Neural implementation of fuzzy K-NN classification for seismic pattern recognition

Fuzzy K-nearest neighbor classification rule is implemented by neural network of the Hamming net and is applied to seismic first arrival picking. Two kinds of feature set are used in two experiments. The first experiment uses amplitude, mean power level, power ratio, and envelope slope as the features. The second experiment uses average amplitude and envelope at the local maximum amplitude as the features. In the training stage, the training wavelets of the first arrival are selected from the training traces, and features are generated. The fuzzy membership is then assigned to each training wavelet. In the testing stage, features of each seismic trace is generated, and each candidate local picks is selected. Then the features of each candidate local pick, through the neural network, are used to determine if the candidate is the first arrival or not. The experimental results are quite encouraging