University of British Columbia
Using a convolutional neural network for potential unexploded ordnance detection with time-domain electromagnetic data
Abstract
dc:descriptionUnexploded ordnances (UXOs) left on or below the ground after wars or military training are posed to be great threat to public safety. Signal processing for remediation of UXOs via is generally divided into two steps, detection and discrimination. Detection aims at detecting all potential UXO targets (including UXO and clutter (fragments of exploded ordnance)), while discrimination aims at separating UXOs out of clutters and classifying different types of UXOs. This thesis focuses on the first step, detection of potential UXO targets, and aims to replace the conventional initial targets picking procedure. The conventional potential UXO target detection method employs a threshold-based design, which identifies potential targets when response signal surpasses a specified threshold. This method is straightforward but fragile due to its limitations in information density and information fusion capability. To overcome these limitations and improve the detection accuracy, this thesis proposes using Machine Learning (ML)-based method for potential UXO target detection (referred to as targets for brevity). First, I introduce the novel design of training dataset, which takes different uncertainties of the problem into account, including different types of UXO and clutter, positional error and background noise. Second, I propose a ML-based network for target detection, which is composed of a convolution neural network to extract the texture feature of the training dataset and a linear transformation for latent feature classification. Third, I apply the proposed ML-based detection network to both synthetic and field data. The results from the proposed ML-based detection method compared with the conventional target picking method, demonstrate the effectiveness and robustness of the proposed methodology.
Degree
thesis:*- Name thesis:degree_name
- Master of Science - MSc
- Level thesis:degree_level
- master's
- Discipline thesis:degree_discipline
- Geophysics
- Grantor dc:publisher
- University of British Columbia
- Year dc:date
- 2025
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Lin, Jingrong
- Contributors dc:contributor
-
- Heagy, Lindsey
Rights
dc:rights- Statement dc:rights
-
- Attribution-NonCommercial-NoDerivatives 4.0 International
- Language dc:language
- eng
Identifiers
dc:identifier.*- Handle dc:identifier
- http://hdl.handle.net/2429/90186