Scholarlink

<p>As society progresses and living standards rise, there is a growing consumer demand for high-quality meat products. Traceability is becoming a subject of significant interest to all those involved in pig production and marketing. The traceability of livestock products is increasingly acknowledged as a vital component of contemporary and comprehensive food safety control systems. Implementing a quality tracing and traceability system throughout the entire production process serves as a crucial technological tool for ensuring the safety of pork. Individual pig identification is a critical aspect of modern livestock management, providing benefits in tracking, breeding, and health monitoring. Pigs are identified principally through ear tags and or tattoos which are the common “traditional” systems. This study presents a novel approach for face-based individual pig identification using supervised machine learning algorithms. The study evaluates machine learning models like Support Vector Machines (SVMs), K-nearest neighbours (KNN), Decision tree, random forest, and Naïve Bayes, along with feature extraction methods like Local Binary Patterns Histograms, Histogram of Oriented Gradients, Principal Component Analysis, and Scale-Invariant Feature Transform, to determine the most effective approach for identifying individual pigs from facial features. The best results (97% accuracy) are obtained with HoG and SVM out of all feature extraction and machine learning techniques. This research evaluates pig identification on a dataset of 1,680 images from 28 pigs, using 1,400 images for training and 280 for testing, across HOG and SIFT feature extractors with five classifiers. HOG–SVM achieved the best accuracy of 97% (272/280 correct predictions), followed by SIFT–SVM at 96% and SIFT–KNN at 95%, outperforming baseline methods such as Naïve Bayes (52–65%) and Decision Trees (60–78%).</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Face Based Individual Pig Identification Using Supervised Machine Learning Algorithm

  • 期刊论文

  • 发表日期：

    2026年5月26日

查看全文
  • Debapriya Sengupta,
  • Sanket Dan,
  • Satyendra Nath Mandal,
  • Pompi Rani Boro,
  • Faijun Toufiki,
  • Seema Rani Pegu,
  • Santanu Banik,
  • Pranab Jyoti Das

摘要

As society progresses and living standards rise, there is a growing consumer demand for high-quality meat products. Traceability is becoming a subject of significant interest to all those involved in pig production and marketing. The traceability of livestock products is increasingly acknowledged as a vital component of contemporary and comprehensive food safety control systems. Implementing a quality tracing and traceability system throughout the entire production process serves as a crucial technological tool for ensuring the safety of pork. Individual pig identification is a critical aspect of modern livestock management, providing benefits in tracking, breeding, and health monitoring. Pigs are identified principally through ear tags and or tattoos which are the common “traditional” systems. This study presents a novel approach for face-based individual pig identification using supervised machine learning algorithms. The study evaluates machine learning models like Support Vector Machines (SVMs), K-nearest neighbours (KNN), Decision tree, random forest, and Naïve Bayes, along with feature extraction methods like Local Binary Patterns Histograms, Histogram of Oriented Gradients, Principal Component Analysis, and Scale-Invariant Feature Transform, to determine the most effective approach for identifying individual pigs from facial features. The best results (97% accuracy) are obtained with HoG and SVM out of all feature extraction and machine learning techniques. This research evaluates pig identification on a dataset of 1,680 images from 28 pigs, using 1,400 images for training and 280 for testing, across HOG and SIFT feature extractors with five classifiers. HOG–SVM achieved the best accuracy of 97% (272/280 correct predictions), followed by SIFT–SVM at 96% and SIFT–KNN at 95%, outperforming baseline methods such as Naïve Bayes (52–65%) and Decision Trees (60–78%).