Advanced search
Start date
Betweenand

Estimation of crop structure and health using heterogeneous robots

Grant number: 18/24526-5
Support type:Scholarships abroad - Research Internship - Doctorate (Direct)
Effective date (Start): March 23, 2019
Effective date (End): March 22, 2020
Field of knowledge:Physical Sciences and Mathematics - Computer Science
Principal Investigator:Roseli Aparecida Francelin Romero
Grantee:Guilherme Vicentim Nardari
Supervisor abroad: Ramakrishnan Vijay Kumar
Home Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
Local de pesquisa : University of Pennsylvania, United States  
Associated to the scholarship:17/17444-0 - Plantation monitoring using heterogeneous robots, BP.DD

Abstract

Brazil has one of the largest economies in the world, with agriculture being one of its pillars. The country is also the largest producer of oranges with 35\% of the global market. Citrus production has been suffering from many diseases, especially Huanglongbing (HLB) which spreads quickly and has no available cure. Producers rely mostly on visual inspection made by humans to detect infected trees and remove them as soon as possible which is time-consuming and prone to error. In this work, we investigate the combination of complementary aerial and ground images of a crop to estimate the structure of trees. While the aerial vehicle can cover a large area in a fast period, the ground vehicle can obtain images of higher resolution, from a different perspective. The combination of both can yield a complete vision of each tree in a crop.The estimated structure, in combination with reflectance images with near-infrared information captured by multispectral cameras for detection of HLB and further analysis such as the quantity of products/ fruits produced by region, detection of diseases in the leaves of the plants, rate of development and degree of maturity of the plants that can help the decision making of producers. Moreover, with the recent success of learning-based feature extractors, this work will explore how a traditional structure from motion pipeline can be improved with the addition of this class of methods.