On Application of Lightweight Models for Rice Variety Classification and Their Potential in Edge Computing.

Area of Science:

• Agricultural Science
• Computer Science
• Data Science

Background:

• Rice classification is crucial for industrial and marketing applications.
• Traditional machine learning and deep learning models exist but are resource-intensive.
• Existing deep learning models often require significant expert supervision for feature engineering.

Purpose of the Study:

• To develop computationally efficient deep learning models for rice variety classification.
• To minimize the need for expert supervision in pre-processing and feature engineering.
• To achieve comparable performance to existing state-of-the-art models with reduced overhead.

Main Methods:

• Proposal of three novel deep learning models for image-based rice classification.
• Training models using an end-to-end approach to streamline the workflow.
• Evaluation of models based on accuracy and computational overhead.

Main Results:

• The proposed models demonstrate comparable performance to existing state-of-the-art methods.
• Achieved a 10% reduction in computational overhead compared to current best models.
• Successfully classified five distinct rice varieties: Arborio, Basmati, Ipsala, Jasmine, and Karacadag.

Conclusions:

• The developed deep learning models offer an efficient alternative for rice variety classification.
• The end-to-end training approach reduces the necessity for extensive expert intervention.
• The models show potential for deployment on edge and mobile devices for autonomous field applications.