چكيده به لاتين
Sonar target classification is considered the wide range of targets and methods for classification. The detected echoes are classified into the target and non-target echoes. The classification of targets and false alarm, which are very similar to target’s features, and also the classification of various floating targets are the challenging points for the researcher, considering the military and commercial sensitivity in this field of study. Received echoes include noise, reverberation, clutter and various targets in the sea. In this thesis, a sonar target classifier system is designed using evolutionary neural networks. This classifier is designed in such a way that it be able to classify various floating targets in addition to target and non-target echoes in real-time and with an accuracy of 94%.
This thesis uses the neural network as a classifier to reduce the impact of environmental conditions, complex naval experiment and high statistical processing load. Because the sonar dataset’s high-dimension disrupt the real-time operation and accuracy of the designed classifier, in this thesis, newly proposed meta-heuristic algorithms are considered for neural network training algorithm. In the next step, those utilized meta-heuristic algorithms were modified to improve the classifier performance in addition to propose a new meta-heuristic algorithm entitled “Chimp Optimization Algorithm (ChOA)”. ChOA is able to classify sonar data set with an accuracy of 97.44% which is 2.24% greater than the best benchmark classifier in the almost equal condition. Then, to satisfy the real-time classification condition, the designed classifier was implemented on Field Programmable Gate Array (FPGA) using Xilinx System Generator (XSG) tool.
