Abstract

This project implements a Convolutional Neural Network (CNN) based on GoogLeNet for recognizing five species of flowers: sunflower, daisy, tulip, rose, and dandelion. Using a dataset from Kaggle, the model achieved a training accuracy of 95.8% and a validation accuracy of 70.5%. The system was deployed via a GUI using Gradio, allowing horticultural beginners to upload images and receive real-time flower identification.

Key Contributions

• Designed and fine-tuned a GoogLeNet-based CNN for flower classification with multiple hyperparameter settings.
• Achieved 95.8% training accuracy and 70.5% validation accuracy across five flower classes.
• Developed a lightweight Gradio-based GUI for real-time flower image recognition.
• Evaluated model performance using confusion matrices and optimization metrics across different optimizers and learning rates.

Conclusion

This project shows the feasibility of deploying CNN-based models for practical flower recognition. By using GoogLeNet and integrating a user-friendly interface, it bridges the gap between AI and accessible tools for amateur botanists. Future work should expand the flower dataset and further improve generalization across broader flower species.

Limitation

The dataset used includes only five flower types, which limits generalizability. Validation accuracy, though decent, suggests room for improved performance on unseen data, possibly due to data bias or model overfitting.

Future Work

Expanding the dataset, incorporating real-time mobile applications, and refining the model using data augmentation and transfer learning techniques are future directions that can boost the utility and scalability of this tool.