Generalized Technique for Potato Leaves Disease Classification Using Convolutional Neural Network

Abstract

In today’s world, there is a growing demand for potatoes that are produced professionally, successfully, and sustainably. This is due to changing climate conditions, worldwide population growth, and changing consumer demand. It is more crucial than ever to increase production capacity in order to satisfy future demand. However, insects and diseases have a significant impact on potato yield stability. Potato leaf disease is one major concern. Early diagnosis of the disease and immediate action can prevent further damage. Therefore, this paper proposes a generalized technique to diagnose late blight and early blight potato leaf disease using a convolutional neural network (CNN). The proposed technique comprises segmentation, augmentation, training and testing, model evaluation, and disease diagnosis components. We used the GrabCut approach, which works in combination with foreground extractions, to segment database images. Data augmentation techniques, including flip, rotation, zoom, and shift operation employed for improving the performance of the proposed approach. Our model evaluated with the three leaves disease datasets, namely potato leaf disease (PLD), new plant disease (NPD), and PlantifyDr (PD). We demonstrated the evaluation performance accuracy of each dataset separately, and then we showed the performance with combined images from all the datasets using the proposed generalized method.