The use of deep learning to solve the inverse problem of ground penetrating radar and its application to potato farming

Lees, Charles

Abstract

This thesis reviews current analytical approaches to solving the inverse problem for Ground Penetrating Radar and then walks the reader through the creation of a 2D image creation method with the aim of demonstrating that the use of deep learning to solve this problem is viable. The next stage is to use a similar methodology in the creation of a mapping between a single radar scan and usable 3D images. The intention is to demonstrate that the technology is viable, and that by using a single scan it is possible to reduce the investment costs and the computational costs required for the hardware and system solution. A final outcome will be to demonstrate that it is possible to generate these images in the field, thus making it viable to the agricultural sector. The 2D results showed great promise and demonstrate that it is possible to utilise the 2D mapping approach to create images based on a single radar scan. The 3D results, whilst the results are not perfect, encouragingly demonstrate that there is a potential in the method contained within this thesis. Overall, the results show that there are severe limitations which still need to be overcome in the current commercial scanning hardware available and in the computational resource. Some solutions to these limitations have been suggested and should be incorporated in future developments of the system.

• Thesis Type: Thesis
• Deposit Date: Apr 17, 2023
• Publicly Available Date: Jun 7, 2024
• Public URL: https://uwe-repository.worktribe.com/output/10624245
• Award Date: Jun 7, 2024