Multiclass Paddy Disease Detection Using Filter-Based Feature Transformation Technique

Pests and diseases are the big issues in paddy production and they make the farmers to lose around 20% of rice yield world-wide. Identification of rice leaves diseases at early stage through thermal image cameras will be helpful for avoiding such losses. The objective of this work is to implement a Modified Lemurs Optimization Algorithm as a filter-based feature transformation technique for enhancing the accuracy of detecting various paddy diseases through machine learning techniques by processing the thermal images of paddy leaves. The original Lemurs Optimization is altered through the inspiration of Sine Cosine Optimization for developing the proposed Modified Lemurs Optimization Algorithm. Five paddy diseases namely rice blast, brown leaf spot, leaf folder, hispa, and bacterial leaf blight are considered in this work. A total of six hundred and thirty-six thermal images including healthy paddy and diseased paddy leaves are analysed. Seven statistical features and seven Box-Cox transformed statistical features are extracted from each thermal image and four machine learning techniques namely K-Nearest Neighbor classifier, Random Forest classifier, Linear Discriminant Analysis Classifier, and Histogram Gradient Boosting Classifier are tested. All these classifiers provide balanced accuracy less than 65% and their performance is improved by the usage of feature transform based on Modified Lemurs Optimization. Especially, the balanced accuracy of 90% is achieved by using the proposed feature transform for K-Nearest Neighbor classifier.