Knowledge-Aware Mamba for Joint Change Detection and Classification from MODIS Times Series

2025-10-08

AI Summary

To address accuracy bottlenecks in MODIS time-series change detection—arising from mixed pixels, spatiotemporal-spectral coupling, and land-cover heterogeneity—this paper proposes a knowledge-enhanced multi-task learning framework. Methodologically, it introduces a knowledge-aware transfer loss mechanism and a triple-loss function (PreC/PostC/Chg), constructs a Spatial-Spectral-Temporal Mamba (SSTMamba) module, and incorporates a Sparse Deformable Mamba (SDMamba) backbone to model long-term temporal dependencies while substantially reducing computational cost. Notably, this is the first work to jointly optimize change detection and land-use/land-cover (LULC) classification. Evaluated on a MODIS dataset from Saskatchewan, Canada, the method achieves 1.5–6% improvement in mean F1-score for change detection and approximately 2% gains in overall accuracy (OA), average accuracy (AA), and Kappa coefficient for LULC classification.

Technology Category

• Machine Learning: Multimodal Learning
• Computer Vision: Multi-modal Vision
• Data Mining & Knowledge Management: Mining of Visual, Multimedia & Multimodal Data

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Although change detection using MODIS time series is critical for environmental monitoring, it is a highly challenging task due to key MODIS difficulties, e.g., mixed pixels, spatial-spectral-temporal information coupling effect, and background class heterogeneity. This paper presents a novel knowledge-aware Mamba (KAMamba) for enhanced MODIS change detection, with the following contributions. First, to leverage knowledge regarding class transitions, we design a novel knowledge-driven transition-matrix-guided approach, leading to a knowledge-aware transition loss (KAT-loss) that can enhance detection accuracies. Second, to improve model constraints, a multi-task learning approach is designed, where three losses, i.e., pre-change classification loss (PreC-loss), post-change classification loss (PostC-loss), and change detection loss (Chg-loss) are used for improve model learning. Third, to disentangle information coupling in MODIS time series, novel spatial-spectral-temporal Mamba (SSTMamba) modules are designed. Last, to improve Mamba model efficiency and remove computational cost, a sparse and deformable Mamba (SDMamba) backbone is used in SSTMamba. On the MODIS time-series dataset for Saskatchewan, Canada, we evaluate the method on land-cover change detection and LULC classification; results show about 1.5-6% gains in average F1 for change detection over baselines, and about 2% improvements in OA, AA, and Kappa for LULC classification.

Problem

• Detecting land-cover changes from MODIS time series data
• Addressing mixed pixels and information coupling in satellite imagery
• Improving classification accuracy and computational efficiency in change detection

Innovation

• Knowledge-driven transition loss enhances detection accuracy
• Multi-task learning with three losses improves model constraints
• Spatial-spectral-temporal Mamba modules disentangle information coupling