Meat Quality Assessment based on Deep Learning

Abstract

Under unsuitable sales conditions, red meat containing rich amount of protein might receive a negative perception from consumers. Importantly, nutrients lose their effectiveness, while at the same time the formation of harmful microorganisms becomes a threat to human health. The main purpose of this study is to keep the quality of the open department sales service offered to the consumers in the retail red meat sector at the highest level, to ensure the sustainability of resources and to provide immediate economic precautions by reducing the disposal of the red meat due to possible deterioration. To do so, one tray of meat cubes has been monitored for a long time with a stable camera mounted on a pilot red meat counter and RGB images have been acquired in every two minutes. In parallel, expert data has been gathered and used as reference labels. After a preprocessing mechanism on the acquired images, a deep convolutional neural networks architecture has been modeled and trained to classify images as fresh or spoiled. The obtained experimental results and comparisons prove that deep learning methods will be very successful in this research field. However, the most important challenge in this subject is to collect large volumes of training datasets of various types of meat and meat products which are individually labeled by food experts.