Advanced search
Start date

Technical feasibility analysis of a solution for real time processing based on the application of artificial intelligence (IA) together with images collected by drones to perform a quantitative inventory of herds

Grant number: 18/08354-0
Support type:Research Grants - Innovative Research in Small Business - PIPE
Duration: February 01, 2019 - October 31, 2019
Field of knowledge:Physical Sciences and Mathematics - Computer Science
Principal Investigator:Lucas Momm Bastos
Grantee:Lucas Momm Bastos
Company:Horus Aeronaves Ltda. - EPP (Filial)
CNAE: Serviços de engenharia
Atividades técnicas relacionadas à arquitetura e engenharia
City: Piracicaba


This research aims at the technical feasibility analysis of a solution for real time processing based on application of Artificial Intelligence (IA) together with images collected by drones to perform a quantitative inventory of herds. The use of aerial images obtained through Unmanned Aerial Vehicles (VANTs, in this project called by drones) have proved to be an important tool in agribusiness due to its great efficiency in agricultural planning and monitoring. In livestock, in particular, the use of drones images can aid in the process of performing the herd inventory. The massive amount of data that a drone collects during a flyover brings some hindrances to the processing and extraction of information that can assist the farmer in decision making. In this way the objective of this research is the development of a solution for real time processing based on the application of Artificial Intelligence in the images collected by drones for the accomplishment of quantitative inventory of herd. The solution consists of a drone, that will capture and storage of images, an embedded system (GPU), that will perform the processing of the obtained images and the algorithms based on computer vision and artificial intelligence (AI) techniques. This project is a continuation of the research carried out in 2017 by the researcher Electronic Engineer Giovanni Cimolin da Silva in the application of I.A in the images obtained by drones to determine the population of individuals. In his research, Giovanni investigated the use of different machine learning techniques and computational vision for an application in the detection and counting of individuals in a eucalyptus plantation. The most accurate convolutional neural network (CNN) models were used through the tensorFlow platform. In his research Giovanni identified that CNN of the single shot detector type (SSD) presented satisfactory results reaching 93% of assertiveness in the detection and counting of pine trees being 7 times faster than other networks tested, with low demand of hardware what it would be possible to embark the processing and detection on the aircraft itself. It is intended to evaluate the performance of these same neural networks now applied to the accomplishment of quantitative inventory of herd.It is known that in this research there is a complex amount of variables and steps necessary to obtain results with real validity. In this way the research will be divided into four stages: 1 - Review of the bibliography and study of the state of the art 2 - Selection of the neural network 3 - Creation of the dataset and Training of the network and 4 - Evaluation of the results. The first stage, which consists of deepening the topic and studying the state of the art, aims to investigate the advances that may have occurred in the application of neural networks to solve complex problems, thus ensuring a better selection of the type of network for this research. The third step consists of creating a dataset with a minimum number of images to guarantee convergence of the detection models and the training of the networks using a computational cluster. In the final stage, a technical report will be carried out based on the evaluation of the results obtained, which will indicate whether or not the proposal for a prototype can be submitted in the FAPESP Small Business Innovation Program (PIPE 2). With the results of this research, we intend to create a new, more assertive form of quantitative inventory of herds. Cattle raising is an economic activity of great importance for the national economy and for its administration reliable information is necessary for an efficient management. Today the methods for herd inventory are manual with high error rate and quite time consuming. This new tool contributes to increase the competitiveness of cattle raising through the increase of control over its assets, thus contributing to the economic development of beef cattle production in Brazil. (AU)