This research presents FruitNet-121, an innovative deep learning framework designed for automated fruit classification, leveraging the Fruits-360 dataset. The system is built upon the DenseNet121 architecture, which was selected after thoroughly evaluating various top-performing models, such as ViT-L/14, DINOv2, VGG, ResNet, and different Inception models. Among these, DenseNet121 demonstrated the highest performance, achieving a notable accuracy of 98.35% and an F1-score of 0.98. To facilitate practical application, the system incorporates an image enhancement module that processes input images by removing backgrounds, reducing noise, and resizing them to a standardized format. Users can access the platform through a user-friendly web interface, which also features modules for model management, data recording, feedback collection, and secure data management. Further assessment of precision and recall validates the model's consistent and balanced performance. Training metrics indicate efficient convergence without signs of overfitting. As a result, the study highlights the effectiveness of DenseNet121 for fruit classification, emphasizing its potential for real-world applications in fields such as agriculture, retail, and education.