A Comparative Study of Convolutional Neural Networks and Vision Transformers for Sugarcane Leaf Disease Classification in Real-World Environments

This study aimed to investigate and compare the performance of two deep learning architecture groups, namely Convolutional Neural Networks (ResNet-50 and DenseNet-121) and Vision Transformers (ViT-Base and Swin-Tiny), for classifying four types of sugarcane leaf conditions, namely healthy leaves, red wilt, ring spot, and rust, under real-world field conditions. The dataset comprised 4,000 high-resolution sugarcane leaf images, balanced across four classes with 1,000 images per class, and was partitioned into training, validation, and testing sets at a ratio of 80:10:10. All models were trained from scratch on an NVIDIA Tesla T4 GPU using PyTorch 1.12 under identical hyperparameter settings, including a batch size of 16, 15 epochs, and the Adam optimizer with a learning rate of 0.001. Experimental results demonstrated that DenseNet-121 achieved the highest performance across all evaluation metrics, with an overall accuracy of 99.25%, and Precision, Recall, and F1-Score values of 0.9926, 0.9925, and 0.9925, respectively, while requiring the shortest training time of only 10 minutes and 56 seconds. ResNet-50 and Swin-Tiny followed with accuracies of 99.00% and 98.50%, respectively, whereas ViT-Base achieved the lowest accuracy of 97.00%. These findings indicate that CNN architectures are more suitable than Vision Transformer architectures for limited datasets because their inherent inductive bias is better suited for extracting discriminative features from small disease lesions. The findings of this study provide a practical foundation for developing automated plant disease diagnosis systems on mobile devices to support smart agriculture.

Authors

Institutions

Publication Details

Journal
The Journal of King Mongkut s University of Technology North Bangkok
Published
2026-11-20
DOI
https://doi.org/10.14416/j.kmutnb.2026.11.004
Primary Topic
Smart Agriculture and AI
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

A Comparative Study of Convolutional Neural Networks and Vision Transformers for Sugarcane Leaf Disease Classification in Real-World Environments

Tinnakorn Kunasit, Jackaphan Sriwongsa, Supachai Baipod
2 links
The Journal of King Mongkut s University of Technology North Bangkok
Smart Agriculture and AI
article

A Comparative Study of Convolutional Neural Networks and Vision Transformers for Sugarcane Leaf Disease Classification in Real-World Environments

Tinnakorn Kunasit, Jackaphan Sriwongsa, Supachai Baipod
article en

Abstract

This study aimed to investigate and compare the performance of two deep learning architecture groups, namely Convolutional Neural Networks (ResNet-50 and DenseNet-121) and Vision Transformers (ViT-Base and Swin-Tiny), for classifying four types of sugarcane leaf conditions, namely healthy leaves, red wilt, ring spot, and rust, under real-world field conditions. The dataset comprised 4,000 high-resolution sugarcane leaf images, balanced across four classes with 1,000 images per class, and was partitioned into training, validation, and testing sets at a ratio of 80:10:10. All models were trained from scratch on an NVIDIA Tesla T4 GPU using PyTorch 1.12 under identical hyperparameter settings, including a batch size of 16, 15 epochs, and the Adam optimizer with a learning rate of 0.001. Experimental results demonstrated that DenseNet-121 achieved the highest performance across all evaluation metrics, with an overall accuracy of 99.25%, and Precision, Recall, and F1-Score values of 0.9926, 0.9925, and 0.9925, respectively, while requiring the shortest training time of only 10 minutes and 56 seconds. ResNet-50 and Swin-Tiny followed with accuracies of 99.00% and 98.50%, respectively, whereas ViT-Base achieved the lowest accuracy of 97.00%. These findings indicate that CNN architectures are more suitable than Vision Transformer architectures for limited datasets because their inherent inductive bias is better suited for extracting discriminative features from small disease lesions. The findings of this study provide a practical foundation for developing automated plant disease diagnosis systems on mobile devices to support smart agriculture.

The Journal of King Mongkut s University of Technology North BangkokVol. 37(1)
Mahasarakham University (TH), Rajabhat Maha Sarakham University (TH)
Zero hunger
Smart Agriculture and AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon
A Comparative Study of Convolutional Neural Networks and Vision Transformers for Sugarcane Leaf Disease Classification in Real-World Environments — Tinnakorn Kunasit, Jackaphan Sriwongsa, et al. · The Journal of King Mongkut s University of Technology North Bangkok (2026) | TGRS Research Map | TGRS