Related Experiment Video
Updated: Jul 1, 2026
03:31
End-To-End Deep Neural Network for Salient Object Detection in Complex Environments
Published on: December 15, 2023
A Validation-Driven Explainable Deep Ensemble Framework for Image-Based Saffron Adulteration Detection.
Syed Nisar Hussain Bukhari1, Kingsley A Ogudo2
1National Institute of Electronics and Information Technology (NIELIT) J&K, Srinagar 191132, India.
Foods (Basel, Switzerland)
|May 27, 2026
Summary
This study introduces a robust deep learning ensemble for detecting saffron adulteration using images. The validated framework ensures reliable authentication, outperforming individual models and offering explainable results for food quality control.
Area of Science:
- Food Science
- Computer Science
- Analytical Chemistry
Background:
- Saffron (Crocus sativus L.) is a high-value spice susceptible to adulteration.
- Conventional authentication methods are limited for rapid, non-destructive analysis.
- Existing deep learning approaches often lack rigorous validation and statistical reliability.
Purpose of the Study:
- To develop a validation-driven and explainable deep ensemble framework for image-based saffron adulteration detection.
- To ensure robust and statistically reliable authentication of saffron quality.
- To provide methodological insights for food adulteration detection using limited data.
Main Methods:
- Integration of pretrained convolutional neural networks (DenseNet169, ResNet50, VGG16) using a validation-driven weighted ensemble.
- Fusion weights computed from validation performance within training folds to prevent information leakage.
- Stratified five-fold cross-validation and statistical tests (McNemar's, 5x2 cv) for performance validation.
- Grad-CAM for explainability and background-invariance analysis for robustness.
Main Results:
- Achieved 98.61% classification accuracy, 98.17% F1-score, and 98.61% AUC.
- Outperformed the best individual base model by up to 1.4% in F1-score.
- Demonstrated stable performance with mean accuracy of 97.81% ± 0.53 via cross-validation.
- Statistical validation confirmed reliable performance improvements.
Conclusions:
- The proposed deep ensemble framework offers a robust, interpretable, and statistically validated solution for saffron authentication.
- The method addresses limitations of conventional techniques and single deep learning models.
- Provides a reliable approach for image-based food adulteration detection, especially under limited data conditions.
