Age Classification of Rice Seeds in Japan Using Gradient-Boosting and ANFIS Algorithms

Area of Science:

• Agricultural Science
• Computer Science
• Data Science

Background:

• Climate change impacts the food industry, affecting staple crops like rice.
• Japanese agriculture relies on aged seeds due to frequent natural disasters.
• A significant research gap exists in identifying seeds based on age.

Purpose of the Study:

• To develop and implement a machine learning model for identifying Japanese rice seeds by age.
• To address the lack of agewise rice seed datasets in existing literature.
• To improve cultivation success rates by enabling accurate seed age classification.

Main Methods:

• Creation of a novel rice seed dataset with six varieties and three age groups using RGB images.
• Extraction of image features using six distinct feature descriptors.
• Implementation of a Cascaded-ANFIS algorithm, integrating XGBoost, CatBoost, and LightGBM for a two-step classification (variety, then age).

Main Results:

• The proposed Cascaded-ANFIS algorithm outperformed 13 state-of-the-art algorithms in accuracy, precision, recall, and F1-score.
• Achieved specific scores for variety classification: 0.7697 (accuracy), 0.7949 (precision), 0.7707 (recall), and 0.7862 (F1-score).
• Demonstrated the model's capability for successful seed age classification.

Conclusions:

• The developed machine learning model, Cascaded-ANFIS, is effective for classifying Japanese rice seeds by age.
• This research contributes a novel dataset and algorithm to the field of seed identification.
• The findings support the application of advanced machine learning in agricultural practices for enhanced crop management.