The Applicability of Machine Learning Methods on Infrared Video Data
Abstract
Situational awareness is a necessity for the warfighter. Using radars, cameras, and sensors is a common surveillance method for the US Army. One type of sensor is known as an Electro-Optical/Infrared (EOIR) Sensors, which use both visible and infrared sensors, allowing them to be useful in both light and dark (day/night) scenarios. These systems can be used to detect Unmanned Aircraft Systems (UAS) that are present in the sky. Recognizing these objects in the sky requires diligence from the human that is monitoring the system. This technical report was completed with the intent of investigating the feasibility of using machine learning algorithms on this type of sensor data to identify UAS in the sensor output. The sensor output was fed into a feed forward convolutional neural network which classified images as either containing a UAS or not. The convolutional model proved to be effective first attempt at working with this data. This report also provides a future direction to expand upon the work done for this report. Recommendations include fine tuning this model, as well as using other machine learning methods on this data set such as object detection and the You only look once (YOLO) algorithm. From this report, future iterations of this project can build off of this work, applying machine learning to similar data and building the Army's machine learning and artificial intelligence capabilities. The data used for this project was provided by the Precision Targeting and Integration group in the Combat Capabilities Development Command Armaments Center.
Document Details
- Document Type
- Technical Report
- Publication Date
- Sep 01, 2020
- Accession Number
- AD1124028
Entities
People
- Victoria R. Gerardi