In the search for optimal multi-view learning models for crop classification with global remote sensing data

This study systematically explores Multi-View Learning (MVL) configurations for global crop and cropland classification using diverse remote sensing data. It investigates the combined effect of five fusion strategies—Input, Feature, Decision, Ensemble, and Hybrid—and five temporal encoders, including LSTM, GRU, TempCNN, TAE, and L-TAE. Using the CropHarvest dataset, the study evaluates optical, radar, weather, and topographic information under different data availability scenarios. Results show that no single MVL configuration consistently performs best, particularly when labeled data are limited. Instead, the optimal combination of encoder architecture and fusion strategy depends on the specific classification task and fusion approach.

Recommended citation: Mena, Francisco, Diego Arenas, and Andreas Dengel. "In the search for optimal multi-view learning models for crop classification with global remote sensing data." International Journal of Applied Earth Observation and Geoinformation 143 (2025): 104823.
Download Paper