A Lightweight Combined Feature Extraction Model for Tomato Leaf Disease Classification Using DWT–FCM and SVM

Abstract Farmers are facing challenges in tomato leaf disease detection and recognition due to similar appearance of disease symptoms among the different diseases. False or late identification worsens the plant growth and hinders its control measures. Hence, researchers are embracing computerized systems inbuilt with an accurate feature extraction process to highlight the nature of diseases. But they are facing difficulties in categorizing the diseases due to the inseparable closeness among the extracted features. Hence, they are pushing towards the hybridization of various feature extraction models. In this study, a lightweight combined feature extraction model is proposed that reduces the state of inseparability among the features with 28 numbers of features and ensures effective classification of tomato leaf diseases. In the proposed framework, the disease signs are initially segmented by fuzzy c-means clustering techniques, from which 28 features are extracted using the hybrid feature extraction framework and fed to the SVM classifier for classification. The subsequent results are exposed in comparison with k-means clustering. Moreover, the proposed framework classifies the tomato leaf diseases without compromising minimum number of features and maximum number of diseases. Additionally, it demonstrates an accuracy of about 92.5% for the classification of real time tomato leaf disease images. This represents a notable enhancement of 2.5%, 2.44%, and 1.25% in accuracy, specificity, and precision over k-means clustering. In addition, the proposed method overrides other previous studies expressing its effectiveness in classifying images from both plant village database and real-time images.