Distributed search and fusion for wine label image retrieval

Area of Science:

• Computer Science
• Artificial Intelligence
• Machine Learning

Background:

• Wine culture is growing, increasing the need for efficient wine label image retrieval.
• Deep learning models for retrieval struggle with the large number of wine brands and imbalanced sample data.

Purpose of the Study:

• To propose effective distributed retrieval frameworks for wine label image retrieval.
• To address the challenge of imbalanced sample data in wine brand identification.

Main Methods:

• Developed two novel distributed retrieval frameworks.
• Utilized a distributed strategy to handle large-scale datasets and data imbalance.
• Evaluated performance on a large-scale wine label dataset and the Oxford flowers dataset.

Main Results:

• Both proposed distributed retrieval frameworks demonstrated significant effectiveness.
• The frameworks outperformed previous state-of-the-art retrieval models.
• Successful retrieval of wine information, including brand and sub-brand, was achieved.

Conclusions:

• The proposed distributed retrieval frameworks are effective solutions for wine label image retrieval.
• These methods successfully overcome the limitations of imbalanced data in deep learning retrieval systems.
• The approach shows strong potential for real-world applications in wine information systems.