Indian Journal of Science and Technology

Abstract

Background/Objectives: Papaya leaf being an excellent source of bioactive compounds plays a crucial role in the formulation of Ayurvedic remedies, irrespective of medicinal usage Papaya leaves are frequently affected by diseases which harm the crop and decrease its productivity. Hence, it urges for disease detection. Methods/Statistical analysis: Utilizing computer vision methods to detect diseases, presents a solution to the limitations of constant human supervision. The study introduces a transfer learning model built upon the Resnet-50 architecture to recognize and classify diseased papaya leaves. On a dataset of 2159 images, 1726 images are allocated for training, 213 for validation and 220 for testing. The classes we distinguish in our study include healthy leaves as well as those afflicted by anthracnose, bacterial spot, curl, and ringspots. Findings: The proposed model has been pre-trained and fine-tuned on this dataset, and when evaluated using a sample set of 220 images, it achieves an impressive accuracy rate of 87.95%. Notably, this model surpasses the performance of the base models, including CNN, VGG 16, Inception V3, ResNet-50, DenseNet 121, MobileNet V2, and EfficientNet B0 in the classification task. Novelty/Applications: This work emphasizes the practicality of the proposed approach in real-world applications and its importance in agriculture and disease management. It paves the way for significant revolutions in food security and contributes to environmental conservation and economic stability.

Keywords: CNN, Papaya Leaf Diseases­ Anthracnose, Bacterial Spot, Curl and Ring Spot, Transfer Learning