Selected work

Representative Papers

Learning to Forecast Crop Growth from Earth Observation Data

Aug 14, 2026

This study addresses trajectory distortion in national-scale crop growth prediction caused by sparse observations. We propose a lightweight unimodal shape-regularized Seq2Seq model that integrates Sentinel-2 remote sensing time series with meteorological drivers, embedding biological priors into a deep learning framework to effectively constrain curve morphology under sparse supervision. Experimental results demonstrate that the model predicts winter wheat LAI trajectories with an R² exceeding 0.8, significantly outperforming conventional methods. This approach achieves high-precision retrieval of growth dynamics consistent with agronomic principles, validating the effectiveness of multi-source data-driven sequence modeling for landscape-scale crop monitoring.

0 citationsRead paper

SwissCrop25: A National Multi-Year Benchmark for Operational Crop Mapping

Aug 10, 2026

Existing crop mapping datasets struggle to jointly evaluate model performance across multi-year generalization, fine-grained classification, and cropland identification. To address this gap, this work introduces a nationwide Swiss benchmark dataset spanning seven growing seasons, integrating Sentinel-2 time-series imagery, phenological features derived from daily temperature, 73 crop classes, and five non-crop land-cover categories. A leave-one-year-out cross-validation protocol is designed to simulate real-world deployment conditions. Experiments on joint cropland delineation and crop classification using spatiotemporal architectures—including U-TAE, TSViT, and Galileo—demonstrate that TSViT achieves the best performance, surpassing U-TAE by 12 percentage points in macro mIoU. Domain-specific models consistently outperform general-purpose remote sensing foundation models, and incorporating phenological information significantly enhances inter-annual robustness, while also revealing pronounced seasonal trade-offs in model performance.

0 citationsRead paper

Machine Learning-based Early Detection of Potato Sprouting Using Electrophysiological Signals

Jul 01, 2025

This study addresses post-harvest losses in potato storage caused by delayed detection of sprouting. We propose a novel early sprouting prediction method based on tuber electrophysiological signals. Custom-designed sensors acquire electrophysiological data from stored tubers; wavelet transform is applied to extract time–frequency domain features; and a supervised machine learning model—integrated with uncertainty quantification—is developed to predict sprouting onset prior to visible eye emergence. Unlike conventional vision-based approaches, our method overcomes temporal limitations, enabling prediction several days in advance, with mean absolute error within an acceptable range. The key innovation lies in the first integration of tuber electrophysiological responses, wavelet-domain feature modeling, and predictive uncertainty quantification. This work establishes a practical, non-invasive paradigm for intelligent, real-time post-harvest storage management.

0 citationsRead paper

Recent publications

Latest Papers

Learning to Forecast Crop Growth from Earth Observation Data

Aug 14, 2026

This study addresses trajectory distortion in national-scale crop growth prediction caused by sparse observations. We propose a lightweight unimodal shape-regularized Seq2Seq model that integrates Sentinel-2 remote sensing time series with meteorological drivers, embedding biological priors into a deep learning framework to effectively constrain curve morphology under sparse supervision. Experimental results demonstrate that the model predicts winter wheat LAI trajectories with an R² exceeding 0.8, significantly outperforming conventional methods. This approach achieves high-precision retrieval of growth dynamics consistent with agronomic principles, validating the effectiveness of multi-source data-driven sequence modeling for landscape-scale crop monitoring.

0 citationsRead paper

SwissCrop25: A National Multi-Year Benchmark for Operational Crop Mapping

Aug 10, 2026

Existing crop mapping datasets struggle to jointly evaluate model performance across multi-year generalization, fine-grained classification, and cropland identification. To address this gap, this work introduces a nationwide Swiss benchmark dataset spanning seven growing seasons, integrating Sentinel-2 time-series imagery, phenological features derived from daily temperature, 73 crop classes, and five non-crop land-cover categories. A leave-one-year-out cross-validation protocol is designed to simulate real-world deployment conditions. Experiments on joint cropland delineation and crop classification using spatiotemporal architectures—including U-TAE, TSViT, and Galileo—demonstrate that TSViT achieves the best performance, surpassing U-TAE by 12 percentage points in macro mIoU. Domain-specific models consistently outperform general-purpose remote sensing foundation models, and incorporating phenological information significantly enhances inter-annual robustness, while also revealing pronounced seasonal trade-offs in model performance.

0 citationsRead paper

Machine Learning-based Early Detection of Potato Sprouting Using Electrophysiological Signals

Jul 01, 2025

This study addresses post-harvest losses in potato storage caused by delayed detection of sprouting. We propose a novel early sprouting prediction method based on tuber electrophysiological signals. Custom-designed sensors acquire electrophysiological data from stored tubers; wavelet transform is applied to extract time–frequency domain features; and a supervised machine learning model—integrated with uncertainty quantification—is developed to predict sprouting onset prior to visible eye emergence. Unlike conventional vision-based approaches, our method overcomes temporal limitations, enabling prediction several days in advance, with mean absolute error within an acceptable range. The key innovation lies in the first integration of tuber electrophysiological responses, wavelet-domain feature modeling, and predictive uncertainty quantification. This work establishes a practical, non-invasive paradigm for intelligent, real-time post-harvest storage management.

0 citationsRead paper