Electronic Thesis and Dissertation Repository

Thesis Format



Master of Engineering Science


Electrical and Computer Engineering

Collaborative Specialization

Planetary Science and Exploration


McIsaac, Kenneth


Lunar regolith, unconsolidated rock on the lunar surface, is made up of various particles. Understanding the quantities and locations of these particles on the lunar surface is of particular interest to planetary scientists for mission planning and science objectives. There is a limited supply of lunar regolith samples available on Earth for planetary scientists to characterize. Lunar rover missions over the next decade are expected to provide high-resolution images of the lunar surface. Deep learning can be leveraged to analyze lunar regolith from image data. An object detection model using transfer learning was developed to identify and classify particles of in-situ lunar regolith. A custom dataset using micro-images of the lunar surface from the Apollo missions was labelled and processed. Pre-trained Faster R-CNN and Single Shot Detector with ResNet backbone architectures were tested. The results were promising for the application of deep learning using transfer learning on lunar regolith imagery.

Summary for Lay Audience

Humanity first set foot on the Moon in 1969 with the Apollo 11 mission. Few missions have explored the surface of the Moon as closely since the Apollo era. With the rise of lunar exploration through the Artemis era, missions have been planned over the next decade and are expected to provide new information about the Moon. Lunar rovers and other robotic systems equipped with specialized instruments will study the lunar surface composition, uncovering a deeper understanding of the formation of the Moon, Earth and the solar system. Missions will also analyze the surface for potential mineral resources that could be used for further exploration, and provide insight for the feasibility of a long-term human presence on the Moon. With this influx of data, there will be many opportunities to use machine learning to reduce the human resource required to process the data. Currently, planetary geologists look at lunar soil under a microscope to understand the composition and characteristics. However, as lunar robotic missions with microscopic imaging equipment begin to return data from the Moon, deep learning models can be used for this task.

Using a convolutional neural network (CNN) based model, commonly used for image classification and object detection tasks in machine learning, this work outlines a process developed to demonstrate the capability of applying deep learning to planetary geology applications. Since deep learning requires large datasets of labelled images, a transfer learning approach, which leverages models that are already trained on large datasets for one task, was used to apply to a smaller image dataset of the lunar surface. Different models were tested to understand the best architectures to use, and the models were tuned to improve performance.

This work showed that despite small amounts of data from the lunar surface, machine learning models can understand the important features of lunar soil and identify different particles, such as glass, melts, and rock and mineral conglomerates called breccias, with reasonable confidence.

Creative Commons License

Creative Commons Attribution-Noncommercial 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License