Study on the Impact of LDA Preprocessing on Pig Face Identification with SVM

Area of Science:

• Agricultural Technology
• Machine Learning
• Animal Science

Background:

• Intelligent management of swine is crucial for efficient livestock production.
• Facial recognition offers a non-invasive method for individual pig identification.
• Traditional machine learning models require optimization for real-world agricultural applications.

Purpose of the Study:

• To evaluate the effectiveness of Linear Discriminant Analysis (LDA) preprocessing on Support Vector Machine (SVM) performance for pig facial recognition.
• To determine optimal kernel functions and coefficients for SVM classifiers in swine identification.
• To assess the impact of LDA on the efficiency (training and testing time) of pig identification systems.

Main Methods:

• Implemented and compared two SVM protocols: one standalone and one with LDA preprocessing.
• Experimentally determined optimal kernel functions (polynomial and RBF) and coefficients (0.03) for SVM.
• Conducted individual identification tests on a cohort of 10 pigs.

Main Results:

• The LDA-enhanced SVM protocol improved identification accuracy from 83.66% to 86.30%.
• Training and testing durations were significantly reduced: 0.7% and 0.3% of original times, respectively.
• Polynomial and RBF kernels with 0.03 coefficients proved effective for both protocols.

Conclusions:

• LDA preprocessing substantially enhances the accuracy and efficiency of SVM-based pig facial recognition.
• The findings provide empirical support for deploying SVM classifiers in mobile and embedded systems for swine management.
• Optimized machine learning models can improve individual animal identification in precision agriculture.