Robust Classification of Beef and Pork Images Using EfficientNet B0 Feature Extraction and Ensemble Learning with Visual Interpretation
Year: 2025
Type: article
Abstract: Distinguishing between beef and pork based on image appearance is a critical task in food authentication, but it remains challenging due to visual similarities in color and texture, especially under varying lighting and capture conditions. To address these challenges, we propose a robust classification framework that utilizes EfficientNet B0 as a deep feature extractor, combined with an ensemble of Regularized Linear Discriminant Analysis (RLDA), Support Vector Machine (SVM), and Random Forest (RF) classifiers using soft voting to enhance generalization performance. To improve interpretability, we incorporate Gradient-weighted Class Activation Mapping (Grad-CAM) to visualize classification decisions and validate that the model focuses on relevant regions of the meat, such as red-channel intensity and muscle structure. The proposed method was evaluated on a public dataset containing 400 images evenly split between beef and pork. It achieved a hold-out accuracy of 99.0% and a ROC-AUC of 0.995, outperforming individual learners and demonstrating strong resilience to limited data and variation in imaging conditions. By integrating efficient transfer learning, ensemble decision-making, and visual interpretability, this framework provides a powerful and transparent solution for binary meat classification. Future work will focus on fine-tuning the CNN backbone, applying GAN-based augmentation, and extending the approach to multiclass meat authentication tasks.
Authors: Ahmad Taufiq Akbar, Shoffan Saifullah, Hari Prapcoyo, Bambang Yuwono, Heru Cahya Rustamaji
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
SDG: Zero hunger
Open Access status: diamond
OpenAlex ID: W4413372465