SaffNet: an ensemble-based approach for saffron adulteration prediction using statistical image features.

Area of Science:

• Agricultural Science
• Computer Science
• Food Science

Background:

• Saffron adulteration is a significant issue due to its high cost and restricted availability, making visual discrimination challenging.
• Existing saffron adulteration prediction systems (SAPS) face limitations in generalization and database scarcity, particularly for specific regional varieties like Kashmiri saffron.

Purpose of the Study:

• To develop a novel ensemble-based model, SaffNet, for accurate prediction of adulteration in Kashmiri saffron.
• To address the challenge of limited benchmark datasets for Kashmiri saffron by creating a new dataset.

Main Methods:

• A novel Saffron dataset (Saff-Kash) was created, comprising authentic and adulterated Kashmiri saffron samples.
• Statistical image features were extracted and pre-processed.
• An ensemble-based model, SaffNet, utilizing gradient boosting, was designed and evaluated.

Main Results:

• The SaffNet model achieved an overall accuracy of 98% in detecting saffron adulteration.
• SaffNet outperformed individual classifiers such as Support Vector Machine (SVM), Decision Tree, and K-Nearest Neighbor (KNN).
• The gradient boosting ensemble in SaffNet demonstrated efficient training times (7.7 ms).

Conclusions:

• The proposed SaffNet model effectively detects adulteration in Kashmiri saffron, offering a reliable solution.
• The creation of the Saff-Kash dataset addresses a critical need for regional saffron analysis.
• SaffNet's ensemble approach enhances prediction accuracy and generalization capabilities for saffron authentication.