Non-invasive automatic beef carcass classification based on sensor network and image analysis

Abstract

The classification of beef carcasses is a task performed by a human expert, where the characteristics of a piece of meat are analyzed visually before being processed. The price and classification of the meat that comes from the inspected piece will depend on this inspection. It is a subjective task based on a visual review carried out by the operator in charge and based on his experience. Factors, such as the lighting of the room, the volume of work, and the type of pieces, can influence the decision of the operator. Currently, there are few and costly automatic systems used to classify beef carcasses. In this document, we propose the design of a computer-vision system in combination with a sensorization system for the real-time classification of beef carcasses. For the first step, Landmark detection techniques are applied for the detection of characteristic points. These points enable the segmentation of the beef carcass. In the second phase, different filters and threshold values are used on the image to segment the fat and proceed to its classification. A case study is carried out that compares the classification of 140 pieces made automatically with the classification of the same parts by a group of human experts with highly relevant results.