Abstract

The rapid rise of e-commerce in the fashion industry has sparked a need for intelligent clothing analysis. This project focuses on attribute label recognition and key point localization using deep learning. An AlexNet-based CNN was implemented, incorporating convolution layers, dropout, and fully connected layers to enhance feature learning. Preprocessed image data were classified using a SoftMax function. Results showed improvements in prediction accuracy and processing speed.

Key Contributions

• Designed and trained a CNN model based on AlexNet for classifying clothing images.
• Applied data augmentation to increase dataset diversity and prevent overfitting.
• Achieved 96.88% accuracy on the test dataset.
• Built a GUI using PyQt5 to visualize and interact with the classification system.
• Demonstrated model evaluation using accuracy, loss curves, and prediction results.

Conclusion

The implementation of AlexNet effectively classified clothing images into categories with high accuracy. The system is robust, fast, and enhanced with a user-friendly GUI. The integration of deep learning and computer vision provides a scalable solution for intelligent fashion analysis.

Limitations

• Dataset limited to 2644 images, potentially reducing generalization to broader fashion items.
• Model focused only on binary classification (clothes vs pants).
• GUI functions are basic and may lack advanced usability features.

Future Work

• Expand dataset with more diverse clothing categories for multi-class classification.
• Integrate Capsule Networks or ResNet for higher-level feature extraction.
• Improve GUI design for real-world deployment in retail platforms.

Student Profile

Name: Zhao Fangding
Motivation: Interested in AI applications in retail and fashion image recognition.
Aspiration: Hopes to further explore neural networks in visual commerce solutions.