Deep Learning-based Automated Mango Variety Classification Using Convolutional Neural Networks

This research presents a comprehensive deep learning framework for automated classification of mango varieties using convolutional neural networks (CNNs). The study addresses the critical agricultural challenge of accurate mango variety identification, which is essential for quality control, pricing, and supply chain management in the fruit industry. We developed and evaluated multiple state-of-the-art CNN architectures including EfficientNet-B3, ResNet50, and Vision Transformers on a dataset comprising seven distinct mango varieties: Aafush, Dasheri, Jamadar, Kesar, Langdo, Rajapuri, and Totapuri. The proposed system employs advanced data augmentation techniques, transfer learning, and class imbalance handling to achieve robust classification performance. Experimental results demonstrate that our optimized EfficientNet-B3 model achieved a remarkable test accuracy of 94.3%, significantly outperforming traditional computer vision approaches. The research contributes to agricultural technology by providing an efficient, scalable solution for automated fruit classification with practical implications for farmers, distributors, and quality control agencies.