EnC-SVMWEL: Ensemble Approach using CNN and SVM Weighted Average Ensemble Learning for Sugarcane Leaf Disease Detection

Plant diseases have always been a challenge for crop yield and plant growth, which can have detrimental consequences on food availability worldwide. For the identification of disease in plants Sugarcane, a major commercial crop, has a high net production value. However, the rate of production is decreasing every year due to multifarious diseases. Detecting the presence of disease and its infestation type becomes crucial for diagnosing such diseases before complete contamination of the crop to prevent financial losses. To accomplish this goal, this paper proposes an Ensemble Convolutional-Support Vector Machine Weighted Average Ensemble Learning (EnC-SVMWEL) based sugarcane plant leaf disease detection approach. For experimentation, sugarcane leaf images containing both healthy and unhealthy classes are taken from a sugarcane dataset. Then, the images are preprocessed through grayscale conversion, resizing, and contrast enhancement procedures. From the preprocessed images, some important feature properties are extracted using the DenseNet201 architecture. Finally, the proposed Support Vector Machine Weighted Average Learning (SVMWEL) classifier trains the extracted image features and classifies them into six classes accordingly. To analyze the ability of the proposed approach in detecting sugarcane plant diseases, metrics such as accuracy, precision, recall, and F-measure are measured. The experimental results prove that the proposed approach achieves higher detection accuracy of about 97.45% than other existing methods.