 0.3855 | 0.8810 | 6.6587 | 0.5026 |
| MuLog | 0.0117 | 0.9481 | 0.5034 | 0.6745 |
| Method | M -index | MOR 1 | VOR 1 | MOI | |
|---|---|---|---|---|---|
| RADARSAT-2 | MEET-SAR | 15.3040 | 0.7947 | 0.4320 | 0.0352 |
| MuLog | 4.7503 | 0.9773 | 0.6071 | 0.1064 | |
| CL-SAR | 11.5405 | 0.9083 | 0.2790 | 0.1182 | |
| PANDA | 6.0333 | 0.9968 | 0.7655 | 0.1474 | |
| BIOMASS | MEET-SAR | 25.4885 | 0.8656 | 0.4733 | 0.0187 |
| MuLog | 5.8145 | 1.0059 | 1.0166 | 0.1390 |
Synthetic Aperture Radar
Synthetic Aperture Radar uses motion-enabled microwave measurements to image Earth and objects, even through clouds and at night.
See all 75 papers