Diagnosis Anthracnose of Chili Pepper Using Convolutional Neural Networks Based Deep Learning Models.

Area of Science:

• Agricultural Science
• Computer Science
• Plant Pathology

Background:

• Chili pepper (Capsicum annuum L.) is a globally significant crop facing economic losses due to anthracnose disease.
• Accurate and early detection of anthracnose is crucial for mitigating yield losses and maintaining marketability.
• Deep learning, specifically image recognition, presents a promising avenue for automated plant disease diagnosis.

Purpose of the Study:

• To apply deep learning models (MobileNet, ResNet50v2, Xception) using transfer learning for chili pepper anthracnose diagnosis.
• To determine the minimum dataset size required for accurate and efficient disease detection using limited data.
• To evaluate the performance of different models across varying dataset sizes.

Main Methods:

• Utilized transfer learning with deep learning architectures: MobileNet, ResNet50v2, and Xception.
• Trained and evaluated models on datasets ranging from 500 to 4,000 chili pepper images.
• Assessed performance using metrics: precision, recall, F1-score, and accuracy.

Main Results:

• Model performance generally improved with increased dataset size.
• ResNet50v2 and Xception models required larger datasets for optimal accuracy.
• MobileNet demonstrated robust generalization capabilities, performing effectively even with smaller datasets.

Conclusions:

• Transfer learning-based deep learning models are effective for diagnosing chili pepper anthracnose.
• MobileNet offers a practical solution for disease detection when dataset size is limited.
• Findings provide guidelines for optimizing data collection and model selection in agricultural disease diagnostics.