Just Project! Multichannel Despeckling, the Easy Way

2024-08-21

IEEE Transactions on Geoscience and Remote Sensing

AI Summary

Addressing the challenges of speckle noise suppression in multi-channel synthetic aperture radar (SAR) imagery—namely, poor generalizability and reliance on clean ground-truth labels—this paper proposes MuChaPro, a generic despeckling framework. MuChaPro projects multi-channel SAR data into numerous single-channel signals, enabling parallel denoising via any pre-trained single-channel despeckling model; outputs are then fused through a learnable reconstruction module. This constitutes the first plug-and-play transfer of single-channel despeckling methods to multi-channel SAR. The framework supports sensor-adaptive self-supervised training, eliminating the need for clean reference labels. Evaluated on polarimetric classification and interferometric height estimation tasks, MuChaPro achieves superior despeckling performance: it preserves fine textural details while maintaining inter-channel coherence and exhibits high computational efficiency. Moreover, it is compatible with diverse state-of-the-art single-channel denoisers.

Technology Category

• Machine Learning: Multimodal Learning
• Computer Vision: Multi-modal Vision
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data

Abstract

Reducing speckle fluctuations in multichannel SAR images is essential in many applications of synthetic aperture radar (SAR) imaging such as polarimetric classification or interferometric height estimation. While single-channel despeckling has widely benefited from the application of deep learning techniques, extensions to multichannel SAR images are much more challenging. This article introduces MuChaPro, a generic framework that exploits existing single-channel despeckling methods. The key idea is to generate numerous single-channel projections, restore these projections, and recombine them into the final multichannel estimate. This simple approach is shown to be effective in polarimetric and/or interferometric modalities. A special appeal of MuChaPro is the possibility to apply a self-supervised training strategy to learn sensor-specific networks for single-channel despeckling.

Problem

• Reduces speckle noise in multi-channel SAR images
• Extends single-channel despeckling to multi-channel SAR
• Uses self-supervised training for sensor-specific networks

Innovation

• Multi-channel SAR despeckling via single-channel projections
• Self-supervised training for sensor-specific networks
• Generic framework combining restored single-channel estimates