A comparative analysis on plant pathology classification using deep learning architecture – Resnet and VGG19

Abstract Agriculture is one of the most significant roles in the growth and development of our nation's economy. The identification of plant disease is the key to prevent the losses in the yield and quantity of the agricultural product. Disease detection on the plant is very critical for sustainable agriculture. It is challenging to monitor the plant diseases manually, especially those who are new to farming. It requires excessive processing time. Therefore, a proper prediction and detection of plant disease will reduce the use of fertilizers in the field, which helps from soil impurities and also helps the farmers to yield the right products. Hence, the food industry will not get affected. Researchers have brought out many techniques, but still, agricultures face many problems related to plant disease. In this paper, we have studied plant diseases in apple plants. The diseases in apple plants are categorized into four different types scab, healthy, multiple diseases, and rust. Several machine learning algorithms developed for early-prediction of disease in the leaves of apple plants, thereby alerting the farmers. In this paper, the performance of two deep learning algorithms, such as ResNet50 and VGG19, is compared in classifying and predicting the apple leaf diseases. The Kaggle dataset is taken for detecting and classifying the disease in apple leaf. The dataset contains 3,651 real-time images to classify four categories: scab, healthy, multiple diseases, and rust in the apple plant. The images are captured in the apple orchid under different environmental conditions such as illumination, varying backgrounds, view-invariant, and various noises. The dataset is riven into 80% training and 20% validation. The popular architectures such as ResNet50 and VGG19 produce a better performance during training, and they can predict the leaf disease with an accuracy of 87.7%.