Unlocking chickpea flour potential: AI-powered prediction for quality assessment and compositional characterisation

Area of Science:

• Agricultural Science
• Food Science
• Data Science

Background:

• Chickpea flour is crucial for plant-based diets, but its quality varies due to genetics, environment, and processing.
• Standardizing chickpea flour quality is challenging, impacting its use in food production.
• Accurate and efficient quality assessment methods are needed for consistent plant-based food products.

Purpose of the Study:

• To integrate deep learning (DL) models with near-infrared (NIR) spectroscopy for improved chickpea flour quality assessment.
• To compare the performance of various DL models (CNNs, ViTs, GCNs) against traditional methods.
• To evaluate the accuracy of DL models in predicting key nutritional and compositional attributes of chickpea flour.

Main Methods:

• Utilized a dataset of 136 chickpea varieties for analysis.
• Applied deep learning models including Convolutional Neural Networks (CNNs), Vision Transformers (ViTs), and Graph Convolutional Networks (GCNs).
• Compared the performance of the most effective DL model (CNN) against Partial Least Squares Regression (PLSR).

Main Results:

• CNN-based deep learning models outperformed the traditional PLSR method in accuracy.
• Accurate predictions were achieved for protein content, starch, soluble sugars, insoluble fibers, total lipids, and moisture levels.
• AI-enhanced NIR spectroscopy demonstrated potential for non-destructive, rapid, and reliable chickpea flour analysis.

Conclusions:

• Deep learning models integrated with NIR spectroscopy offer a promising approach for revolutionizing chickpea flour quality control.
• AI-driven methods can lead to more consistent and higher-quality plant-based food products.
• Further advancements in DL models could enable industrial applicability for enhanced food quality assessment.