A Transfer Learning Approach for Leaf Image based Classification of Healthy and Diseased Leaves

Background: India is mainly an agriculturist country where agriculture is a very important sector for Indian economy. Different varieties of crops are grown in huge acres of land yielding sufficient quantity. But due to frequent attacks of pests and pathogens the plants gets infected and develops different diseases. These diseases affect the quality of crop production and put the consumer’s health at risk. Thus, it is important to keep a track on the health of the plants on a regular basis so that required action can be taken without wasting any time. Therefore the study aimed to propose a method for automatic classification of healthy and diseased plant leaves using transfer learning approach. 
Methods: Deep learning is gaining popularity among researchers due to its several advantages. Two approaches were used in deep learning: one is to build a model from scratch and the other to use pre-trained models (PTMs). This concept is known as transfer learning. It is based on the concept where the network learns from existing models and uses the knowledge to solve a completely different problem. These models are trained on huge datasets to solve a task that is similar to our desired problem. Mobile Net V2 has been used for our research problem and the beans dataset which is a publicly available dataset. 
Result: Mobile Net V2 is a very light weight model which is trained on more than 1.4 M images. We have downloaded the pre-trained model and fine tuned the last layer to solve our desired task. The results observed that in just only 30 epochs our proposed model could achieve a classification accuracy of 98%. Also, we have compared our proposed method with other existing methods in literature and found that our proposed method had higher classification accuracy than other existing methods.