Computer vision for identifying tomatoes in images of tomato plants captured by low-cost cameras

• Grant number: 18/20537-2
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Effective date (Start): December 01, 2018
• Effective date (End): July 31, 2019
• Field of knowledge: Interdisciplinary Subjects
• Principal Investigator: Luiz Henrique Antunes Rodrigues
• Grantee: Yuri Figueiredo Borrmann
• Host Institution: Faculdade de Engenharia Agrícola (FEAGRI). Universidade Estadual de Campinas (UNICAMP). Campinas , SP, Brazil

Abstract

The availability of microelectronics and low-cost sensors simplifies the development of systems capable of acquiring, processing and modeling various factors inside a greenhouse. One of such sensors is the camera, which can be used to identify fruits, diseases, and pests in crops. In this work, an image acquisition and processing system will be implemented with the objective of identifying images that contain mature or immature tomato fruits grown in a greenhouse using only low-cost devices and open source software. The system will be developed using deep learning techniques, which have performed well in similar works found in the literature. In particular, convolutional neural networks and transfer learning will be used and, additionally, data augmentation techniques will be evaluated. The results will be evaluated according to classification precision, sensitivity, and F1-score obtained by the models.