EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

A Bayesian Conditional Diffusion Regression Framework with Hierarchical Global-Local Priors for Robust Agricultural Image-to-Yield Prediction

Ch. Shashi Kumar, G. Upender Reddy, K. Varalakshmi

Boletim da Sociedade Paranaense de Matemática · 19.06.2026 · 44 (17) · 1-13

DOI: 10.5269/bspm.82344 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Boletim da Sociedade Paranaense de Matemática, online erschienen am 19.06.2026. Verfasst von Ch. Shashi Kumar und 2 weiteren Autor:innen. Die Arbeit nennt 37 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Accurate estimation of crop yield from image-derived data remains difficult due to the combined effects of high predictor dimensionality, strong inter-feature dependence, non-standard noise behavior, and the presence of anomalous observations. Many commonly used regression approaches, including Bayesian models with Gaussian error assumptions, struggle to remain stable under these conditions and often provide unreliable uncertainty estimates. This work presents Bayesian Conditional Diffusion Regression (BCDR), a probabilistic modeling framework that approaches regression through conditional response generation. The method integrates a diffusion-based mechanism for modeling the response variable with hierarchical global–local shrinkage to control model complexity, while a heavy tailed likelihood improves robustness to outliers. Posterior inference is carried out using a hybrid Gibbs and Hamiltonian Monte Carlo strategy to enable tractable estimation in high-dimensional settings. Experiments conducted on synthetic datasets and a real agricultural case study involving pomegranate images and fruit weight measurements demonstrate improved predictive accuracy, bet-ter uncertainty calibration, and stronger resistance to data contamination when compared with established statistical and machine-learning baselines. These results suggest that BCDR is a robust Bayesian framework for image-based yield prediction.

Zitieren

APA

Kumar, C. S., Reddy, G. U., & Varalakshmi, K. (2026). A Bayesian Conditional Diffusion Regression Framework with Hierarchical Global-Local Priors for Robust Agricultural Image-to-Yield Prediction. Boletim da Sociedade Paranaense de Matemática, 44(17), 1-13. https://doi.org/10.5269/bspm.82344

Vancouver

Kumar CS, Reddy GU, Varalakshmi K. A Bayesian Conditional Diffusion Regression Framework with Hierarchical Global-Local Priors for Robust Agricultural Image-to-Yield Prediction. Boletim da Sociedade Paranaense de Matemática. 2026;44(17):1-13. doi:10.5269/bspm.82344

BibTeX
@article{Kumar2026,
  author = {Kumar, Ch. Shashi and Reddy, G. Upender and Varalakshmi, K.},
  title = {A Bayesian Conditional Diffusion Regression Framework with Hierarchical Global-Local Priors for Robust Agricultural Image-to-Yield Prediction},
  journal = {Boletim da Sociedade Paranaense de Matemática},
  year = {2026},
  volume = {44},
  number = {17},
  pages = {1--13},
  doi = {10.5269/bspm.82344}
}
Kostenlos starten →