Meat Image: evaluation of meat quality by image

• Grant number: 21/10536-1
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Start date: March 01, 2023
• End date: March 31, 2025
• Field of knowledge: Agronomical Sciences - Animal Husbandry - Animal Production
• Principal Investigator: Gregori Alberto Rovadoscki
• Grantee: Gregori Alberto Rovadoscki
• Company: Brazil Beef Quality Ltda. - ME
• CNAE: Criação de bovinos
• City: Piracicaba
• Pesquisadores principais: Marcelo Aranda da Silva Coutinho
• Associated scholarship(s): 23/04205-8 - Development of an algorithm applied to beef evaluation, BP.TT; 23/02983-3 - Mobile APP for carcass evaluation traits, BP.TT; 23/01220-6 - Meat Image: evaluation of meat quality traits by image, BP.PIPE

Abstract

The meat industry receives animals of different categories and carcass traits (marbling, age, sex, fat thickness, breeds, etc.), consequently, produces different meat cuts related to the organoleptic attributes perceived by consumers. Associated the deficiency in the evaluation of carcasses in Brazil, Brazil Beef Quality (2017) was created, offering a carcass evaluation and meat quality prediction service based on the Meat Standard Australia (MSA). Evaluating traits such as: rib eye area, subcutaneous fat thickness, marbling , meat and fat colors. Currently, these measurements are collected manually and subjectively (classic method), despite showing good results, this type of assessment is time-consuming and subject to errors, due human perception is sensitive to several factors, such as stress and fatigue, recurring situations in a slaughterhouse environment. Thus, the objective of this project is to develop an algorithm based on Machine Learning for image evaluation using a mobile device for carcass classification for meat quality in real time. Images of the sirloin steaks of 2,000 animals previously evaluated for rib eye area, subcutaneous fat thickness, marble, meat and fat colors will be used, which will be obtained from the slaughterhouses in São Paulo state. For the construction of this algorithm via Machine Learning, the YOLO method, well known and with good performance for object identification, will be used, which provides fast and real-time object detection. The algorithm will be adapted into a system via API, always prioritizing the demands of slaughterhouses. (AU)