Efficiency of Identification of Blackcurrant Powders Using Classifier Ensembles

Area of Science:

• Food Science and Technology
• Computer Science
• Image Analysis

Background:

• Assessing food product quality requires efficient and accurate analytical methods.
• Non-invasive techniques are crucial for real-time food quality evaluation.
• Artificial intelligence (AI) offers innovative solutions for food industry challenges.

Purpose of the Study:

• To evaluate machine learning algorithm performance for identifying blackcurrant powders using SEM images.
• To compare the effectiveness of single classifiers versus metaclassifiers for food product analysis.
• To explore texture feature extraction using the gray-level co-occurrence matrix (GLCM) for quality assessment.

Main Methods:

• Microscopic images of blackcurrant powders were acquired using scanning electron microscopy (SEM).
• Texture features were extracted using the gray-level co-occurrence matrix (GLCM).
• Various machine learning classifiers, including single models and a metaclassifier, were trained and evaluated.

Main Results:

• The metaclassifier and a single random forest (RF) classifier demonstrated the highest accuracy in identifying blackcurrant powders.
• Machine learning models utilizing image texture features proved effective for quality evaluation.
• Ensembles of classifiers outperformed traditional single neural models in this application.

Conclusions:

• Machine learning, particularly classifier ensembles, offers a powerful approach for objective food quality assessment.
• This method can support real-time quality control and accelerate the selection of appropriate AI algorithms for food analysis.
• The developed technique shows promise for enhancing the efficiency and accuracy of food product evaluation.