Adaptive classification based on compressed data using learning vector quantization

Abstract

Classification problems using compressed data are becoming increasingly important in many applications with large amounts of sensory data and large sets of classes. These applications range from aided target recognition (ATR), to medical diagnosis, to speech recognition, to fault detection and identification in manufacturing systems. In this paper, we develop and analyze a learning vector quantization (LVQ) based algorithm for the combined compression and classification problem. We show convergence of the algorithm using techniques from stochastic approximation, namely, the ODE method.

Development of a tree structured hierarchical wavelet representation of synthetic database to NCTR

“…Number of clusters must be large to have a computational time decrease, but probability of false classification must not be degraded. Number of clusters on each decomposition level must be defined as a function of the distortion on the entire population of data vectors [1]. This distortion can be determined using a mean squared distance metric and is computed using the finest representation of the data vectors.…”

“…It is therefore important to develop efficient methods to decrease the size of high resolution data of radar targets. One way to compress these data is to use tree structured representation using clustering algorithm coupled with a multiresolution wavelet representation to decrease the data size and the number of RCS (Radar Cross Section) signature [1].…”

Development of multiresolution hierarchical trees to non-cooperative target recognition

“…Number of clusters must be large to have a decrease of computational time, but probability of false classification must not be degraded. The clusters number on each decomposition level must be defined as a function of the distortion on the entire population of data vectors [3]. This distortion can be determined using a mean squared distance metric and is computed using the finest representation of the data vectors.…”

“…One way to compress these representations is to use multiresolution signal decomposition allied with data clustering techniques, and then to merge them to build hierarchical tree structured representations to decrease the data size and the number of RCS signature [1][2][3].…”

Estimation of efficiency of a tree structured hierarchical wavelet representation of synthetic database applied to non-cooperative target recognition

“…One way to compress these data is to use tree structured representation based on clustering algorithm coupled with a multiresolution wavelet representation to reduce the data size and the number of radar cross section (RCS) signature [1]. The disadvantage is that these detailed characteristics require much computer memory to be stored, computer resources, and increase the search time to non-cooperative target recognition (NCTR) association.…”