Technical gazette, Vol. 32 No. 4, 2025.
Original scientific paper
https://doi.org/10.17559/TV-20241018002074
Classification of Paddy Crop Disease Using SVM from GLCM Features Obtained from Wavelet Sub-Band Decomposed Images
Karthick Mookkandi
orcid.org/0000-0001-8360-4122
; Department of Electronics and Communication Engineering, National Institute of Technology Puducherry, Karaikal, India
*
Malaya Kumar Nath
; Department of Electronics and Communication Engineering, National Institute of Technology Puducherry, Karaikal, India
Mathumathi Mookkandi
; Department of Computer Science and Engineering, K.Ramakrishnan College of Technology, Trichy, India
Vinoth Kumar Kalimuthu
orcid.org/0000-0002-8920-4936
; Department of CSE (AI&ML), SSM Institute of Engineering & Technology, Dindigul, Tamil Nadu, India
* Corresponding author.
Abstract
Rice serves as a primary dietary staple in many nations, prompting the establishment of expansive rice fields to meet escalating food requirements. Yet, biotic and abiotic stressors impose various health afflictions on crops, significantly compromising both quality and yield. Early-stage disease detection and preventive measures are thus critical in agriculture. Traditional diagnostic approaches, however, are often labor-intensive and financially burdensome. Advanced image processing techniques offer refined solutions for disease detection, severity grading, and prompt intervention, proving highly suitable for extensive farming practices. This work rigorously explores the application of wavelet decomposition (5-level and 7-level) using the db4 mother wavelet, feature extraction by Grey-Level Co-occurrence Matrix (GLCM), and classification through a range of classifiers (such as: support vector machine (SVM), KNN, ensemble, decision tree, and Naiveꞌs Baye) for efficient and precise disease identification. Wavelet decomposition helps capturing the disease information in different wavelet sub-bands. By selecting proper sub-band for a particular level of decomposition helps identifying disease features. Fourteen GLCM features are computed from the selected wavelet sub-band for classification. This approach was experimented on customized dataset, which is comprising of healthy class and seven distinct diseased classes. The performance of the proposed method is evaluated by accuracy, area under the curve (AUC), true positive rate (TPR), false negative rate (FNR), positive predictive value (PPV), false discovery rate (FDR) and MCC. SVM with cubic kernel achieved a training-accuracy and testing-accuracy of 98.12% and of 90.64% for 5-level decomposition. For 7-level decomposition, the training and testing accuracy are found to be 94.48% and 84.45%, respectively.
Keywords
gray level co-occurrence matrix (GLCM); image processing in agriculture; rice disease diagnosis; wavelet decomposition (db4); support vector machine (SVM)
Hrčak ID:
332860
URI
Publication date:
29.6.2025.
Visits: 885 *