Leveraging Deep Learning to Automate the Approval Process for Durum Wheat Grain Quality

This study addresses the urgent need for accurate and efficient quality control of durum wheat grains in Algeria, where traditional inspection methods are often insufficient. To tackle this, we trained and evaluated seven convolutional neural network (CNN) models on a comprehensive dataset of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{1 8, 3 6 2}$</tex> images, aiming to classify wheat species, varieties, and impurities. Our findings demonstrate the ResNet50 model's superior performance in species and variety classification, achieving accuracies of <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$99.59 \%$</tex> and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$98.11 \%$</tex>, respectively. Although its accuracy for impurity detection was comparatively lower, likely due to the limited size of the impurity's dataset. ResNet50 still demonstrated its potential in addressing this challenging classification task. Future directions for this study include integrating the ResNet50 model into an automated quality control system, representing a significant step forward in enhancing the efficiency, accuracy, and consistency of wheat quality assessments within the Algerian agricultural sector. This adaptable solution could extend to other cereal types, offering a scalable impact on agricultural quality control practices more broadly.