Chapter 9 - Earth observation image denoising

Abstract

Earth observation (EO) provides valuable information about the Earth's surface through passive (e.g., optical) and active (e.g., radar) sensing modalities. Image quality is critical for downstream analysis and applications, such as classification, segmentation, and change detection, yet EO imagery is frequently degraded by noise arising from sensor characteristics, acquisition geometry, and environmental conditions. Image denoising aims to recover an underlying clean signal from noisy observations and constitutes a fundamental preprocessing step in EO image analysis. In this chapter, we first formalize the image denoising problem and introduce common modeling and algorithmic frameworks for designing denoisers. We then review the principal noise types encountered in EO data, including modality-specific characteristics. State-of-the-art denoising approaches—ranging from classical filtering and model-based methods to statistical learning and deep learning techniques—are discussed and compared. An application-oriented case study illustrates the integration of denoising methods within supervised and unsupervised learning pipelines for EO imagery. The chapter concludes with best practices and a discussion of how current and emerging advances in artificial intelligence influence the design, evaluation, and deployment of EO denoising solutions.