Abstract

This project focuses on developing a convolutional neural network-based method for flower image classification using the VGG16 model. A dataset of 17 flower species from Oxford’s Visual Geometry Group was used, achieving a classification accuracy of 92.04%. The model uses ReLU activation, SGD optimizer, and 13 convolutional layers followed by 5 pooling layers. A GUI was developed using PyQt5 to make the system accessible for real-time flower recognition.

Key Contributions

• Applied VGG16 architecture for classifying 17 flower species with 92.04% accuracy.
• Utilized Oxford VGG Flower Dataset with well-structured data preprocessing and augmentation.
• Developed a user-friendly PyQt5 GUI for real-time flower classification and model interaction.
• Visualized training progress using accuracy and loss curves, aiding model evaluation.

Conclusion

The study demonstrates the effectiveness of deep CNNs, particularly VGG16, in flower classification tasks. The integrated GUI allows for user engagement and educational use in botanical classification systems.

Limitation

VGG16 requires significant memory and computational power due to its large number of parameters, making deployment on low-resource devices challenging.

Future Work

Future improvements include migrating to lightweight architectures like MobileNet for deployment on mobile platforms and expanding the dataset with more flower categories for better generalization.