Detection of tree defects via ground penetrating radar

• Main Author: Dang, Thanh Nhan
• Other Authors: Abdulkadir C. Yucel
• Format: Final Year Project
• Language: English
• Published: Nanyang Technological University 2024
• Subjects: Engineering; Physics; Ground penetrating radar; Deep learning; Tree
• Online Access: https://hdl.handle.net/10356/176837
• Institution: Nanyang Technological University

This report highlights the use of radar based, deep learning driven tree health assessment system. Traditional methods relying on resistograph is deemed insufficient for the challenge, prompting recent research into on using radar as a non-invasive approach to scan and acquire information on the severity of tree defects. Available technologies can construct a permittivity map of the target from the radargram, such as migration or inversion algorithms, but at the expense of computational power. While deep learning driven approaches to permittivity mapping can reduce the computing time, a considerable dataset is needed. The framework suggested in this project involves training an encoder decoder convolutional neural network to perform regression, predicting parameterized defect geometry. The low dimensional representation of the defect helps in reduce problem’s complexity. Numerical simulations as well as real measurements suggests the framework’s high accuracy in predicting the position and extent of defect.