Kernel Model Applied in Kernel Direct Discriminant Analysis for the Recognition of Face with Nonlinear Variations

A kernel-based discriminant analysis method called kernel direct discriminant analysis is employed, which combines the merit of direct linear discriminant analysis with that of kernel trick. In order to demonstrate its better robustness to the complex and nonlinear variations of real face images, such as illumination, facial expression, scale and pose variations, experiments are carried out on the Olivetti Research Laboratory, Yale and self-built face databases. The results indicate that in contrast to kernel principal component analysis and kernel linear discriminant analysis, the method can achieve lower （7%） error rate using only a very small set of features. Furthermore, a new corrected kernel model is proposed to improve the recognition performance. Experimental results confirm its superiority （1% in terms of recognition rate） to other polynomial kernel models.

FIXED/PREASSIGNED-TIME SYNCHRONIZATION OF QUATERNION-VALUED NEURAL NETWORKS INVOLVING DELAYS AND DISCONTINUOUS ACTIVATIONS: A DIRECT APPROACH

The fixed-time synchronization and preassigned-time synchronization are investigated for a class of quaternion-valued neural networks with time-varying delays and discontinuous activation functions. Unlike previous efforts that employed separation analysis and the real-valued control design, based on the quaternion-valued signum function and several related properties, a direct analytical method is proposed here and the quaternion-valued controllers are designed in order to discuss the fixed-time synchronization for the relevant quaternion-valued neural networks. In addition, the preassigned-time synchronization is investigated based on a quaternion-valued control design, where the synchronization time is preassigned and the control gains are finite. Compared with existing results, the direct method without separation developed in this article is beneficial in terms of simplifying theoretical analysis, and the proposed quaternion-valued control schemes are simpler and more effective than the traditional design, which adds four real-valued controllers. Finally, two numerical examples are given in order to support the theoretical results.

A direct coupling analysis method and its application to the Scientific Research and Demonstration Platform

Design of a very large floating structure(VLFS)deployed near islands and reefs,different from those in the open sea,inevitably faces new technical challenges including numerical analysis methods.In this paper,a direct coupling analysis method(DCAM)has been established based on the Boussinesq equations and the three-dimensional hydroelasisity theory with Rankine source method to analyze the responses of a VLFS in shallow sea with complicated geographical environment.Model tests have been carried out to validate the DCAM.To further verify the numerical methods and investigate the performance of such a VLFS,a“Scientific Research and Demonstration Platform(SRDP)”was built and deployed in 2019 at the site about 1000 m off an island with water depth around 40m in South China Sea.It is a simplified small model of a two-module semi-submersible-type VLFS.The numerical simulation of its responses on severe waves with focus on motions and connector forces is conduct by DCAM,and compared with the on-site measurements.Good agreement has been achieved.This approves the DCAM as a feasible tool for design and safety assessment of a VLFS deployed near islands and reefs.

ESTIMATION OF DIRECTIONAL SPECTRUM AND REFLECTED COEFFICIENT OF INCIDENT AND REFLECTED WAVE IN PHASE-LOCKED WAVE FIELD

This paper analyses and compares the property of the Modified Bayesian Directional spectrum analysis Method （MBDM） and the Modified Maximum Lkelihood Method （MMLM） that can he used to estimate directional spectrum and reflected coefficient of phase-locked wave field overlapped by multi directional irregular incident and reflected waves. The numerical test verifies the results under different wave conditions, different measurement systems, and different reflection features. The computation speed and stability of the two methods is also compared. The analysis addresses that the MBDM is better than the MMLM for directional spectrum estimating, while the MMLM is better than the MBDM for reflected coefficient estimation and calculating speed and stability.