Analysis Methodology of Image Classifiers in Stressed Environments

Abstract

This report defines the methodology used to analyze 2D (image) classifiers in the form of convolutional neural network models in the presence of battlefield stresses. Battlefield stresses are defined in this context as those that alter the sensor data relative to the images used to train the neural networks. Battlefield stresses can be intentional, such as camouflage or distortion, or unintentional, such as obscurants or excessive background (or foreground) clutter. Parts needed to perform the analysis are described first, followed by the methodology steps used to rank and rate the neural network classifiers, and the report concludes with a practical analysis case.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Sep 01, 2022
Accession Number
AD1180069

Entities

People

  • Patrick S. Debroux

Tags

Communities of Interest

  • Materials and Manufacturing Processes
  • Weapons Technologies

DTIC Thesaurus Topics

  • Army
  • Artificial Intelligence
  • Artificial Intelligence Software
  • Classification
  • Computer Languages
  • Convolutional Neural Networks
  • Data Sets
  • Deep Learning
  • Directories
  • Distortion
  • Graphics Processing Unit
  • Image Classification
  • Images
  • Machine Learning
  • Military Vehicles
  • Networks
  • Neural Networks
  • Three Dimensional
  • Training
  • Two Dimensional
  • Video
  • Video Clips
  • Video Frames
  • Video Images
  • Visible Spectra
  • Visualizations

Readers

  • Computer Vision.
  • Neural Network Machine Learning.
  • Sensor Fusion and Tracking Systems.

Technology Areas

  • AI & ML
  • AI & ML - Neural Networks