Image Meat Classification Using Deep Learning Approach

Abstract

In many countries meat is one of the most popular food in the world, but not all kind of meat can be consumed by most people such as pork for muslim and jewish also beef, pork and mutton tend to have several similar characteristics, namely color and texture, this can create confusion in distinguishing the three types of meat when see it by naked eye, therefore we need a system that can help identify the type of meat. In this study the authors used CNN-based Deep Learning approached with pre-defined architecture ResNet50 and ResTet101 combined with transfer learning techniques, the authors also used several types of hyperparameters including learning rate, momentum and epoch to see the effect of these hyperparameter values on the performance of deep learning model, the results of this study show that ResNet50 can outperformed ResNet101 by producing less loss and by 1 % more in F1 score with 95.96% F1 score 95.96% precision and 95.96% recall. In addition the momentum of 0.3 in both architectures has the tendency to produce high loss with a low F1 score, while for a learning rate of 0.0001 on both ResNet50 and ResNet101 architectures also has a tendency to produce high loss and a low F1 score, optimal hyperparameter values range of learning rate is between of 0.001 to 0.01 and 0.6 - 0.9 for momentum.