Agricultural innovation through deep learning: a hybrid CNN-Transformer architecture for crop disease classification

Smitha Padshetty,Ambika Umashetty
DOI: https://doi.org/10.1080/14498596.2024.2355225
2024-06-08
Journal of Spatial Science
Abstract:A novel deep learning approach for crop disease classification using a Hybrid Convolutional Neural Network (CNN)-Transformer architecture is proposed. Utilizing the Plant Village, corn/maize leaf, and rice disease image datasets, the preprocessing retains essential features to enhance classification accuracy. The CNN extracts local features, while the Vision Transformer (ViT) captures global contexts by segmenting input images into patches. The Feature Attention Module (FAM) enhances crucial features and captures global aspects. The multilevel feature fusion network integrates these features, converting the fused vector into class probabilities. The proposed crop disease classification model provided a higher classification accuracy of 98.92%.
geography, physical,remote sensing
What problem does this paper attempt to address?