Flower Classification and Key Parameter Analysis Based on ViT

Ruochen Deng

2024

Abstract

Flower classification holds significant implications for various fields, including plant resource survey and plant taxonomy education. This paper proposes employing the Vision Transformer (ViT) model for flower classification tasks. The study aims to investigate the impact of varying depth and head parameters in ViT model on their performance. Through an analysis of accuracy performance and attention properties, the research explores optimal strategies for setting depth and head parameters. Additionally, it delves into the phenomenon of attention collapse within the multi-head attention mechanism, utilizing mean attention distance plots for in-depth analysis. Results reveal a positive correlation between model depth, number of heads, and classification accuracy. Moreover, insights gleaned from attention collapse observations provide valuable guidance for optimizing depth and head parameter settings. This study offers valuable insights into the performance of ViT models in flower classification tasks, while also contributing to the understanding of depth and head parameters in self-attention mechanisms for future research endeavors.

Paper Citation

in Harvard Style

Deng R. (2024). Flower Classification and Key Parameter Analysis Based on ViT. In Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI; ISBN 978-989-758-713-9, SciTePress, pages 194-199. DOI: 10.5220/0012922900004508

in Bibtex Style

@conference{emiti24,
author={Ruochen Deng},
title={Flower Classification and Key Parameter Analysis Based on ViT},
booktitle={Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI},
year={2024},
pages={194-199},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0012922900004508},
isbn={978-989-758-713-9},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 1st International Conference on Engineering Management, Information Technology and Intelligence - Volume 1: EMITI
TI - Flower Classification and Key Parameter Analysis Based on ViT
SN - 978-989-758-713-9
AU - Deng R.
PY - 2024
SP - 194
EP - 199
DO - 10.5220/0012922900004508
PB - SciTePress