Wavelet transform implemented classification of sensor data

Abstract

An improved classifier estimates a classification of a sensed object. Data representing the sensed object are received by a sensor, such as a radar, and transformed into wavelet transform coefficients. A subset of the wavelet transform coefficients are selected, the number of coefficients in the subset being fewer in number than the size of the original data. The subset of wavelet transform coefficients is then used in place of the original data by a classifier that generates the classification of the sensed object. The classifier may be a correlation (profile matching) classifier or a quadratic classifier. In another embodiment of the invention, a wavelet fusion classifier further improves classifier performance by taking data from a single sensor, and transforming it into a reduced subset of wavelet transform coefficients for use with a correlation classifier, and also transforming the data into a reduced subset of wavelet transform coefficients for use with a quadratic classifier. These respective reduced subsets of wavelet transform coefficients are supplied to corresponding correlation and quadratic classifiers. A confidence level is determined for the output of each of the correlation and quadratic classifiers. The output of the classifier having the highest confidence level is then selected as the estimated classification of the sensed object.