TY - GEN
T1 - Transfer Learning and Explainable Artificial Intelligence Enhance the Classification of Date Fruit Varieties
AU - Zaki, Nazar
AU - Singh, Harsh
AU - Krishnan, Anusuya
AU - Alnaqbi, Aisha
AU - Alneyadi, Shamma
AU - Alnaqbi, Sara
AU - Alhindaassi, Sarah
AU - Alam, Muneeba
AU - Eldin, Afaf Kamal
N1 - Publisher Copyright:
© 2023 IEEE.
PY - 2023
Y1 - 2023
N2 - This study proposes a novel approach for the classification of date fruits using image processing, convolutional neural networks (CNN), and explainable artificial intelligence (AI). A unique, manually curated dataset was assembled for this research, including a diverse range of date fruit types. Rigorous image pre-processing techniques, including grayscale conversion, normalization, augmentation, and resizing, were applied to prepare the dataset for effective model training. Various renowned transfer learning models-VGG16, VGG19, MobileNetV2, EfficientNetV2L, and ResNet50-were evaluated for their performance on this dataset, and assessed based on accuracy, precision, recall, and F1-score. The ResNet50 model demonstrated superior results with an accuracy of 94%. Importantly, the study incorporated Gradient-weighted Class Activation Mapping (Grad-CAM), an explainable AI technique, providing a visually interpretable rationale for the model's decisions by highlighting significant image regions contributing to classification. To the best of our knowledge, this implementation of Grad-CAM to address this research challenge is pioneering, offering significant insights for future research in agricultural produce classification.
AB - This study proposes a novel approach for the classification of date fruits using image processing, convolutional neural networks (CNN), and explainable artificial intelligence (AI). A unique, manually curated dataset was assembled for this research, including a diverse range of date fruit types. Rigorous image pre-processing techniques, including grayscale conversion, normalization, augmentation, and resizing, were applied to prepare the dataset for effective model training. Various renowned transfer learning models-VGG16, VGG19, MobileNetV2, EfficientNetV2L, and ResNet50-were evaluated for their performance on this dataset, and assessed based on accuracy, precision, recall, and F1-score. The ResNet50 model demonstrated superior results with an accuracy of 94%. Importantly, the study incorporated Gradient-weighted Class Activation Mapping (Grad-CAM), an explainable AI technique, providing a visually interpretable rationale for the model's decisions by highlighting significant image regions contributing to classification. To the best of our knowledge, this implementation of Grad-CAM to address this research challenge is pioneering, offering significant insights for future research in agricultural produce classification.
KW - Classification
KW - Convolutional Neural Networks (CNN)
KW - Date fruit
KW - Grad-CAM
KW - ResNet50
KW - explainable AI
KW - transfer learning models
UR - http://www.scopus.com/inward/record.url?scp=85182946822&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85182946822&partnerID=8YFLogxK
U2 - 10.1109/IIT59782.2023.10366495
DO - 10.1109/IIT59782.2023.10366495
M3 - Conference contribution
AN - SCOPUS:85182946822
T3 - 2023 15th International Conference on Innovations in Information Technology, IIT 2023
SP - 222
EP - 227
BT - 2023 15th International Conference on Innovations in Information Technology, IIT 2023
PB - Institute of Electrical and Electronics Engineers Inc.
T2 - 15th International Conference on Innovations in Information Technology, IIT 2023
Y2 - 14 November 2023 through 15 November 2023
ER -