Plant Disease Detection using Transfer Learning in Precision Agriculture

1304'57?N 8016'30?E Many plants exhibit apparent disease symptoms and thus the method used so far is that a plant pathologist detects the disease through microscopic and optimal observation. The problem with the diagnosis process is that it is slow to perform, and the level of accuracy depends on the pathologist's capability, here this provides an area for computer-aided diagnostic systems in precision agriculture. Thus, deep learning offers a robust tool with highly accurate results. But however, there are several drawbacks like overfitting, and low performance when tested on datasets from new environments. All the above disadvantages can be overcome by capturing images from different angles, in different weather conditions, and at different backgrounds which replicates practical situations. The augmentation techniques like rotation and zooming are used to generate more data. Finally, the model is trained and developed by transfer learning, using pre-trained models like InceptionV3, ResNet and Inception Resnetv2. From the experimentation, it is found that Resnet performed well than other models.