Development of an automated system for plant diseases identification based on RGB and thermal images

• Grant number: 19/00433-0
• Support Opportunities: Research Grants - Innovative Research in Small Business - PIPE
• Start date: May 01, 2020
• End date: May 31, 2021
• Field of knowledge: Agronomical Sciences - Agronomy - Plant Health
• Principal Investigator: Mateus Tonini Eitelwein
• Grantee: Mateus Tonini Eitelwein
• Company: Smart Consultoria Agronômica e Serviços Agrícolas Ltda
• CNAE: Cultivo de soja
• City: Piracicaba
• Associated researchers: Fernando Dill Hinnah ; Jayme Garcia Arnal Barbedo ; João Felipe Manfrinato Mariano
• Associated scholarship(s): 20/04791-6 - Development of an automated system for plant diseases identification based on RGB and thermal images, BP.PIPE; 20/04493-5 - Development of an automated system for plant diseases identification based on RGB and thermal images, BP.TT; 20/04498-7 - Development of an automated system for plant diseases identification based on RGB and thermal images, BP.TT; 20/04590-0 - Development of an automated system for plant diseases identification based on RGB and thermal images, BP.TT

Abstract

Plant diseases are responsible by yield reductions, social and economical problems worldwide, in multiple pathosystems. When the plant diseases control is not well performed in wide scale, it becomes an epidemic, as recently seen with tomato late blight in USA and coffee leaf rust in Central América. These examples of totally different growing conditions, technology and agricultural inputs used, evidence that as in higher as in lower agricultural practicing levels, the epidemics are recurring, and its monitoring is not well performed. Moreover, both diseases are registered since old times, therefore extensively studied, however still heavily reducing yields. Due to these problems, we believe in the constant plant diseases intensity monitoring as a fundamental compound of the diseases integrated management (DIM). The DIM rarely is applied in agriculture, due to the difficulties and specificities in its assessments. With the aim of rationalizing the plant diseases management, the present project will develop algorithms based in convolutional neural networks (CNN) is order to identify soybean diseases as early as possible, using images analyzing technologies. The soybean diseases will be used for this project development due to the large growing area, several diseases susceptibility, and the control costs. Different image optical sensors for plant diseases analysis have been researched, as RGB, spectral, thermal and fluorescent. Among them, due to the literature review, the thermal technology will be assessed in this project, aiming early detection. A commercial product composed by a camera, global positioning system, a data logger and the algorithm can solve on-field disease intensity assessments, for commercial and field trials, thus enabling rational fungicide sprays, avoiding yield losses or excessive number of sprays, besides stablishing assessment standards independent of human error. The algorithm to be developed can be boarded on specific machinery, as a terrestrial autonomous vehicle with on-field tracked movements, performing autonomous disease assessments. For that, it is necessary a trained algorithm for several diseases of the soybean crop, and different types of cameras to verify the possibility of early diseases detection (before visible symptoms). The potential market is composed by medium to large soybean growers, whom are an important part of Brazilian agriculture. This market can be amplified in the case of success in the algorithm identification for other pathosystems, using the knowledge acquired in the present project. Associated with other projects already submitted, the Smart Agri intends to build an agricultural digital platform, based on the researcher's group in precision agriculture, agrometeorology and plant diseases, and machine learning. (AU)