Chicken Meat Freshness Classification Based on VGG16 Architecture
Mary Bettina P. Garcia, Eugene A. Labuac, Carlos C. Hortinela
Abstract
The freshness and quality of the meat is one of the most important factors to consider. To assess the standard quality these products, a certain level of knowledge is required. This opens the possibility of using the VGG16 architecture of Convolutional Neural Network to categorize chicken meats based on freshness. The Raspberry Pi 3B+ board and a Raspberry Pi Camera Module V1.3 are the study's main hardware components. Additionally, image pre-processing aids in improving object classification accuracy by removing the darker parts of the image. Thresholding and morphological transformation of the image is applied for the image mask in which unnecessary regions are removed. To evaluate the model, 102 photos of chicken meat were used, which were separated into two categories: fresh and old. A total of 578 photos were used in the training of the network, with the remaining 102 images being used for testing. The system correctly predicted 50 fresh chicken meats out of 51 photos, with a 98.04% accuracy. Meanwhile, the system correctly classified 46 out of 51 aged chicken meats, resulting in a 90.2% accuracy. The algorithm has a 94.11% overall accuracy, with 96 out of 102 photos properly categorized as fresh or old chicken meats.