Soil gradation is an important characteristic for soil mechanics. Traditionally soil gradation is performed by sieve analysis using a sample from the field. In this research, we are interested in the application of hyperspectral remote sensing to characterize soil gradation. The specific objective of this work is to explore the application of hyperspectral remote sensing to be used as an alternative to traditional soil gradation estimation. The advantage of such an approach is that it would provide the soil gradation without having to obtain a field sample. This work will examine five different soil types from the Keweenaw Research Center within a laboratory-controlled environment for testing. Our study demonstrates a correlation between hyperspectral data, the percent gravel and sand composition of the soil. Using this correlation, one can predict the percent gravel and sand within a soil and, in turn, calculate the remaining percent of fine particles. This information can be vital to help identify the soil type, soil strength, permeability/hydraulic conductivity, and other properties that are correlated to the gradation of the soil.

Soil strength characterization is essential for any problem that deals with geomechanics, including terramechanics/terrain mobility. Presently, the primary method of collecting soil strength parameters through in situ measurements but sending a team of people out to a site to collect data this has significant cost implications and accessing the location with the necessary equipment can be difficult. Remote sensing provides an alternate approach to in situ measurements. In this lab study, we compare the use of Apparent Thermal Inertia (ATI) against a GeoGauge for the direct testing of soil stiffness. ATI correlates with stiffness, so it allows one to predict the soil strength remotely using machine-learning algorithms. The best performing regression algorithm among the ones tested with different predictor variable combinations was found to be KNN with an R² of 0.824 and a RMSE of 0.141. This study demonstrates the potential for using remote sensing to acquire thermal images that characterize terrain strength for mobility utilizing different machine-learning algorithms.

Terrain traversability is critical for developing Go/No-Go maps for ground vehicles, which significantly impact a mission’s success. To predict the mobility of terrain, one must understand the soil characteristics. In-situ measurements performed in the field are the current method of collecting this information, which is time-consuming, costly, and can be lethal for military operations. This paper investigates an alternative approach using thermal, multispectral, and hyperspectral remote sensing from an unmanned aerial vehicle (UAV) platform. Remotely sensed data combined with machine learning (linear, ridge, lasso, partial least squares (PLS), support vector machines (SVM), and k nearest neighbors (KNN)) and deep learning (multi-layer perceptron (MLP) and convolutional neural network (CNN)) are used to perform a comparative study to estimate the soil properties, such as the soil moisture and terrain strength, used to generate prediction maps of these terrain characteristics. This study found that deep learning outperformed machine learning. Specifically, a multi-layer perceptron performed the best for predicting the percent moisture content (R²/RMSE = 0.97/1.55) and the soil strength (in PSI), as measured by a cone penetrometer for the averaged 0–6” (CP06) (R²/RMSE = 0.95/67) and 0–12” depth (CP12) (R²/RMSE = 0.92/94). A Polaris MRZR vehicle was used to test the application of these prediction maps for mobility purposes, and correlations were observed between the CP06 and the rear wheel slip and the CP12 and the vehicle speed. Thus, this study demonstrates the potential of a more rapid, cost-efficient, and safer approach to predict terrain properties for mobility mapping using remote sensing data with machine and deep learning algorithms.

Knowledge about the terrain properties is vital for off-road mobility planning. The Next Generation NATO Reference Mobility Model (NG-NRMM) uses many different terrain and vehicle inputs to develop Go/No-Go maps for mobility planning. When dealing with off-road scenarios, soil strength prediction is one of the key inputs for building mobility maps. Current methods utilize in-situ

Terrain traversability is critical for developing Go/No Go maps, significantly impacting a mission's success. To predict the mobility of a vehicle over a terrain, one must understand the soil characteristics. In situ measurements performed by soldiers in the field are the current method of collecting this information, which is time-consuming, are only point measurements, and can put soldiers in harm's way. Therefore, this study investigates using remote sensing as an alternative approach to characterize terrain properties. This approach will explore the relationships between electromagnetic radiation and soil types with varying properties. Optical, thermal, and hyperspectral sensors will be used to collect remote data and compare it against ground truth measurements for validation. Machine learning (linear, ridge, lasso, partial least squares, support vector machines, and k nearest neighbors) and deep learning (multi-layer perceptron and convolutional neural network) algorithms will be used to build prediction models. Results showed that soil properties such as soil gradation, moisture content, and soil strength measured by a geogauge and averaged cone penetrometer for 0-6” and 0-12” (CP06 and CP12) can be estimated remotely. Deep learning provides the best models for estimating terrain characteristics compared to machine learning. It is shown that this method can produce much finer spatial resolution coverage than traditional geospatial point-based interpolation approaches and yield a higher prediction accuracy. Predictions maps can be used to generate threshold-based Go / No Go maps using a vehicle cone index or as a cost map for vehicle performance. A Polaris MRZR vehicle was used to test the application of these prediction maps for mobility purposes, and correlations were observed between the CP06 and rear wheel slip and CP12 and vehicle speed. This study demonstrates the potential of using remote sensing data for more rapid and finer spatial resolution predictions of terrain properties with higher accuracies compared to traditional in situ mapping methods implementing machine and deep learning algorithms. The remote sensing approach allows the generation of Go/No Go and vehicle cost maps and, most importantly, provides a safe alternative to keep soldiers out of harm’s way.

Project 1.29 Characterizing the Terrain Strength and Type Using Remote Sensing for Mobility