Rose Diseases Recognition Using MobileNet

Abstract

Plants always prove a great assessment of human life for many years in many sectors. Nowadays plant diseases are affecting our agricultural sector very badly. As a result, farmers are facing huge losses. For developing an early treatment process, the exact and fastest detection of plant diseases can help to reduce huge economical suffering. To detect rose diseases manually we need expert knowledge about rose diseases which is very complex, time taking and tiring. In this paper, we have used transfer learning and without transfer learning technique by using a MobileNet model to detect rose diseases. Augmentation has been performed on the collected image data for the lack of many images. For experimental purpose, 1600 data images are used to train the model and 400 data images are used to test the model. For evaluating our empirical eventuality we have reckoned the F1 score beside the model's exactitude and used the ROC curve to compare the result generated using both techniques. Using MobileNet with transfer learning technique for each class we get better accuracy and F1 score than without transfer learning. Within two approaches, MobileNet with transfer learning omits the MobileNet without transfer learning technique and achieves 95.63% accuracy. The acquired result exhibits that the working method for recognizing rose diseases is appeasement and feasible.