Advancing Early Blight Detection in Potato Leaves Through ZeroShot Learning.

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Potato cultivation faces significant yield loss due to early blight, a fungal disease caused by *Alternaria solani*.
• Accurate and early detection of plant diseases is crucial for effective crop management and yield preservation.

Purpose of the Study:

• To introduce a novel deep learning framework, ZeroShot CNN, for classifying potato diseases, including previously unseen classes.
• To evaluate the performance of ZeroShot CNN against conventional methods for early blight detection.

Main Methods:

• Developed a hybrid deep learning model integrating convolutional neural networks (CNNs) for feature extraction and ZeroShot Learning (ZSL) for classification.
• Utilized semantic embedding techniques within the CNN architecture to enable identification of untrained disease classes.
• Trained and tested the model on the Kaggle potato disease dataset.

Main Results:

• ZeroShot CNN achieved high accuracy: 98.50% for seen disease categories and 99.91% for unseen categories.
• The model demonstrated superior generalization capabilities compared to traditional methods.
• The framework provides a scalable, real-time solution for agricultural disease detection.

Conclusions:

• The developed ZeroShot CNN framework offers a powerful and efficient tool for identifying potato diseases, including novel strains.
• Deep learning combined with ZeroShot inference holds significant potential for advancing plant pathology and crop protection strategies.
• This approach can contribute to improved food security by mitigating crop losses due to diseases.