Journals Information
Food Science and Technology Vol. 13(2), pp. 165 - 177
DOI: 10.13189/fst.2025.130205
Reprint (PDF) (2426Kb)
Deep Learning-Powered Precision: A CNN-Based Approach for Postharvest Classification of Indian Banana Varieties in Supermarket Supply Chains
Ashoka Kumar Ratha 1, A. Geetha Devi 2, Prabira Kumar Sethy 1, Nalini Kanta Barpanda 1, Santi Kumari Behera 3, Aziz Nanthaamornphong 4,*
1 Department of Electronics, Sambalpur University, Odisha, India
2 Department of Electronics and Communication Engineering, PVP Siddhartha Institute of Technology, Vijayawada, Andhra Pradesh 520007, India
3 Department of Computer Science and Engineering, Veer Surendra Sai University of Technology, Odisha, India
4 College of Computing, Prince of Songkla University, Phuket, Thailand
ABSTRACT
Banana classification holds vital importance in India, the world's largest banana producer, as accurate variety identification is crucial for optimizing postharvest handling, pricing, and supply chain management, ensuring quality consistency and market efficiency across diverse regional and commercial demands. This study proposes an advanced classification technique using an integrated deep convolutional neural network model tailored to classify ten prominent Indian banana varieties. The model strategically combines ResNet and DenseNet architectures enhanced with attention mechanisms as feature extractors, while employing a Support Vector Machine (SVM) for final classification. A distinctive aspect of this approach is its patch-wise image analysis strategy, wherein banana images are divided into sequences of patches to effectively capture both fine-grained local details and broader global patterns. The incorporation of self-attention mechanisms enables the model to learn intricate global relationships within these patches, substantially improving feature extraction efficiency. Additionally, DenseNet's well-connected architecture facilitates the systematic elimination of redundant or irrelevant features, optimizing the classification workflow. Remarkably, the proposed integrated model achieved an accuracy rate of 98.3%, outperforming conventional deep learning networks. This significant improvement in classification precision holds considerable promise for enhancing postharvest processes in supermarket supply chains. By ensuring reliable and efficient banana variety identification, this method can greatly contribute to streamlined inventory management, pricing strategies, and overall supply chain efficiency.
KEYWORDS
Deep Learning, ResNet, DenseNet, Banana Classification, Support Vector Machine (SVM)
Cite This Paper in IEEE or APA Citation Styles
(a). IEEE Format:
[1] Ashoka Kumar Ratha , A. Geetha Devi , Prabira Kumar Sethy , Nalini Kanta Barpanda , Santi Kumari Behera , Aziz Nanthaamornphong , "Deep Learning-Powered Precision: A CNN-Based Approach for Postharvest Classification of Indian Banana Varieties in Supermarket Supply Chains," Food Science and Technology, Vol. 13, No. 2, pp. 165 - 177, 2025. DOI: 10.13189/fst.2025.130205.
(b). APA Format:
Ashoka Kumar Ratha , A. Geetha Devi , Prabira Kumar Sethy , Nalini Kanta Barpanda , Santi Kumari Behera , Aziz Nanthaamornphong (2025). Deep Learning-Powered Precision: A CNN-Based Approach for Postharvest Classification of Indian Banana Varieties in Supermarket Supply Chains. Food Science and Technology, 13(2), 165 - 177. DOI: 10.13189/fst.2025.130205.