Nonparametric Classifier of Buried Mines Using MWIR Images

Abstract

Under Army SBIR Phase I funding, we have developed a nonparametric buried mine classifier using MWIR images. We start with our new image segmentation method based on the wavelet transform. Instead of thresholding the original MWIR images, we first apply the wavelet transform to MWIR image and estimate a threshold value in the corresponding wavelet domain. The small wavelet coefficients are associated with the noise and background clutters appeared in the original image. We then map this threshold in the wavelet domain back to MWIR image domain by applying the inverse wavelet transform. This new threshold is subsequently used to segment the MWIR images and extract small image chips (patches) containing potential buried mines for further detection and classification. In order to perform the statistical classification, we have applied Kolmogorov-Smirnov (KS) test, a powerful nonparametric statistical hypothesis test procedure. One major advantage of using KS test for buried mine detection is that we don't need to make any assumptions of the underlying statistical distributions associated with the cluster intensity variation profiles.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Nov 01, 2006
Accession Number
ADA481448

Entities

People

  • Anh H. Trang
  • Bo Ling
  • Chung Phan

Tags

Communities of Interest

  • Counter IED
  • Materials and Manufacturing Processes
  • Sensors

DTIC Thesaurus Topics

  • Algorithms
  • Computer Vision
  • Detection
  • Detectors
  • Feature Extraction
  • Image Processing
  • Image Segmentation
  • Information Science
  • Information Theory
  • Intensity
  • Land Mines
  • Machine Learning
  • Pattern Recognition
  • Signal Processing
  • Statistical Distributions
  • Warning Systems
  • Wavelet Transforms

Readers

  • Neural Network Machine Learning.
  • Quantum Dot Semiconductor Device Photonics and Graphene Optoelectronic Materials and THz Physics.
  • Statistical inference.