Visual Intelligence in Precision Agriculture: Exploring Plant Disease Detection via Efficient Vision Transformers.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Agricultural development is vital for economic growth, but plant diseases significantly impede crop yield and quality.
• Accurate plant disease identification is challenging due to a lack of experts and low-contrast visual information, necessitating automated early detection methods.
• Convolutional Neural Network (CNN) models face limitations in preserving crucial spatial information during dimensionality reduction, impacting precise feature localization.

Purpose of the Study:

• To introduce GreenViT, a novel fine-tuned technique for detecting plant infections and diseases.
• To leverage the capabilities of Vision Transformers (ViTs) to address the information loss issues inherent in CNN-based models.
• To evaluate the performance of GreenViT against state-of-the-art CNN models on benchmark datasets.

Main Methods:

• The proposed GreenViT method adapts Vision Transformers (ViTs) for plant disease detection.
• Input images are segmented into smaller patches, analogous to word embedding, and processed sequentially by the ViT.
• The approach is designed to capitalize on ViT's strengths to overcome CNN limitations in feature localization.

Main Results:

• GreenViT was evaluated on widely used benchmark datasets for plant disease detection.
• Experimental results demonstrate that GreenViT significantly outperforms state-of-the-art CNN models.
• The proposed method shows superior performance in accurately identifying plant diseases.

Conclusions:

• GreenViT offers a more effective approach to plant disease detection compared to traditional CNN models.
• The use of Vision Transformers (ViTs) in GreenViT enables precise feature identification, crucial for early disease diagnosis.
• This advancement supports the development of improved agricultural management systems for enhanced crop health and productivity.