Analysis of technical feasibility of embedded computer vision system for identification of poultry carcass with possible contamination

• Grant number: 18/13213-6
• Support Opportunities: Research Grants - Research Program in eScience and Data Science - PIPE
• Start date: February 01, 2019
• End date: October 31, 2019
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computer Systems
• Principal Investigator: Fernando Antonio Torres Velloso da Silva Neto
• Grantee: Fernando Antonio Torres Velloso da Silva Neto
• Company: Mvisia Desenvolvimentos Inovadores Ltda
• CNAE: Fabricação de equipamentos e instrumentos ópticos, fotográficos e cinematográficos ; Desenvolvimento e licenciamento de programas de computador customizáveis
• City: São Paulo
• Associated researchers: Douglas Fernandes Barbin ; Elza Iouko Ida ; Sylvio Barbon Junior
• Associated scholarship(s): 19/00633-0 - Analysis of technical feasibility of embedded computer vision system for identification of poultry carcass with possible contamination, BP.eScience.PIPE

Abstract

Motivated by the successful implementation of process PIPE FAPESP Phase I (number 2012/50974-9) and P Phase II (number 2015/08706-5), the leading researcher and his team submit the current research project aiming to develop a new technology for the aviculture industry, which has a great importance for the national economy. Brazil is the largest exporter of poultry meat in the world, and the quality of this product is an important factor for the maintenance of the Brazilian consumer markets. The analysis of the quality of poultry carcasses, which represents more than 90% of the slaughtered poultry in the country, is done by visual inspection, with employees analyzing parameters such as color, shape and size of each individual carcass. Given the visual nature of the inspection performed on the qualification of poultry carcasses, a computer vision system can be employed. Still, it is generally not trivial for experts to explain exactly what parameters they use in this inspection. Thus, attribute selection techniques help in the process of choosing the most relevant characteristics. The construction of the classifier can be done with algorithms of machine learning, from examples previously determined by specialists. In addition, features that are not visible to the human eye, such as the presence of pathogens and contamination, can be identified by the use of hyperspectral images in the range of ultraviolet or near-infrared electromagnetic radiation. Considering that there are several criteria for the qualification of poultry carcasses and the limitation of time and funds at this stage of the research proposal, the present work will have the objective of classifying poultry carcasses according to the criterion of fecal, gastric contamination or biliary. After the demonstration of technical feasibility, it is intended to increase the scope of the research, with the inclusion of other parameters of carcass classification. At the end of this research project, it is expected to obtain a computer vision system capable of correctly classifying the poultry carcasses in an industrial line, allowing an increase in productivity, quality and profitability of domestic slaughterhouses. (AU)