A New Siamese Network Loss for Cattle Facial Recognition in a Few-Shot Learning Scenario

Author:

Porto João12,Higa Gabriel1,Weber Vanessa23,Weber Fabrício2ORCID,Loebens Newton14ORCID,Claure Pietro3,de Almeida Leonardo5,Porto Karla2,Pistori Hemerson15ORCID

Affiliation:

1. Inovisão Department, Universidade Católica Dom Bosco, Campo Grande 79117-900, Brazil

2. Kerow Soluções de Precisão, Campo Grande 79117-900, Brazil

3. Faculty of Engineering, Universidade Estadual de Mato Grosso do Sul, Campo Grande 79115-898, Brazil

4. Faculty of Applied Mathematics, Universidade Federal do Pampa, Itaqui 97650-000, Brazil

5. Faculty of Engineering, Universidade Federal de Mato Grosso do Sul, Campo Grande 79070-900, Brazil

Abstract

This study explores the use of a Siamese neural network architecture to enhance classification performance in few-shot learning scenarios, with a focus on bovine facial recognition. Traditional methodologies often require large datasets, which can significantly stress animals during data collection. In contrast, the proposed method aims to reduce the number of images needed, thereby minimizing animal stress. Systematic experiments conducted on datasets representing both full and few-shot learning scenarios revealed that the Siamese network consistently outperforms traditional models, such as ResNet101. It achieved notable improvements, with mean values increasing by over 6.5% and standard deviations decreasing by at least 0.010 compared to the ResNet101 baseline. These results highlight the Siamese network’s robustness and consistency, even in resource-constrained environments, and suggest that it offers a promising solution for enhancing model performance with fewer data and reduced animal stress, despite its slower training speed.

Funder

Dom Bosco Catholic University

Foundation for the Support and Development of Education, Science and Technology from the State of Mato Grosso do Sul, FUNDECT

Federal University of Pampa, UNIPAMPA

Brazilian National Council of Technological and Scientific Development, CNPq

Coordination for the Improvement of Higher Education Personnel, CAPES

Publisher

MDPI AG

Reference40 articles.

1. Das Indústrias Exportadoras de Carne (ABIEC) (2024, January 21). Perfil da Pecuária no Brasil. Associação Brasileira das Indústrias Exportadoras de Carne, Available online: https://www.abiec.com.br/.

2. Milanez, A.Y., Mancuso, R.V., Maia, G.B.d.S., Guimarães, D.D., Alves, C.E.A., and Madeira, R.F. (2024, January 21). Conectividade Rural: Situação Atual e Alternativas para Superação da Principal Barreira à Agricultura 4.0 No Brasil, Available online: https://web.bndes.gov.br/bib/jspui/handle/1408/20180.

3. Neto, A., Nicola, S., Moreira, J., and Fonte, B. (2021, January 16–18). Livestock Application: Naïve Bayes for Diseases Forecast in a Bovine Production Application: Use of Low Code. Proceedings of the International Conference on Innovations in Bio-Inspired Computing and Applications, Cham, Switzerland.

4. Identification and Recognition of Animals from Biometric Markers Using Computer Vision Approaches: A Review;Cihan;Kafkas Univ. Vet. Fak. Derg.,2023

5. Usage of few-shot learning and meta-learning in agriculture: A literature review;Dorsa;Smart Agric. Technol.,2023

同舟云学术

1.学者识别学者识别

2.学术分析学术分析

3.人才评估人才评估

"同舟云学术"是以全球学者为主线,采集、加工和组织学术论文而形成的新型学术文献查询和分析系统,可以对全球学者进行文献检索和人才价值评估。用户可以通过关注某些学科领域的顶尖人物而持续追踪该领域的学科进展和研究前沿。经过近期的数据扩容,当前同舟云学术共收录了国内外主流学术期刊6万余种,收集的期刊论文及会议论文总量共计约1.5亿篇,并以每天添加12000余篇中外论文的速度递增。我们也可以为用户提供个性化、定制化的学者数据。欢迎来电咨询!咨询电话:010-8811{复制后删除}0370

www.globalauthorid.com

TOP

Copyright © 2019-2024 北京同舟云网络信息技术有限公司
京公网安备11010802033243号  京ICP备18003416号-3