Applying Multiclass Classification For Leaf Disease Detection Using Hybrid Feature Extraction Approach

Abstract

The roots of agriculture in India are planted since the era of Indus Valley Civilization and in producing farm output, India ranks second highest. In Indian economy, agriculture sector provides employment to 50% of the workforce and it contributes 18% in the GDP of the country.  Diseases and infection are the major constraints towards crop yield and quality. Timely identification of symptoms can significantly control spreading of Infectious Diseases and help in optimum yield generation. Automated way of identification and detection of disease requires less efforts and able to produce more accurate result. Computer vision and machine learning are two promising field to develop the automated system that identifies and classifies the leaf disease. The goal of the proposed work is to diagnose the disease using image processing and classification techniques. A leaf image is pre-processed, segmented and 31 features including color, shape, vein, GLCM and Gabor texture and zernik are extracted. A prediction model is developed and different classifiers are experimented on the two different dataset containing 10,000 and 3,000 images of disease affected leafs. The experimental results are evaluated and compared with Random Forest, SVM, KNN and ANN. The proposed method with Random Forest classifier achieved highest accuracy amongst all other classifiers.