Abstract：Seismic first arrival pickup can be taken as a pattern recognition process to be analysed by using artificial neural network theory.One three-layer perceptron is nicely used in first arrival pickup,Error back propagation algorithm is taken in neural network training.The learning rate is automatically adjusted according to the error of output nodes,and hidden-layer nodes may be increased automatically in the learning course.As a result ,the training speeds up,and the local minimum trouble of the neural netwark can be reduced to a certain degree.Five kinds of characteristic quantity such as peak amplitude,mean square root amplitude ratio,signal/noise rado,peak-lobe difference etc.,are used in signal choice.The trained neural network results in 97% success rate of first arrival pickup for different exploration areas.
