Application of LVQ neural network combined with the genetic algorithm in acoustic seafloor classification

The Learning Vector Quantization (LVQ) neural network approach has been widely used in acoustic seafloor classification. However, one of the major weak points of LVQ is its sensitivity to the initialization, affecting the seafloor classification accuracy. In this paper, Genetic Algorithm (GA) is used to optimize the initial values of LVQ. The GA-based LVQ can rapidly provide the most optimized initial reference vectors and accurately identify many types of seafloor, such as rock, gravel, sand, fine sand and mud in survey areas. The proposed new approach has been applied to sea-floor classification using Multibeam Echo Sounder (MBES) backscatter data in Jiaozhou Bay near Qingdao City of China. Comparing the evolving LVQ with the standard LVQ, the experiment results indicate that the approach of GA-based LVQ has improved the seafloor classification speed and accuracy.