Detection of Diseases in Potato Leaves Using Transfer Learning

Abstract

The problem of food shortage has grown rampant in the recent times in developing countries. In a tropical country like India, potato is one of the major staple food that is eaten throughout the year. Recently the production of potato is falling short due to various diseases like Early Blight and Late Blight which cause a huge loss of the cropped plants. This also leads to a major loss in the national economy as well. The emergence of deep learning has affected many fields of machine learning research. Since it is not required in deep learning to develop hand-crafted features, it has found widespread adoption in the scientific community. To tackle the need for a huge amount of data for deep learning, another heavily implemented technique is used, namely, transfer learning, to make the training process faster and more accurate with a relatively small dataset at hand. The performance of the model is demonstrated both quantitatively by computing the accuracy metric as well as visually. The model is lightweight and robust and thus can be added to an application in a handheld device like smartphone so that crop growers could spot the disease affected crops on the go and save them from getting ruined.