A zero-shot learning framework for chilli leaf disease detection, classification and severity estimation using...

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Late identification of chilli leaf diseases impacts agricultural productivity.
• Conventional deep learning models struggle with novel or emerging disease patterns.
• Manual inspection and model retraining are current limitations.

Purpose of the Study:

• To develop a framework for identifying unseen chilli leaf diseases.
• To estimate disease severity using lesion area ratios.
• To improve real-time agricultural disease monitoring.

Main Methods:

• Proposed a Zero-Shot Dual-Encoder Framework integrating a Vision Transformer (ViT) and RoBERTa-based semantic encoder.
• Utilized a curated chilli leaf dataset under diverse environmental conditions.
• Evaluated the model's performance in recognizing unseen disease categories and estimating severity.

Main Results:

• Achieved 98.7% accuracy, 98.0% precision, and 98.0% recall.
• Outperformed five existing state-of-the-art approaches.
• Demonstrated enhanced recognition accuracy and scalability.

Conclusions:

• The Zero-Shot Dual-Encoder Framework effectively identifies unseen chilli leaf diseases and estimates severity.
• The model offers a scalable solution for real-time agricultural disease monitoring.
• No retraining is required for new or emerging disease patterns.