Research And Improvement For Genetic Matching Pursuits Algorithm

Sparse decomposition based on the matching pursuit algorithm looks for the atom which has the best Inner product with local signal in the dictionary ,and chooses finite atoms to approximate signals. It has a lot of attention signal compression and feature extraction etc widely attention. But the exhausted searching of matching pursuit algorithm dues to the application scope limited, aiming at this problem there is something done as follow based on genetic matching pursuit:First, a new selection operator has been proposed which combined the proportion selection, optimal elitist, and ranking selection ,which protects optimal individuals by optimal elitist protection to suppress the randomness from proportion selection firstly, then redesign probability table, which uses linear probability initially to improve population diversity, Later chooses proportional selection to accelerate the convergence. This improved operator introduced by genetic matching pursuits is to reduce searching time.Synthetic speech signal and the actual signal simulation result indicates that genetic matching pursuit algorithm through the improved selection operators is effective in terms of residual energy and search time.Second, the algorithm is improved by the laplace crossover in which the parent of the Laplace distribution density function coefficients replace the arithmetic crossover operator coefficients, through the parent control of offspring production. Simulation results show that the improved genetic matching pursuit algorithms is effect.from the residual energy and searching time.Third, a new mutation operator named diversity mutation operator is proposed to improve genetic matching pursuits algorithm, which is moderated by the colony diversity. Mutation higher the population diversity is, lower the probability of mutation is. When the population diversity is small, big mutation rate is needed to raise the population diversity. However, When the population diversity is big, small mutation rate is needed to avoid destroy optimal individuals. Simulation results show that searching time is down by 5.54% by the improved genetic matching algorithms compared with genetic matching pursuits in conference 15.