The Hybrid Deep Learning Framework to Identify Real World Papaya Leaf to Detect Disease by Approximating the Severity

View at publisher Summarise with VIRAT
Authors: Joan Yazhini J, Nikhita Sherlin Christopher, Sowmiya L
Cited by: 0
References: 15

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Despite the fact that papaya is a staple of tropical farming, the growth of the crop is frequently crippled by viral infections which affect the foliage and loss of output is a norm. The need to detect and accurately quantify the severity of infection in time comes to the fore, so as to reduce the cost of loss, and to take corrective actions on time. A second problem of the real-world detection is clutter field scene, changeable illumination patterns and partial foliage coverage. To overcome this challenge, the disease classification and severity levels are combined to present the Hybrid Deep Learning Framework of intelligent papaya leaf disease management. It consists of a transfer learning-explained joint of both Xception and DenseNet201 backbones to learn representations of non-uniform vein patterns and surface textures. The mask-aware severity estimator applies the infection proportionally to the surfaces of the leaves after classification to provide actionable and transparent disease information. To enable the capacity to make decisions at a contextual level, a context sensitive feature such as climatic variations and geo-spatial tags are incorporated thus allowing a series of agro-ecological zones to adapt. This methodology opposes to exhaustive augmentation methods and generative approaches, which enables data augmentation of such challenging and uncontrolled real-field situations which require increasing ground truth data that impiles generalization directions, even though the data insufficiency problem exists. The framework should also offer AI-based treatment recommendations in the local languages, using an edge-computing-based mobile platform which can be implemented in rural settings with limited connectivity. The research paper builds a farmer friendly, interpretable solution, which not only helps in preserving sustainable farming, but also helps in precision agriculture, by implementing the deep learning models in practical contexts..

Fields of Study & Subjects

Smart Agriculture and AI Advanced Neural Network Applications Leaf Properties and Growth Measurement

Publication details

Journal / source
International Conference on Communication, Computing and Electronics Systems (ICCCES)
Pages
1292–1297
Publication date
21 January 2026
Language
EN