Optimization of Automatic Target Recognition with a Reject Option Using Fusion and Correlated Sensor Data

Abstract

This dissertation examines the optimization of automatic target recognition (ATR) systems when a rejection option is included. First, a comprehensive review of the literature inclusive of ATR assessment, fusion, correlated sensor data, and classifier rejection is presented. An optimization framework for the fusion of multiple sensors is then developed. This framework identifies preferred fusion rules and sensors along with rejection and receiver operating characteristic (ROC) curve thresholds without the use of explicit misclassification costs as required by a Bayes' loss function. This optimization framework is the first to integrate both "vertical" warfighter output label analysis and "horizontal" engineering confusion matrix analysis. In addition, optimization is performed for the true positive rate, which incorporates the time required by classification systems. The mathematical programming framework is used to assess different fusion methods and to characterize correlation effects both within and across sensors. A synthetic classifier fusion-testing environment is developed by controlling the correlation levels of generated multivariate Gaussian data. This synthetic environment is used to demonstrate the utility of the optimization framework and to assess the performance of fusion algorithms as correlation varies. The mathematical programming framework is then applied to collected radar data. This radar fusion experiment optimizes Boolean and neural network fusion rules across four levels of sensor correlation. Comparisons are presented for the maximum true positive rate and the percentage of feasible thresholds to assess system robustness. Empirical evidence suggests ATR performance may improve by reducing the correlation within and across polarimetric radar sensors. Sensitivity analysis shows ATR performance is affected by the number of forced looks, prior probabilities, the maximum allowable rejection level, and the acceptable error rates.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Apr 25, 2005
Accession Number
ADA437220

Entities

People

  • Trevor I. Laine

Organizations

  • Air Force Institute of Technology

Tags

Communities of Interest

  • C4I
  • Human Systems
  • Materials and Manufacturing Processes
  • Sensors
  • Weapons Technologies

DTIC Thesaurus Topics

  • Artificial Intelligence Software
  • Computational Science
  • Data Mining
  • Data Science
  • Databases
  • Detection
  • Detectors
  • Information Processing
  • Information Science
  • Intelligence Cycle
  • Machine Learning
  • Mathematical Programming
  • Neural Networks
  • Operations Research
  • Pattern Recognition
  • Surveillance
  • Target Recognition

Readers

  • Computational Modeling and Simulation
  • Neural Network Machine Learning.
  • Sensor Fusion and Tracking Systems.

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Machine Learning Algorithms