Fruit Detection and Classification Using CNN with Deep Transfer Learning

Abstract: Accurate and efficient fruit detection and classification are critical in modern precision agriculture, supply chain automation, and quality control. Manual inspection methods are time-consuming, error-prone, and impractical at scale. This paper proposes a deep learning framework that integrates Convolutional Neural Networks (CNN) with deep transfer learning techniques to automatically detect and classify multiple fruit categories from images. We fine-tune three pre-trained models — VGG19, ResNet-50, and MobileNetV2 — on the publicly available Fruits-360 dataset consisting of 90,380 images across 131 fruit classes. Data augmentation strategies including random flipping, rotation, and brightness adjustment were applied to improve model generalization. Experimental results demonstrate that the fine- tuned ResNet-50 model achieves the highest classification accuracy of 98.74%, while MobileNetV2 provides the best trade-off between accuracy and computational efficiency at 97.91% accuracy with significantly reduced inference time. The proposed approach outperforms several existing methods and shows strong potential for real-world agricultural deployment.

Keywords: Fruit Detection, Image Classification, Convolutional Neural Networks, Transfer Learning, Deep Learning, VGG19, ResNet-50, MobileNetV2, Fruits-360.