Publications

This paper investigates conformal prediction as a framework for uncertainty quantification in crop production systems. We analyze how conformal prediction can enhance decision support by providing reliable prediction sets with coverage guarantees, comparing it against established uncertainty methods across various agricultural monitoring tasks.

We propose the Confident Naturalness Explanation (CNE) framework that combines explainability methods with uncertainty-aware confidence measures to explain and assess the patterns that form naturalness in remote sensing imagery. The framework bridges the gap between model interpretability and ecological assessment.

This work presents a comprehensive framework for integrating explainable machine learning into digital agriculture decision support systems. We demonstrate how explainability methods can enhance trust and understanding in crop monitoring predictions, evaluated across multiple agricultural datasets.

We present a comparative study of uncertainty quantification methods for harvest-readiness classification in cauliflower plants. Our evaluation focuses on inductive conformal prediction, comparing its calibration properties, computational efficiency, and coverage guarantees against ensemble-based and Bayesian uncertainty methods.

This paper proposes an automatic severity classification system for diabetic retinopathy using DenseNet enhanced with Convolutional Block Attention Module (CBAM). The attention mechanism improves the network's ability to focus on relevant retinal features, achieving competitive classification accuracy across multiple severity levels.