DeepVarCaneNet: Attention-Driven CNN–Transformer Model for Fine-Grained Sugarcane Variety Recognition

DOI

10.2991/978-94-6239-789-7_33How to use a DOI?

Keywords

Hybrid Deep Learning Model; Attention Mechanism (CBAM); Contrastive Learning; Precision Agriculture; Image-Based Crop Classification

Abstract

To effectively classify the sugarcane varieties using the images of the leaves, this paper proposes a new deep learning architecture, DeepVarCaneNet. The conventional approaches of sugarcane variety identification are very dependent on manual examination of morphological characteristics, which are time consuming, knowledge-based and subject to error particularly when there is a high inter-class similarity. To address these drawbacks, the model suggested is an EfficientNet-based CNN to extract local features, a Vision Transformer (ViT) to acquire global contextual relationship, and CBAM to refine better features.Also, a contrastive learning approach is integrated to enhance the feature separability and model generalization in low-data settings.The model is trained using a self-curated dataset of five sugarcane varieties CO0118, CO0238, COOP8272, COS13235, and PB99. The experimental findings show that DeepVarCaneNet has a higher classification accuracy of 98.72 than the traditional architectures like MobileNetV2, ResNet50, EfficientNet and single ViT models. Confusion matrix, ROC curve and precision-recall performance evaluation show that the class separability is high and the misclassification error is low, even between visually similar varieties. In addition, validation curves and training curves ensure the stability of convergence and low overfitting.The hybrid architecture proposed is an effective way to combine both local and global feature learning and attention mechanisms, which can be extremely applicable in real-life contexts related to precision agriculture.

Copyright

© 2026 The Author(s)

Open Access

Open Access This chapter is licensed under the terms of the Creative Commons Attribution-NonCommercial 4.0 International License (http://creativecommons.org/licenses/by-nc/4.0/), which permits any noncommercial use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.

Download article (PDF)

TY  - CONF
AU  - Gautam Tyagi
AU  - Satender Kumar
PY  - 2026
DA  - 2026/10/08
TI  - DeepVarCaneNet: Attention-Driven CNN–Transformer Model for Fine-Grained Sugarcane Variety Recognition
BT  - Proceedings of theInternational Conference on Innovation and Practices in Artificial Intelligence (ICIPAI 2026)
PB  - Atlantis Press
SP  - 484
EP  - 494
SN  - 1951-6851
UR  - https://doi.org/10.2991/978-94-6239-789-7_33
DO  - 10.2991/978-94-6239-789-7_33
ID  - Tyagi2026
ER  -