A Hybrid Adaptive Representation and Margin-Based Approach for Rice Variety Classification
编号:219
访问权限:仅限参会人
更新:2026-11-16 11:09:43 浏览:8次
Online
摘要
Rice variety classification is essential for seed quality assessment, agricultural management, authentication, and supply-chain applications. However, existing Machine Learning (ML) and Deep Learning (DL) approaches are often limited by handcrafted feature dependence, incomplete feature representation, and inadequate handling of class imbalance. Hence, this study addresses these limitations by developing an enhanced ML-based approach for accurate rice variety classification. The proposed HARM-Rice (Hybrid Adaptive Representation and Margin-based Rice Classifier) integrates multi-scale deep feature extraction with statistical rice-grain characteristics to construct a comprehensive representation. An adaptive feature-gating mechanism is employed to emphasize discriminative features, while adaptive class weighting and margin-based classification improve class separation and robustness to imbalanced data. Experiments conducted on the Rice MSC Dataset containing 75,000 images across five rice varieties demonstrate that HARM-Rice achieves 99.92% accuracy. The findings show effectiveness of HARM-Rice for reliable rice variety classification.
关键词
Rice Variety Classification,Machine Learning,Deep Learning,Feature Extraction,Class Imbalance,HARM-Rice
稿件作者
Madhu Shree R
DR. AMBEDKAR INSTITUTE OF TECHNOLOGY
Nandini N
DR. AMBEDKAR INSTITUTE OF TECHNOLOGY
发表评论