A hybrid vision transformer and ResNet18 based model for biotic rice leaf disease detection

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Rice cultivation is vital for global food security.
• Early detection of rice leaf diseases is critical for maintaining crop yields.
• Existing methods for disease detection require improvement in accuracy and efficiency.

Purpose of the Study:

• To develop a highly accurate hybrid model for rice leaf disease prediction.
• To leverage the strengths of Vision Transformer (ViT) and ResNet18 architectures.
• To enhance crop productivity through improved disease identification.

Main Methods:

• A hybrid model combining Vision Transformer (ViT) and pre-trained ResNet18 (ViT-ResNet18) was proposed.
• Input images were processed independently by ViT and ResNet18.
• Features from both models were fused and classified using a Fully Connected (FC) layer and Softmax.

Main Results:

• The ViT-ResNet18 model achieved an accuracy of 94.4%.
• Performance metrics included a precision of 0.948, recall of 0.944, F1-Score of 0.942, and AUC of 0.985.
• The hybrid model demonstrated significant accuracy improvements over VGG16, Inception V3, and SqueezeNet classifiers.

Conclusions:

• The proposed ViT-ResNet18 model offers a robust and accurate solution for rice leaf disease detection.
• This approach can significantly contribute to increasing agricultural productivity.
• The hybrid deep learning model shows superior performance compared to other established neural network classifiers.