Identification of meat types using wireless digital microscope 1000X WIFI and deep learning

Identification of meat types is necessary to overcome the issue of meat adulteration in the market which can harm consumers because each type of meat has a different price level and there are also several types of meat that certain consumers should not consume due to religious factors. Some types of meat have a similar external appearance, so it is difficult to distinguish using human vision. The purpose of this study was to characterize and identify four types of meat i.e. beef, goat meat, buffalo meat, and wild boar meat using a wireless digital microscope 1000X WIFI as a computer vision sensor and a pre-trained convolutional neural network (CNN) as a classifier. In the image characterization process, 286 image features were extracted from color and textural features. Based on the average value and standard deviation data, some image features can classify several types of meat, but none of the image features can classify the four types of meat precisely because there was an overlap in the standard deviation range. Therefore, CNN was used to classify the four types of meat accurately. This study used four pre-trained CNNs i.e., AlexNet, ResNet-50, GoogLeNet, and Inception-V3 with 30 epochs and 20 mini-batch sizes. The data used was 500 data for training and validation in each type of meat. Sensitivity analysis was performed by varying the learning rate (0.0001 and 0.00005) and the type of optimizer (Adam, RMSProp, and SGDM). The results of the pre-trained CNNs showed the best validation accuracy of 100%.