Urban flooding, image processing and deep learning

• Grant number: 24/23110-0
• Support Opportunities: Scholarships abroad - Research Internship - Scientific Initiation
• Start date: April 01, 2025
• End date: July 31, 2025
• Field of knowledge: Physical Sciences and Mathematics - Computer Science - Computing Methodologies and Techniques
• Principal Investigator: Jó Ueyama
• Grantee: Otávio Ferracioli Coletti
• Supervisor: Raja Jurdak
• Host Institution: Instituto de Ciências Matemáticas e de Computação (ICMC). Universidade de São Paulo (USP). São Carlos , SP, Brazil
• Institution abroad: Queensland University Of Technology, Australia
• Associated to the scholarship: 24/07514-4 - A flood monitoring and forecasting model supported by cameras and ultrasonic sensors, BP.IC

Abstract

In recent years, the computational cost of training and deploying artificial intelligence models has increasedsignificantly, making these models impractical for use on low-cost platforms such as mobile devices and em-bedded systems. This challenge has prompted researchers to explore ways to optimize models, reducing boththeir energy consumption and computational demands. As a result, several solutions have emerged, such asMobileNets, a convolutional neural network (CNN) with reduced computational cost, and platforms like Ten-sorFlow Lite, designed to facilitate the development of lightweight neural networks. In the context of ourresearch, which focuses on monitoring urban rivers and detecting and predicting floods, we are working withlow-cost hardware and narrowband communication networks. Given these constraints, we propose a study onthe latest advancements in tiny machine learning (TinyML), a field focused on creating optimized machinelearning models for low-power edge devices, to enhance our system with methods tailored for resource-limitedenvironments. By implementing lightweight image classification techniques, we can process data at the edgedirectly at the river site thereby significantly reducing network traffic and improving the efficiency of our floodmonitoring system.