Full Text Available
Note: Clicking the button above will open the full text document at the original institutional repository in a new window.
Initial attempt: classification of Simmental cattle based on body conformation using machine learning
Thesis (MSc)--Stellenbosch University, 2025.
Saved in:
| Main Author: | |
|---|---|
| Other Authors: | |
| Format: | Thesis |
| Published: |
Stellenbosch : Stellenbosch University
2026
|
| Subjects: | |
| Tags: |
No Tags, Be the first to tag this record!
|
| _version_ | 1867613985128316928 |
|---|---|
| access_status_str | Open Access |
| author | Schulenburg, Suzette |
| author2 | Grobler, Trienko |
| author_browse | Grobler, Trienko Schulenburg, Suzette |
| author_facet | Grobler, Trienko Schulenburg, Suzette |
| author_sort | Schulenburg, Suzette |
| collection | Thesis |
| dc_rights_str_mv | Stellenbosch University |
| description | Thesis (MSc)--Stellenbosch University, 2025. |
| format | Thesis |
| id | oai:scholar.sun.ac.za:10019.1/134814 |
| institution | Stellenbosch University (South Africa) |
| last_indexed | 2026-06-10T12:44:50.018Z |
| license_str | Other — see source repository |
| provenance_str_mv | Harvested via OAI-PMH from SUNScholar — Stellenbosch University Repository |
| publishDate | 2026 |
| publishDateRange | 2026 |
| publishDateSort | 2026 |
| publisher | Stellenbosch : Stellenbosch University |
| publisherStr | Stellenbosch : Stellenbosch University |
| record_format | dspace |
| source_str | SUNScholar — Stellenbosch University Repository |
| spelling | oai:scholar.sun.ac.za:10019.1/134814 Initial attempt: classification of Simmental cattle based on body conformation using machine learning Schulenburg, Suzette Grobler, Trienko Ngxande, Mkhuseli Stellenbosch University. Faculty of Science. Dept. of Computer Science. Simmental cattle -- Data processing Cattle breeds -- Classification Machine learning Graphics processing units Thesis (MSc)--Stellenbosch University, 2025. Schulenburg, S. 2025. Initial Attempt: Classification of Simmental Cattle Based on Body Conformation Using Machine Learning. Unpublished masters thesis. Stellenbosch: Stellenbosch University [online]. Available: https://scholar.sun.ac.za/items/a2a14b01-5ed7-44e8-903e-9745066ac558 ENGLISH ABSTRACT: The classification of cattle quality, particularly body conformation, is a critical task in livestock management, especially for breeds such as the Simmental, where phenotypic traits play a decisive role in both economic value and breeding decisions. The central contribution of this study is the use of a newly created high‐quality image dataset of Simmental cattle to enhance predic‐ tion accuracy through 5‐fold cross‐validation and ensemble learning methods. This dataset was developed in direct collaboration with expert Simmental breeders and judges, ensuring that the ground‐truth labels accurately reflected established breed standards and quality criteria. The study explores the use of machine learning techniques to automate cattle quality classification from image‐based data. Only the Good and Bad classes were used, and the Average class was excluded to maintain clearer visual separation between categories. Overall, this study represents an initial attempt at developing a scalable and data‐driven method for evaluating cattle quality, illustrating methodological mastery rather than claiming a definitive solution. The first models trained made use of VGG16, ResNet50, and MobileNetV2. Among these, MobileNetV2 consistently achieved the best results. Using 5‐fold cross‐validation, an average test accuracy of 67.8% was achieved. When predictions were aggregated through an ensemble approach, the average test accuracy improved to 79.33%. While this represents a promising outcome, it should be viewed as an initial demonstration of how ensemble methods can enhance predictive stability within this context. The study then explored whether a model trained on cow images could be effectively employed for bull classification via transfer learning, as compared to training a bull‐specific model directly, also based on MobileNetV2. Experiments using five‐fold cross‐validation yielded inconclusive evidence on the effectiveness of transfer learning for this task, primarily because the bull dataset was too small to provide sufficient statistical power. In addition, applying majority voting across multiple images of the same individual, both cows and bulls, further improved classification reliability, yielding accuracy gains of more than 10 percentage points for both cohorts. This underscores the effectiveness of ensemble‐based evaluation in accounting for variation due to animal posture and camera angle. To demonstrate practical application, a pro‐ totype web‐based system was developed with a Flutter frontend and a Python backend, enabling real‐time classification of uploaded cow images for experiments. AFRIKAANSE OPSOMMING: Die klassiikasie van beeste se kwaliteit, veral met betrekking tot liggaamsbou, is ’n belangrike faktor in veebestuur, veral vir ’n ras soos die Simmentaler, waar waarneembare eienskappe beide die ekonomiese waarde van beeste sowel as teelbesluite beïnvloed. Die datastel wat in hierdie studie gebruik is, is in noue samewerking met ervare Simmentaler‐telers en beoordelaars saamgestel om te verseker dat die etikettering die amptelike rasstandaarde en kwaliteitkriteria weerspieël. ’n Belangrike bydrae van hierdie studie lê in die verbetering van voorspellingsakkuraatheid deur die gebruik van ’n nuutontwikkelde hoëgehalte‐beeldedatastel van Simmentalerbeeste, waaraan 5‐voudige kruisvalidering en ensemble‐leertegnieke toegepas is. Hierdie studie verteenwoordig ’n eerste poging om ’n skaalbare, data‐gedrewe metode vir die evaluering van beeskwaliteit te ontwikkel. Met behulp van hierdie datastel ondersoek die studie die moontlikheid om masjienleer toe te pas om die gehalte van Simmentalerbeeste outomaties vanaf foto’s te bepaal. Slegs die Goeie en Swak klasse is gebruik, en die Gemiddeld‐klas is uitgesluit om sterker visuele onderskeid tussen kategorieë te handhaaf tydens hierdie eerste benadering. Daar is met verskeie modelle geëksperimenteer, insluitend VGG16, ResNet50 en MobileNetV2. Deur middel van vyfvoudige kruisvalidering is ’n gemiddelde toetsakkuraatheid van 67.8% behaal. Wanneer meer as een foto per koei gebruik is deur middel van ’n ensemblebenadering, het die gemiddelde toetsakkuraatheid tot 79.33% toegeneem. Hierdie verbetering dui voorlopig op die potensiaal van ensemble‐metodes om voorspellings te stabiliseer. Ná die resultate van die koei‐model verkry is, het die studie verder ondersoek of ’n model wat op koeie opgelei is, effektief na bulle oorgedra kan word, en dit vergelyk met ’n model wat slegs op bulle opgelei is. Eksperimente wat gebruik gemaak het van vyfvoudige validasie het minder oortuigende resultate gelewer, hoofsaaklik omdat die datastel baie min unieke bulle bevat het. Verdere eksperimente het getoon dat wanneer meer as een foto per individuele dier gebruik word en ’n meerderheidstem‐benadering toegepas word, die akkuraatheid van beide koei en bulmodelle met meer as 10% verbeter het. Hierdie bevinding bevestig die doeltreffendheid van ’n meerderheidmetode wat die dier se postuur en die kamera‐posisie in ag neemen die resultate oor verskeie foto’s saamvat. Vir verdere eksperimentele doeleindes en om die praktiese toepaslikheid van die navorsing te demonstreer, is ’n prototipe‐stelsel ontwikkel wat regstreekse klassiikasie van koeifoto’s kan uitvoer. 2026-01-09T07:16:34Z 2026-01-09T07:16:34Z 2025-12 Thesis https://scholar.sun.ac.za/handle/10019.1/134814 Stellenbosch University xvii, 139 pages : illustrations application/pdf Stellenbosch : Stellenbosch University |
| spellingShingle | Simmental cattle -- Data processing Cattle breeds -- Classification Machine learning Graphics processing units Schulenburg, Suzette Initial attempt: classification of Simmental cattle based on body conformation using machine learning |
| title | Initial attempt: classification of Simmental cattle based on body conformation using machine learning |
| title_full | Initial attempt: classification of Simmental cattle based on body conformation using machine learning |
| title_fullStr | Initial attempt: classification of Simmental cattle based on body conformation using machine learning |
| title_full_unstemmed | Initial attempt: classification of Simmental cattle based on body conformation using machine learning |
| title_short | Initial attempt: classification of Simmental cattle based on body conformation using machine learning |
| title_sort | initial attempt classification of simmental cattle based on body conformation using machine learning |
| topic | Simmental cattle -- Data processing Cattle breeds -- Classification Machine learning Graphics processing units |
| url | https://scholar.sun.ac.za/handle/10019.1/134814 |
| work_keys_str_mv | AT schulenburgsuzette initialattemptclassificationofsimmentalcattlebasedonbodyconformationusingmachinelearning |