Enhanced Flower Recognition via Transfer Learning with ResNet-50

Abstract-This paper proposes a flower recognition system using transfer learning with the ResNet-50 architecture. By utilizing pre-trained weights from ResNet-50, the system classifies ten species of flowers, drawing on an extended dataset with over 8,000 labelled images. The study addresses challenges in deep convolutional neural networks, such as overfitting and local optimality, by fine-tuning the ResNet-50 model. Initially, only the final layers of the model are retrained on the flower dataset, while the pre-trained layers remain frozen. After achieving initial convergence, all layers are unfrozen for full model fine-tuning. The dataset is divided into training, validation, and test sets to evaluate the model’s performance, which is measured using accuracy, and F1-score. The experimental results demonstrate that the transfer learning approach significantly improves classification accuracy and generalization, outperforming traditional methods. This approach proves especially effective in handling visually similar flower species and diverse environmental conditions. The study highlights the potential of transfer learning in enhancing the efficiency and robustness of flower recognition systems, contributing to broader applications in image classification tasks.