Infectious diseases of Rice plants classified using a deep learning-powered Least Squares Support Vector Machine Model

Abstract

In the field of agriculture, identifying and classifying Oryza Sativa diseases is always a hot research topic. Almost in all the states of India, Oryza Sativa (rice) is cultivated as a necessary food crop. However, rice plants are highly caused due to diseases that drastically affect the agricultural sectors. Researchers are searching for solutions for reliable and exact detection approaches for crop leaf diseases. A Oryza Sativa plant leaf disease classification system that is modelled relying on the concept of deep learning-driven Least Square Support Vector Machine is investigated in the present study. This work focus on the Oryza Sativa plant leaf diseases like Brown Leaf spot(BS), Bacterial leaf blight (BLB), Leaf smut(LS), Tungro(LT), and Leaf Blast(LB). To create a contrast-enhanced image dataset, the input Oryza Sativa plant disease image dataset is resized as well as histogram equalization and the Kuwahara filter are applied. The diseased part is segmented using spatial fuzzy clustering (SFC). Then, a multi-model LeNet5 is used in extracting features, and the least square support vector machine(LSSVM) is employed to classify the Oryza Sativa plant leaf diseases images. Further, to improve the accuracy of the disease classification Weighted Sparrow Search Optimization (WSSO) is utilized. The proposed multi-model LSSVM-WSSO outperformed the existing models on the two benchmark datasets when evaluated on the terms of accuracy, precision, recall, and F1-measure values.

Performance Comparison between SVM and LS-SVM for Rice Leaf Disease detection

“…For multi class classification, performance is approximately same based on Euclidian distance. N V Raja Reddy et al [9] proposed a model for infectious rice plant diseases using a hybrid deep learning and machine learning approach where they investigated 5 diseases of Oryza Sativa like Brown Spot, Leaf Blast, Leaf Smut, Tungro, Bacterial Leaf Blight. In pre-processing stage, they resized the image followed by histogram equalization and Kuwahara filtering technique.…”

“…Similarly, Chen et al [15] used DenseNet 201 for feature extraction followed by SVM for DS1 dataset and achieved an accuracy of 94%. Rajareddy et al [9] used DS1 and DS2 where LeNet network is used for feature extraction following LS-SVM classifier along with Weighted Sparrow Search Optimization (WSSO) for fine tuning of one of the parameters and achieved an improved accuracy of 98.55% for DS1 and 98.37% for DS2 respectively. Sethy et al [10] applied 13 number of transfer learning models followed by SVM where best performance is achieved by ResNet 50 with an accuracy of 98.3% for DS2 dataset.…”