• Version
• Download 0
• File Size 799.26 KB
• File Count 1
• Create Date 5 March 2026
• Last Updated 5 March 2026

Vision Transformer Based Automated Detection of Harmful Farm Insects

Ketan Kanjiya1, Piyush Sonani2, Upendrasinh Zala3

1Chief Research Officer, Kshatrainfotech, Ahmedabad, Gujarat, India

2Chief Technology Officer, Kshatrainfotech, Ahmedabad, Gujarat, India

3Chief Executive Officer, Kshatrainfotech, Ahmedabad, Gujarat, India

Abstract - Early and accurate detection of harmful farm insects is essential for ensuring agricultural productivity, food security, and sustainable farming practices. This study proposes a Vision Transformer based deep learning framework for automated multi class insect detection in agricultural environments. The model leverages global self-attention mechanisms and transfer learning from large scale pretraining to capture both local visual features and long range contextual relationships, enabling robust fine grained insect classification. A balanced learning strategy incorporating random oversampling and data augmentation is employed to address class imbalance and improve model generalization. The framework is evaluated on a multi-class insect image dataset containing 15 agriculturally significant species under realistic field conditions. Experimental results demonstrate stable convergence, strong generalization, and reliable classification performance across diverse insect categories. The proposed system provides a scalable and intelligent solution for precision agriculture, supporting early pest identification, targeted intervention, and data driven crop protection strategies. This work highlights the potential of transformer based architectures for advancing automated pest monitoring and sustainable agricultural management.

Key Words:  Vision Transformer, Agricultural Pest Detection, Farm Insect Classification, Deep Learning, Computer Vision, Smart Farming