Detecting meteors in images with deep learning

• Grant number: 18/20508-2
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Effective date (Start): January 01, 2019
• Effective date (End): June 30, 2020
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Ana Carolina Lorena
• Grantee: Yuri Oliveira Galindo
• Host Institution: Instituto de Ciência e Tecnologia (ICT). Universidade Federal de São Paulo (UNIFESP). Campus São José dos Campos. São José dos Campos , SP, Brazil

Abstract

Various space agencies are interested in monitoring the passage of meteors in Earth's atmosphere, such as NASA, that developed the CAMS project (Cameras for Allsky Meteor Surveillance). In Brazil, the EXOSS (Exploring the Southern Sky) organization uses citizen science to monitor the night sky, relying on low budget stations that are mounted by professional and amateur astronomers across the country. This work proposes to use deep neural networks in identifying the presence of meteors in images captured by the EXOSS system. A program for classifying in real time images as meteors or not meteors will be developed, expanding on a previous work and applying it in a practical setting. Taking advantage of the large quantity of data captured by EXOSS, different approaches to the problem will be studied and compared, such as training the networks solely on EXOSS data and using transfer learning. Deep neural networks are currently the best performing algorithms in the fields of image classification, text translation and voice recognition, representing a relevant research field. The work accomplished in this Scientific Initiation project will contribute to the country's research in meteor detection and deep neural networks, and will also directly aid the monitoring of meteors in Brazil.