Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Energy-Aware Ensemble Learning for Coffee Leaf Disease Classification

Larissa Ferreira Rodrigues Moreira, Rodrigo Moreira, Leonardo Gabriel Ferreira Rodrigues

TL;DR

This work addresses the challenge of on-device coffee leaf disease diagnosis under constrained compute and intermittent connectivity by combining knowledge distillation with energy-aware ensemble learning. High-capacity CNNs are used in data centers to guide compact backbones, which are then fused through simple averaging, optimized weighted averaging, and stacking to maximize accuracy within a tight compute budget. On the RoCoLe dataset, distilled tiny ensembles achieve approximately 94.25% accuracy with significantly reduced energy use and carbon footprint, demonstrating practical, low-power deployment for IoT agricultural sensing. The results provide actionable guidance for edge intelligence in plant-disease monitoring, including transparent energy reporting alongside predictive performance and latency.

Abstract

Coffee yields are contingent on the timely and accurate diagnosis of diseases; however, assessing leaf diseases in the field presents significant challenges. Although Artificial Intelligence (AI) vision models achieve high accuracy, their adoption is hindered by the limitations of constrained devices and intermittent connectivity. This study aims to facilitate sustainable on-device diagnosis through knowledge distillation: high-capacity Convolutional Neural Networks (CNNs) trained in data centers transfer knowledge to compact CNNs through Ensemble Learning (EL). Furthermore, dense tiny pairs were integrated through simple and optimized ensembling to enhance accuracy while adhering to strict computational and energy constraints. On a curated coffee leaf dataset, distilled tiny ensembles achieved competitive with prior work with significantly reduced energy consumption and carbon footprint. This indicates that lightweight models, when properly distilled and ensembled, can provide practical diagnostic solutions for Internet of Things (IoT) applications.

Energy-Aware Ensemble Learning for Coffee Leaf Disease Classification

TL;DR

This work addresses the challenge of on-device coffee leaf disease diagnosis under constrained compute and intermittent connectivity by combining knowledge distillation with energy-aware ensemble learning. High-capacity CNNs are used in data centers to guide compact backbones, which are then fused through simple averaging, optimized weighted averaging, and stacking to maximize accuracy within a tight compute budget. On the RoCoLe dataset, distilled tiny ensembles achieve approximately 94.25% accuracy with significantly reduced energy use and carbon footprint, demonstrating practical, low-power deployment for IoT agricultural sensing. The results provide actionable guidance for edge intelligence in plant-disease monitoring, including transparent energy reporting alongside predictive performance and latency.

Abstract

Coffee yields are contingent on the timely and accurate diagnosis of diseases; however, assessing leaf diseases in the field presents significant challenges. Although Artificial Intelligence (AI) vision models achieve high accuracy, their adoption is hindered by the limitations of constrained devices and intermittent connectivity. This study aims to facilitate sustainable on-device diagnosis through knowledge distillation: high-capacity Convolutional Neural Networks (CNNs) trained in data centers transfer knowledge to compact CNNs through Ensemble Learning (EL). Furthermore, dense tiny pairs were integrated through simple and optimized ensembling to enhance accuracy while adhering to strict computational and energy constraints. On a curated coffee leaf dataset, distilled tiny ensembles achieved competitive with prior work with significantly reduced energy consumption and carbon footprint. This indicates that lightweight models, when properly distilled and ensembled, can provide practical diagnostic solutions for Internet of Things (IoT) applications.
Paper Structure (15 sections, 6 equations, 5 figures, 3 tables)

This paper contains 15 sections, 6 equations, 5 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Proposed Approach
  4. IoT-Driven Data Acquisition Framework
  5. Dataset
  6. Lightweight Base Classifiers (Base Models)
  7. Meta-Model Analysis
  8. Ensemble Strategies
  9. Simple Averaging Ensemble
  10. Optimized Weighted Averaging Ensemble
  11. Stacked Generalization (Stacking)
  12. Results and Discussion
  13. Energy Consumption and Carbon Footprint
  14. Ensemble Evaluation
  15. Concluding Remarks

Figures (5)

  • Figure 1: Top row: healthy leaves. Bottom row: disease leaves with rust symptoms from the RoCoLe dataset.
  • Figure 2: Training and test accuracy/loss over epochs for baseline CNNs.
  • Figure 3: Energy breakdown (CPU/GPU/RAM) for training vs. two-model ensembles.
  • Figure 4: Prediction confidence distributions for the Stacking meta-models.
  • Figure 5: Ensemble for each experiment.