Abstract:
Identification of plant disease is beneficial not to protect plants from diseases but also to increase the budgetary advancement of a country. The study of plant disease identification means the study of visually pattern seen on the plant. The traditional approach adopted for plant disease detection was naked eye observation of expert but it is very complex to find the plant disease manually because it requires high processing time and expertise in plant’s diseases. So, it was necessary to develop a system that detect the plant disease in less time and cost effective manner. Hence, we have applied convolution neural networks (CNNs) architectures VGG19 and ResNet50 on both laboratory and natural images of plants for the identification of plant disease. We have trained and tested our CNN models on 10,066 images which contains the steps likes image capturing, image pre-processing, data augmentation, CNN training and testing. Furthermore, we have trained and tested our models on mixture of laboratory and natural images but the amount of natural images plus classes was greater than laboratory images and classes of plants which contain diseases. Results in that ResNet50 outperformed in detection of plant disease.