Vision-Based Artificial Intelligence in Agriculture: Transformer-Based Plant Health and Disease Detection

Vision-Based Artificial Intelligence in Agriculture

Abstract

Agricultural productivity is increasingly threatened by plant diseases, leading to significant economic losses and food security concerns. Traditional disease detection methods rely on visual inspections by farmers and experts, which are often time-consuming, labor-intensive, and prone to human error. Recent advancements in artificial intelligence (AI), particularly Transformer-based models, offer a promising solution to enhance the accuracy and efficiency of plant disease identification.

This study explores the application of Vision Transformers (ViTs), Swin Transformer and Convolutional Neural Network for plant health monitoring and disease detection. Unlike traditional Convolutional Neural Networks (CNNs), Transformer-based models excel in capturing long-range dependencies within images, enabling more precise and context-aware predictions. By leveraging large-scale agricultural datasets, these models can learn complex visual patterns associated with various plant diseases.

The research methodology includes data collection from publicly available datasets such as PlantVillage, along with custom-labeled images obtained through drones and mobile devices. The images undergo pre-processing, augmentation, and model training using PyTorch and TensorFlow frameworks. Performance metrics such as accuracy, F1-score, and Intersection over Union (IoU) are used to evaluate the effectiveness of Transformer-based models compared to CNN-based approaches.

Preliminary results indicate that ViTs and Swin Transformers outperform CNNs in detecting plant diseases, demonstrating superior generalization capabilities across different crop types and environmental conditions. This research contributes to the field of precision agriculture by showcasing how AI-driven vision models can revolutionize plant disease management. Future work will focus on Edge AI implementations to enable real-time disease detection on low-power mobile devices and drones. Integrating multimodal data sources, such as hyperspectral imaging and soil health indicators, will further enhance model robustness. The findings emphasize the importance of AI in sustainable agriculture, helping farmers make data-driven decisions and reduce pesticide use while ensuring global food security.