10 May 2021

Automated Terrain Classification for Vehicle Mobility in Off-Road Conditions

ABSTRACT:  The U.S. Army is increasingly interested in autonomous vehicle operations, including off-road autonomous ground maneuver. Unlike on-road, off-road terrain can vary drastically, especially with the effects of seasonality. As such, vehicles operating in off-road environments need to be informed about the changing terrain prior to departure or en route for successful maneuver to the mission end point. The purpose of this report is to assess machine learning algorithms used on various remotely sensed datasets to see which combinations are useful for identifying different terrain. The study collected data from several types of winter conditions by using both active and passive, satellite and vehicle-based sensor platforms and both supervised and unsupervised machine learning algorithms. To classify specific terrain types, supervised algorithms must be used in tandem with large training datasets, which are time consuming to create. However, unsupervised segmentation algorithms can be used to help label the training data. More work is required gathering training data to include a wider variety of terrain types. While classification is a good first step, more detailed information about the terrain properties will be needed for off-road autonomy.

10 May 2021

Evaluation of Automated Feature Extraction Algorithms Using High-resolution Satellite Imagery Across a Rural-urban Gradient in Two Unique Cities in Developing Countries

Abstract: Feature extraction algorithms are routinely leveraged to extract building footprints and road networks into vector format. When used in conjunction with high resolution remotely sensed imagery, machine learning enables the automation of such feature extraction workflows. However, many of the feature extraction algorithms currently available have not been thoroughly evaluated in a scientific manner within complex terrain such as the cities of developing countries. This report details the performance of three automated feature extraction (AFE) datasets: Ecopia, Tier 1, and Tier 2, at extracting building footprints and roads from high resolution satellite imagery as compared to manual digitization of the same areas. To avoid environmental bias, this assessment was done in two different regions of the world: Maracay, Venezuela and Niamey, Niger. High, medium, and low urban density sites are compared between regions. We quantify the accuracy of the data and time needed to correct the three AFE datasets against hand digitized reference data across ninety tiles in each city, selected by stratified random sampling. Within each tile, the reference data was compared against the three AFE datasets, both before and after analyst editing, using the accuracy assessment metrics of Intersection over Union and F1 Score for buildings and roads, as well as Average Path Length Similarity (APLS) to measure road network connectivity. It was found that of the three AFE tested, the Ecopia data most frequently outperformed the other AFE in accuracy and reduced the time needed for editing.

20 Aug 2020

PUBLICATION NOTICE: Understanding State-of-the-Art Material Classification through Deep Visualization

Abstract: Neural networks (NNs) excel at solving several complex, non-linear problems in the area of supervised learning. A prominent application of these networks is image classification. Numerous improvements over the last few decades have improved the capability of these image classifiers. However, neural networks are still a black-box for solving image classification and other sophisticated tasks. A number of experiments conducted look into exactly how neural networks solve these complex problems. This paper dismantles the neural network solution, incorporating convolution layers, of a specific material classifier. Several techniques are utilized to investigate the solution to this problem. These techniques look at specifically which pixels contribute to the decision made by the NN as well as a look at each neuron’s contribution to the decision. The purpose of this investigation is to understand the decision-making process of the NN and to use this knowledge to suggest improvements to the material classification algorithm.