Date Fruit Varieties Identification Using Convolutional Neural Networks: A Comparative Study

Abstract

Date fruit is one of the most important economical and cultural agricultural crops in the Middle East that plays a critical role in trade and food sustainability. These merits have attracted increasing interest from researchers and the food industry to improve food sustainability. The advent and integration of computer vision and artificial intelligence (AI) technologies have rapidly accelerated the progress in the development of automated classification, quality assessment, and grading for date fruits. This study offers an inclusive comparative analysis for ten pre-trained convolutional neural network (CNN) models used to classify fifteen different date fruit cultivars. The dataset obtained from two publicly available datasets. They contain images of cultivars from Saudi Arabia and Pakistan. The images are first preprocessed to enhance their quality, segmented, augmented to overcome the imbalance problem, standardized, and normalized to be fed then to the CNNs. Transfer learning was applied to fine-tune the pre-trained models using MATLAB 2023a software package. The performance of models were evaluated based on the overall accuracy, per-class accuracy, training time, execution time, and average inference time per image. Results showed that DarkNet-50 achieved the highest accuracy (99.33%), while MobileNet-V2 and ShuffleNet provided the best balance between accuracy and efficiency, hence they are well-suited for real-time or embedded applications.