Terrain characterization for site selection and preparation

Lobato, Ana Michaela

Permalink

https://hdl.handle.net/2142/113028 Copy

Description

Title

Terrain characterization for site selection and preparation

Author(s)

Lobato, Ana Michaela

Issue Date

2021-07-14

Director of Research (if dissertation) or Advisor (if thesis)

Norris, William R

Department of Study

Industrial&Enterprise Sys Eng

Discipline

Systems & Entrepreneurial Engr

Degree Granting Institution

University of Illinois at Urbana-Champaign

Degree Name

M.S.

Degree Level

Thesis

Keyword(s)

• terrain characterization
• remote sensing
• image processing
• machine learning
• deep learning

Abstract

Terrain characterization is a key component in autonomous base camp site selection and preparation. Aerial terrain characterization allows for large areas of interest to be characterized in a safe and efficient manner. In this work three terrain characteristics, terrain elevation/slope, land cover/land use classes, and soil moisture content were determined using UAV-mounted sensors to inform base camp site selection and preparation decisions. To determine accurate and real-time elevation/slope values, a stale a priori digital elevation model (DEM) was merged with a high-resolution, updated LIDAR DEM using the mblend method. The mblend method achieved better results than the traditional cover method by ensuring fewer height discontinuities along the edge of the two DEMs. To perform land cover/land use mapping, three semantic segmentation models (PSPNet, U-Net, and Segnet) and three base models (VGG, ResNet, and MobileNet) were modified to include multispectral imagery and compared. Seven land cover classes were determined with an accuracy of 82.71% by model ResNet/SegNet. To determine soil moisture content (SMC), ten models were developed to predict soil moisture – two machine learning models, support vector machine (SVM) and extremely randomized trees (ET), were paired with 5 input variables. The results indicated that SMC could be predicted with greater accuracy by reducing the dimensionality of a hyperspectral dataset to resemble a standard multispectral dataset. The ET model produced better estimations of SMC when trained with the reduced dimensionality (RD) input set and concatenated multispectral (CM) set – obtaining an increase of 1.3% (RD) and 5.4% (CM) in R-squared values and a decrease of .13 and .22 in mean absolute error (MAE) when compared to the baseline set. Finally, a process overview and use case is presented to illustrate the terrain characterization process as a whole.

Graduation Semester

2021-08

Type of Resource

Thesis

Permalink

http://hdl.handle.net/2142/113028

Copyright and License Information

Copyright 2021 Ana Lobato

Owning Collections