NNU Logo

2020 Research

Computer Science


Title: Autonomous Fire Detection Using Artificial Intelligence and a Network of Fire Lookout Camera.

Thesis Adviser: Dr. Dale Hamilton

As the frequency and severity of wildfires increase, there is a growing need for improved techniques of combating these fires. This project aimed to address this issue by starting the initial work of creating a network of cameras that are able to utilize a recursive region-convolutional neural network (r-cnn) to automatically detect a forest fire by constantly scanning the area looking for smoke plums. And when a smoke plume is detected by multiple cameras, calculating the approximate location of the fire is using triangulation. This project aimed to train an rcnn to detect smoke plumes of wildfires as they are rising into the sky. As well as formulate and validate an equation that would return the latitude and longitude of a smoke plume given the latitude, longitude and heading of two cameras. The project concluded with moderate success of classifying smoke plumes and the creation of an equation to calculate the location. Future work includes gathering more training data, integrating the AI agent in a server to route and classify the camera video feed, and training/evaluating if a mask r-cnn would produce higher accuracy.


Title: Mapping Dirt Roads from Imagery Using Deep Learning.

Thesis Adviser: Dr. Dale Hamilton

Northwest Nazarene University FireMAP research team is in the process of developing a deep learning approach to finding various archaeological features. This approach is being constructed through the use of a mask region-based convolutional neural network (Mask R-CNN) using Google’s TensorFlow. Over the past few summers NNU has been gathering hyperspatial drone imagery containing these archaeological features. This aerial imagery is then fed into the Mask R-CNN in hopes of making a more dynamic approach. In the past the only way to map out these features was through a manual approach. This research project hopes to create a dynamic approach to finding and accurately mapping old roads and rail grades so that these maps can remain historically accurate.


Title: Integration of Artificial Intelligence with Mission Integration Analysis.

Thesis Adviser: Dr. Dale Hamilton

This project involved the research for how Washington River Protection Solutions, specifically Mission Integration Analysis, could use machine learning to help the planning of their radioactive waste clean up project. After researching both the field of machine learning and Mission Integrations Analysis’s needs, a plan for a machine learning tool was made. This tool would be able to make predictions for which tanks should be involved when a transfer would be made in a given situation. This tool was then developed using python and sklearn for the machine learning library along with Tkinter for the UI. Data for machine learning was queried from a database using SQL and stored in Exel sheets. After testing of the tool was performed, it showed the ability to predict an optimal tank (out of 60 possible tanks) 12%-77% of the time depending on what data was available and which classifier was used. It is theorized that once updates are made to the system which determines optimal tanks, these accuracy percentages could rise significantly.


Title: Implementation of Deep Learning to Map Dredge Tailings from Hyperspatial Aerial Imagery.

Thesis Adviser: Dr. Dale Hamilton, Dr. Barry Myers

Northwest Nazarene University’s FireMAP’s research team is developing deep learning to identify archaeological sites including roads, dredge tailings, and hand-stacked tailings in support of a collaborative relationship with the Boise National Forest. Through the implementation of TensorFlow, a software library developed by Google, a mask region-based convolutional neural network (Mask R-CNN) has been trained to identify the desired landmarks. This project focuses on using the trained Mask R- CNN and the collection and labeling of hyperspatial, aerial photos of dredge tailings extracted from a provided orthomosaic in order to provide a georeferenced shape feature. The Mask R-CNN was able to detect numerous dredge tailings from provided testing imagery with high accuracy. Obtaining additional aerial imagery of dredge tailings would likely improve the Mask R-CNN’s performance further, allowing for increased accuracy in detection.


Title: Tree Crown Classification from Hyperspatial Imagery using Machine Learning.

Thesis Adviser: Dr. Dale Hamilton

The use of imagery from small unmanned aircraft systems (sUAS) has enabled the production of more accurate data about the effects of wildland fire, enabling land managers to make better informed decisions. The ability to detect trees in imagery enables the calculation of canopy cover, improves the accuracy of fire extent calculations, and enables the calculation of tree mortality rates. We compare two machine learning algorithms for detecting trees, one that classifies each pixel independently using an SVM and one that uses a Mask R-CNN to look for trees as groups of pixels. Our object-based Mask R- CNN model is more accurate than our pixel-based SVM model.


Title: Developing a Web Application to Automate Software Testing.

Thesis Adviser: Dr. Barry Myers

Testing a software application is a necessary step in the software development process. It is especially important when the software is configuring devices that run and protect the electrical power grid. At Schweitzer Engineering Laboratories all software is thoroughly tested before being available to customers. With entire teams being dedicated to testing software any tools or processes that can enhance this process provided an excellent rate of return in terms of value. During my time at SEL, I developed an internal web application that the QuickSet testing team utilizes to automate some of their testing processes. The processes this web application automates includes powering on virtual machines, installing software onto the machines, and running scripts that start tests against the installed software. The application provides additional functionality that previously was not available in any capacity, such as running groups of tests together and being able to cancel them in batches. The finished application allows testers to spend more time writing new tests or improving old ones resulting in increased coverage and test depth.


Title: Updating the Chapel Scanning and Reporting System.

Thesis Adviser: Dr. Barry Myers

Accurate recording of student attendance of chapel is a crucial part of the chapel process on the campus of NNU. For the last few years, NNU has used a system for allowing students to scan in and out of chapel, which they only have the executable for and requires a connection to a specific network to use. The purpose was to create a replacement for the current scanning software where source code and all documentation is lost. The goals of the project were to develop a replacement program that would slot in place of the current software, add features such as images of students and encryption, and allow for updates further on through well-documented source code and documentation on every stage of the scanning process. The resulting software was a Java (GUI-based) application that is simpler to use, uses encryption methods to protect sensitive student data, and is well-documented in every aspect, from building the application to chapel scanning instructions. Future work could be done in conjunction with the NNU CX team to update the database tables for chapel credit and the application to allow for multiple chapel sessions each day.


Title: Tracking Stats of College Basketball Players During the Summer Using a Mobile Web Application.

Thesis Adviser: Dr. Barry Myers

There is a strong desire for college basketball players to get better every day. During the season they use “The Gun.” The Gun is a rebounding machine that rebounds basketball players shots and tracks their makes and their misses. However, during the off-season most athletes go home and do not have access to The Gun. Keeping track of stats manually with a pencil and paper is challenging since most athletes only bring their basketball shoes, a basketball, and their phone to the gym. This web application was created so basketball players could keep track of their makes and misses over time. This benefits the athlete and the coach because the coach can also keep track of the athlete’s performance. Athletes and coaches must sign in to have access to the table where stats are kept. The results were promising. The athletes and coaches can login and track/view stats. Future work includes displaying the data in different ways and marketing the product so it can be sold to teams.