SmartUS: artificial intelligence and machine vision for precision livestock feeding

• Grant number: 19/09084-9
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Start date: April 01, 2020
• End date: April 30, 2021
• Field of knowledge: Agronomical Sciences - Animal Husbandry
• Principal Investigator: Mateus Modesto
• Grantee: Mateus Modesto
• Company: Avaltech Inovações para Agropecuária Ltda
• CNAE: Atividades de apoio à pecuária ; Desenvolvimento de programas de computador sob encomenda ; Desenvolvimento e licenciamento de programas de computador customizáveis
• Pesquisadores principais: Roberto Daniel Sainz Gonzalez
• Associated researchers: Alan Caio Rodrigues Marques ; Fabiano Rodrigues da Cunha Araújo ; Tiago Zanett Albertini ; Yuri Baldini Farjalla
• Associated scholarship(s): 20/08400-1 - SmartUS: artificial intelligence and machine vision for precision livestock feeding, BP.TT; 20/08399-3 - SmartUS: artificial intelligence and machine vision for precision Livestock feeding, BP.PIPE

Abstract

Advances in information technology, automation and real-time measurement, coupled with software to manage this information, create opportunities for individual management of beef cattle, that is, precision livestock production. Due to the heterogeneity of the animals in each lot, each one arrives at its Economically Optimum End Point (OEP) on different dates, resulting in losses in productivity and profitability. In order to overcome these losses, each individual should be marketed at their OEP, either by drafting individuals as they reach their OEP or by sorting into homogeneous lots before or at the beginning of the feedlot phase. The first option creates management issues with negative impacts on animal performance and well-being and raising the cost of production. The second requires a decision-aid tool (e.g., AvalTech's NanoBeef) that uses information about the diet, animal (i.e., sex, weight, age, breed, body condition score, conformation, frame, among others) and carcass (subcutaneous fat and loin eye area) via ultrasonography, as inputs to a dynamic model that simulates animal growth and carcass composition (and value) to estimate the OEP for each individual. The main challenge in both cases is to obtain accurate and accurate information on the size, weight and body composition of animals at entry or during fattening. The use of ultrasound limits the commercial scale of the technology because of the cost and time to process each animal. To overcome this challenge, this proposal aims to replace ultrasound with a camera system and intelligent algorithms to evaluate body composition. The first step will be the collection and organization of a database (DB) with complete phenotypic data (animals and carcass commented above) using trained evaluators, ultrasonography and intelligent camera imaging of 2,000 animals. In the second step, this DB will be used to develop algorithms to estimate the degree of carcass finishing (FAT) by imaging with intelligent cameras applying machine learning techniques based on deep convolutional networks. The animals will have the data separated into m groups for cross validation, allowing to perform m-1 rounds evaluating the quality of the prediction of the FAT and reducing the bias. This algorithm (named SmartUS) will be incorporated into systems already under development by AvalTech (NanoBeef) and partner companies such as BeefTrader (Process FAPESP 2015 / 07855-7) and Brazil Beef Quality (2016 / 15395-9). The main outputs of the project will be: (i) the first FAT training and assessment database using visual, ultrasonographic and machine learning assessments; ii) the development of the SmartUS algorithm to predict the degree of carcass finish without the use of ultrasound to predict the optimal economic end point; iii) at least one international scientific publication. (AU)