Share Email Print
cover

Proceedings Paper

Multispectral image fusion for detecting land mines
Author(s): Gregory A. Clark; Sailes K. Sengupta; William D. Aimonetti; Frank Roeske; John G. Donetti; David J. Fields; Robert J. Sherwood; Paul C. Schaich
Format Member Price Non-Member Price
PDF $14.40 $18.00

Paper Abstract

Our system fuses information contained in registered images from multiple sensors to reduce the effects of clutter and improve the ability to detect surface and buried land mines. The sensor suite currently consists of a camera that acquires images in six bands (400nm, 500nm, 600nm, 700nm, 800nm, and 900nm). Past research has shown that it is extremely difficult to distinguish land mines from background clutter in images obtained from a single sensor. It is hypothesized, however, that information fused from a suite of various sensors is likely to provide better detection reliability, because the suite of sensors detects a variety of physical properties that are more separable in feature space. The materials surrounding the mines can include natural materials (soil, rocks, foliage, water, etc.) and some artifacts. We use a supervised learning pattern recognition approach to detecting the metal and plastic land mines. The overall process consists of four main parts: preprocessing, feature extraction, feature selection, and classification. These parts are used in a two step process to classify a subimage. The first step, referred to as feature analysis, determines the features of sub-images which result in the greatest separability between the two classes, 'mine' and 'background'. The second step, image labeling, uses the selected features and the decisions from a pattern classifier to label the regions in the image which are likely to correspond to mines. We extract features from the images, and use feature selection algorithms to select only the most important features according to their contribution to correct detections. This allows us to save computational complexity and determine which of the spectral bands add value to the detection system. The most important features from the various sensors are fused using a supervised learning pattern classifier (the probabilistic neural network). We present results of experiments to detect land mines from real data collected from an airborne platform, and evaluate the usefulness of fusing feature information from multiple spectral bands. We show that even with preliminary data and limited testing, the performance (specified in terms of probability of detection and probability of false alarm) is very promising. The novelty of the work lies mostly in the combination of the algorithms and their application to the very important and currently unsolved operational problem of detecting minefields from an airborne standoff platform.

Paper Details

Date Published: 20 June 1995
PDF: 15 pages
Proc. SPIE 2496, Detection Technologies for Mines and Minelike Targets, (20 June 1995); doi: 10.1117/12.211378
Show Author Affiliations
Gregory A. Clark, Lawrence Livermore National Lab. (United States)
Sailes K. Sengupta, Lawrence Livermore National Lab. (United States)
William D. Aimonetti, Lawrence Livermore National Lab. (United States)
Frank Roeske, Lawrence Livermore National Lab. (United States)
John G. Donetti, Lawrence Livermore National Lab. (United States)
David J. Fields, Lawrence Livermore National Lab. (United States)
Robert J. Sherwood, Lawrence Livermore National Lab. (United States)
Paul C. Schaich, Lawrence Livermore National Lab. (United States)


Published in SPIE Proceedings Vol. 2496:
Detection Technologies for Mines and Minelike Targets
Abinash C. Dubey; Ivan Cindrich; James M. Ralston; Kelly A. Rigano, Editor(s)

© SPIE. Terms of Use
Back to Top