Carica papaya Leaf Disease Prediction and Classification Based on Hyperparameter-Tuned ResNet18

Abstract

One of the essential components of human civilisation is agriculture. It contributes to economic development by providing food security, generating employment, and much more. According to the National Horticulture Board (NHB), India is one of the top suppliers of papaya, accounting for nearly 38.6% of worldwide production and exporting to the Gulf countries for its nutritional appeal and industrial usage. Regarding the economic welfare of any country, recognition of leaf diseases plays a crucial role. Plant organs are infested by viruses, fungi, bacteria, and other organisms. The existing model performs classification using Convolutional Neural Networks (CNN), Transfer Learning, and the MobileNet architecture to categorise papaya leaf diseases. This paper makes use of deep learning techniques to identify and classify papaya leaf diseases based on a deep learning approach, i.e., the ResNet18 architecture. ResNet18 architecture was retrained using images of papaya leaves with Ring Spot, Anthracnose, Curl, and Bacterial Spot, including Healthy and was employed it to create a model successfully. A total of 1010 samples were trained, and 210 samples were tested. The work made use of classification accuracy to find the F1 score, precision, recall, and accuracy for each disease and got 93.55% as the final accuracy.