Based on a detailed discussion on the energy function representation of the structure extraction problem and through proper transformaton and arrangment of the energy function representation, the present paper has introduced beher order associated nonlinear continuous time neural net to the extraction problem of characteristic structure.This method is directly perceivable and understandable. It makes the characteristic structure to be extracted directly correspond to the output when the net is stable. and can conduct self-adaptive tracking and evaluation, thus opening up a new field in the extraction of charatrtidtic structure.
虞水俊, 孔铁生, 梁甸农. 特征结构的高阶神经网计算[J]. 物探与化探, 1996, 20(5): 359-364.
Yu Shuijun, Kong Tiesheng, Liang Diannong. CALCULATION OF HIGHER ORDER NEURAL NET OF CHARACTERISTIC STRUCTURE. Geophysical and Geochemical Exploration, 1996, 20(5): 359-364.
［1］ G. W. Stewart, An Updating Algorithm for Subspace Tracking, IEEE T-SP Vo1.40, P.1535-1541, 1992.
［2］ T. F. Chan, Alternative to the SVD: Rank Revealing QR-factorization, SPIE Vo1.696, 1986.
［3］ G. Mathew, V. U. Reddy and S. Deagupta, Gaum-Newton Adaptive Subspace estimation, ICASSP, IV-205, 1992.
［4］ B. Champagne, A New Adaptive Eigendecompoeition Aorithm Based on a first Order Perturbation Criterion, INRSTecomm,(Oan).
［5］ J. F. Yang&M. Kaveh, Adaptive Eigensubepace Aorithme for Direction or Frequency Estimation and Tracking,ICASSP, 1988.
［6］ J. F. Yang etc A Robust end High Resolution Adaptive Eigeneubepace Algorithm for Nonstationary Environments,ICASSP, 1990.
