Severity identification of Potato Late Blight disease from crop images captured under uncontrolled environment

Plant disease management is an important factor in agriculture as it causes a significant yield loss in crops. Late Blight is the most devastating disease for Potato in most of the potato growing regions in the world. For optimum use of pesticide and to minimize the yield loss, the identification of disease severity is essential. The key contribution here is an algorithm to determine the severity of Potato Late Blight disease using image processing techniques and neural network. The proposed system takes images of a group of potato leaves with complex background as input which are captured under uncontrolled environment. In this proposed approach decorrelation stretching is used to enhance the color differences in the input images. Then Fuzzy C-mean clustering is applied to segment the disease affected area which also include background with same color characteristics. Finally we propose to use the neural network based approach to classify the disease affected regions from the similar color textured background. The proposed algorithm achieves an accuracy of 93% for 27 images captured in different light condition, from different distances and at different orientations along with complex background.