Optimized Multi-Class Rice Leaf Disease Classification Framework Using Rice Feature Selection (RiceFS) and Ensemble Machine Learning: Towards Sustainable Agriculture

Abstract

Abstract Sustainable agriculture has substantial share on improvement of food security and optimization of resources utilization particularly for high value crops like rice leaf. Rice varieties should be properly classified in order to benefit the harvest management, reduced loss after harvest and improved agriculture methods. The traditional classification method usually brings the low precision and the traditional classification method is also subjected to human error, which is difficult to bring about reliable output. This study introduces an optimized multi-class rice leaf disease classification system utilizing “Rice Feature Selection” (RiceFS) and ensemble machine learning approaches. RiceFS is realized based on a feature selection mechanism based on Recursive Feature Elimination. Selected classifiers such as KNN, Random Forest, Gradient Boosting, Ensemble Learning and Optimized SVM are analyzed based on the extracted subset of features and the proposed system is used to classify the seven classes of rice leaf disease. The experimental results show that the Optimized SVM has the best classification results among the different classifiers with accuracy of 92.10%, Precision of 92.20%, balanced Recall and F1 Score, which shows that Optimized SVM is very effective in multi-class rice leaf disease classification. The performance can be improved by feature reduction, generalization capability and computational complexity reduction, which are realized with the help of RiceFS. The proposed framework is designed to provide an intelligent decision support system for timely intervention, loss minimization and sustainable agriculture. Results indicate that these algorithms are applicable for rice leaf disease classification since they are accurate, reliable and scalable.

Plant phenotyping relevance

イネ葉の病害状態を観察データから分類する計算手法が研究の中心であり、RiceFSと複数の機械学習器を用いた分類フレームワークを開発・評価しているため。

Code and data availability

The paper's RiceFS phenotyping/classification experiments are built on two public Kaggle rice leaf disease image datasets, explicitly cited with URLs and a data availability statement. No author code or models are deposited.