Non-Destructive Prediction of Fruit Ripeness and Firmness Using Hyperspectral Imaging and Lightweight Machine Learning Models

Under Review Computers and Electronics in Agriculture 2025

P. M. Konrad, C. Kunstmann-Olsen, J. Fiutowski, S. Ayvaz

Critical-difference diagram comparing traditional ML algorithms across ripeness-classification and firmness-prediction tasks. Tree-based methods match or exceed published deep-learning benchmarks at a fraction of the cost.

Headline result

Tree-based machine-learning models outperform the published deep-learning Fruit-HSNet baseline at orders-of-magnitude lower compute cost, and just three visible-range wavelengths recover over 94% of full-spectrum accuracy. Low-cost multispectral sensors are a viable alternative to expensive hyperspectral cameras for practical fruit-quality sorting.

Method in brief

Twenty classical machine-learning algorithms are benchmarked on hyperspectral imaging data for joint ripeness classification and firmness prediction across five fruit species, using a cross-validated experimental design with Bayesian hyperparameter optimisation. Preprocessing strategy (class balancing, spectral transformation) is studied as an explicit factor alongside algorithm choice.

Key Contributions

Systematic benchmark of 20 classical machine-learning algorithms on hyperspectral imaging data for joint ripeness classification and firmness prediction across five fruit species.
Tree-based models match or outperform the state-of-the-art deep-learning Fruit-HSNet baseline at orders-of-magnitude lower compute cost and without GPU dependence.
Demonstrates that preprocessing strategy (class balancing, spectral transformation) contributes as much to accuracy as model choice, with detailed ablations across the design space.
Identifies that only three visible-range wavelengths recover over 94% of full-spectrum accuracy, showing that low-cost multispectral sensors are a practical alternative to expensive hyperspectral cameras.

Abstract

Post-harvest fruit quality assessment is essential for reducing food waste, yet reliable non-destructive methods typically depend on expensive hyperspectral cameras and computationally intensive deep learning models. These systems typically require GPU resources, large-scale training data, and domain expertise, limiting their feasibility for many real-world agricultural settings. This study systematically evaluates 20 classical machine learning algorithms on hyperspectral imaging data for simultaneous ripeness classification and firmness prediction across five fruit species, using cross-validated experimental design with Bayesian hyperparameter optimisation. Data preprocessing strategy, particularly class balancing and spectral transformations, contributes as much to prediction accuracy as algorithm choice. Our results show that tree-based machine learning models can outperform state-of-the-art deep learning models reported in Fruit-HSNet. Moreover, the findings indicate that only three visible-range wavelengths are needed to recover over 94% of full-spectrum accuracy, demonstrating that low-cost multispectral sensors combined with lightweight machine learning models can serve as practical alternatives to expensive hyperspectral cameras and complex deep learning approaches for practical fruit-quality sorting.