Development of a precision agriculture system applied to real time detection of soy plants and weeds in a controlled environment using multi spectral cameras and artificial intelligence

• Grant number: 22/06153-2
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Start date: December 01, 2022
• End date: October 31, 2023
• Field of knowledge: Agronomical Sciences - Agricultural Engineering - Agricultural Machinery and Implements
• Principal Investigator: Leonardo Bruno Renzo Milani
• Grantee: Leonardo Bruno Renzo Milani
• Company: Agrio Tecnologia Ltda
• CNAE: Desenvolvimento de programas de computador sob encomenda ; Consultoria em tecnologia da informação ; Serviços de engenharia
• City: São Carlos
• Associated researchers: Jarbas Caiado de Castro Neto ; Samuel de Paula ; Yuri Sarreta Oda
• Associated research grant(s): 23/10681-7 - Equipment for localized herbicides spraying for weed control in pre-planting and after soybean crop emergence using multispectral images and computer vision, AP.PIPE
• Associated scholarship(s): 22/15892-3 - Development of a precision agriculture system applied to real time detection of soy plants and weeds in a controlled environment using multi spectral cameras and artificial intelligence., BP.TT; 22/15921-3 - Development of a Precision Agriculture system applied to real time detection of soy plants and weeds in a controlled environment using multi spectral cameras and artificial intelligence, BP.PIPE

Abstract

Currently, precision agriculture is one of the most promising areas for the development of nacional technologies. Some technologies from this area are, for example, the mapping of productive areas and the development of sensors for climate and soil analysis systems, always aiming the intelligent use of resources during the management of the cultures and helping the producer during the decision-making steps. Among modern agriculture problems, is the intensive and non-localized use of herbicides, that besides being prejudicial to the environment, contributes to the development of herbicide-resistant weeds. To avoid this problem, this project aims to, trough methods of computer vision and artificial intelligence, develop a proof of concept that is capable of obtaining images from the top layers of plants in different wave-lengths in a controlled environment, construct multi spectral images with the individually collected images and classify the detected plants as soy plants or weeds, in a way that the whole process happens in real-time. To obtain the multi spectral images, two approaches will be used: The first using four cameras with monochromatic CMOS sensors and four individual filters covering each sensor. The second approach will utilize a single camera with a monochromatic CMOS sensor coupled with a multi spectral filter made on special specification. The images obtained during the experiments will be used for the construction of a multi spectral image bank containing the classes SOY and WEED, that will be used to train an artificial intelligence to perform detection and classification tasks. Embedded systems will be employed to capture the multi spectral images and detect weeds and soy plants on those images. With that, by the end of the project, we will have a proof of concept for the real-time detection of weeds and soy plants in a controlled environment with multi spectral images, making it possible for the development of technologies in the future, aiming for interconnected systems in pulverizing tractors of famous brands, contributing to the reduction of herbicide costs for the producer, environment damage reduction caused by general herbicide use and preventing the surge of new cases of herbicide-resisting weeds. (AU)