Temporal-Spatial Tubelet Embedding for Cloud-Robust MSI Reconstruction using MSI-SAR Fusion: A Multi-Head Self-Attention Video Vision Transformer Approach

2025-12-10

AI Summary

Cloud contamination severely distorts the spectral signatures of multispectral imagery (MSI), impeding reliable early-stage crop mapping. To address this, we propose a spatio-temporal tubular embedding framework that fuses synthetic aperture radar (SAR) and MSI for robust reconstruction. Departing from conventional vision transformer (ViT) approaches that employ coarse temporal aggregation, our method introduces a novel non-overlapping, short-term (t=2) 3D tubular slicing mechanism—preserving local temporal consistency while mitigating inter-day information decay. We further pioneer the adaptation of video vision transformers (ViViT) to MSI–SAR temporal fusion, integrating 3D convolutional feature extraction, cross-modal alignment, and multi-head self-attention fusion. Evaluated on the 2020 Traill County dataset, our method reduces mean squared error (MSE) by 2.23% over single-source MSI reconstruction; incorporating SAR yields an additional 10.33% relative improvement over the baseline, significantly enhancing the robustness of agricultural remote sensing under cloud occlusion.

Technology Category

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

Application Category

• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Representation, reconstruction, and subgraph or motif discovery in Web-related graphs
• Systems and Infrastructure for Web, Mobile and WoT: Applied ML and AI for Web-based mobile applications

Abstract

Cloud cover in multispectral imagery (MSI) significantly hinders early-season crop mapping by corrupting spectral information. Existing Vision Transformer(ViT)-based time-series reconstruction methods, like SMTS-ViT, often employ coarse temporal embeddings that aggregate entire sequences, causing substantial information loss and reducing reconstruction accuracy. To address these limitations, a Video Vision Transformer (ViViT)-based framework with temporal-spatial fusion embedding for MSI reconstruction in cloud-covered regions is proposed in this study. Non-overlapping tubelets are extracted via 3D convolution with constrained temporal span $(t=2)$, ensuring local temporal coherence while reducing cross-day information degradation. Both MSI-only and SAR-MSI fusion scenarios are considered during the experiments. Comprehensive experiments on 2020 Traill County data demonstrate notable performance improvements: MTS-ViViT achieves a 2.23% reduction in MSE compared to the MTS-ViT baseline, while SMTS-ViViT achieves a 10.33% improvement with SAR integration over the SMTS-ViT baseline. The proposed framework effectively enhances spectral reconstruction quality for robust agricultural monitoring.

Problem

• Reconstructs cloud-covered multispectral imagery for crop mapping
• Improves temporal embedding to reduce information loss in sequences
• Integrates SAR data to enhance reconstruction accuracy in agriculture

Innovation

• Video Vision Transformer with temporal-spatial tubelet embedding
• 3D convolution extracts non-overlapping tubelets for local coherence
• Multi-head self-attention fuses MSI and SAR data