Morphological shared-weight neural network for pattern classification and automatic target detection

A shared-weight neural network which performs a novel gray-scale morphological hit-miss transform operation for feature extraction is introduced. The network is applied to general pattern classification and automatic target detection (ATD) problems. The network is compared to the linear shared-weight network and a minimum average correlation energy (MACE) matched filter approach. A training method designed to suppress the background output for ATD problem is presented. Experimental results show that this morphological network is fast in training and is superior for gray-scale pattern classification and ATD.