Improved Landmine Detection by Complex-Valued Artificial Neural Networks
Abstract
This research report presents a procedure for landmine classification using an artificial neural network that can respond to complex- valued input data. This is because the acquired data in the form of scattering parameters at different frequencies are complex-valued and disregard of the phase works against the proven importance of phase in multidimensional signal processing. The complex-valued backpropagation algorithm and its variants are implemented on acquired data to classify mines of different types and shapes. The importance of noise-contaminated phase as well as the role of partial phase information in image reconstruction is also investigated. The role of wavelet superresolution for multiresolution analysis of landmines in particular and image analysis in general is also reported. Analysis of an image acquisition system composed of an array of sensors, where each sensor has a subarray of sensing elements of suitable size, is provided for increasing the spatial resolution and implement filtering of image sequences beyond the performance bound of technologies that constrain the manufacture of imaging devices. Military and commercial applications of the research are highlighted by videomosaicing and superresolution of regions of interest from a real noisy and blurred video sequence.
Document Details
- Document Type
- Technical Report
- Publication Date
- Dec 04, 2002
- Accession Number
- ADA429138
Entities
People
- N. K. Bose
Organizations
- Pennsylvania State University