Sonar Data Set Classification Using MLP Neural Network Trained By Modified Biogeography Based Optimization

Clutter is physically very similar to the real sonar targets in active sonar. So classification and detection of the real targets from the clutter is a complex and challenging problem for the researchers. One of the most applicable instruments to classify sonar datasets is Multi-layer Perceptron Neural Network (MLP NN). Learning is a vital part of all NNs. The use of heuristic and meta-heuristic algorithms is recently becoming very popular. This paper proposes a new migration operator in Biogeography-based Optimization (BBO) for training an MLP NN. Poor balance of exploration and exploitation is the weakness of original version of BBO. Migration, mutation and elitism are three operators in BBO. Migration operator is responsible for the information sharing among candidate solutions (islands). In this way, the migration operator plays an important role for the design of an efficient BBO. This paper proposes a new migration operator in BBO. The obtained BBO shows better diversified search process and hence finds solutions more accurately with high convergence rate. The simulation results indicate that new porposed migration model (modified BBO) has faster convergence speed and greater classification rate than other meta-heuristic algorithms.