Monotone Piecewise-Linear Harmonization of Multitemporal Multispectral Surface Reflectance With Unsupervised Deep Multiview Pseudoinvariant Features Learning

Key points

• Surface reflectance products derived from optical satellite may exhibit systematic inconsistencies across acquisition dates, because atmospheric correction is generally performed independently for each image.1
• Such residual biases distort the temporal behavior of invariant ground targets and reduce the reliability of multitemporal surface reflectance analysis.1
• The framework learns deep invariant features (DIFs) directly from pixel-weighted canonical variates via iterative reweighting strategy, while a reconstruction regularizer preserves surface reflectance information and prevents degenerate collapse from suppressing genuine temporal variability.1
• These findings demonstrate the potential of deep invariant feature learning for multitemporal surface reflectance bias removal while highlighting the importance of preserving physically meaningful temporal and spectral information during nonlinear representation learning.1

Sources (1)

• [1]Monotone Piecewise-Linear Harmonization of Multitemporal Multispectral Surface Reflectance With Unsupervised Deep Multiview Pseudoinvariant Features Learning
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 15, 03:37 AM

  This study proposes Deep Multitemporal Generalized Canonical Correlation Analysis (DeepMGCCA), an unsupervised deep multiview pseudoinvariant feature learning framework that integrates Deep Generalized Canonical Correlation Analysis (GCCA) with Multivariate Alteration Detection (MAD) to identify deep-pseudoinvariant features (DIFs) for the removal of surface reflectance biases across multitemporal

  Surface reflectance products derived from optical satellite may exhibit systematic inconsistencies across acquisition dates, because atmospheric correction is generally performed independently for each image.

Extractive summary: sentences quoted from the sources.