BP神经网络的优化算法研究

BP学习算法通常具有收敛速度慢,易陷入局部极小值等缺点;遗传算法是全局优化算法,具有较强的全局搜索性能,但它在实际应用中容易产生早熟收敛的问题,且在进化后期搜索效率较低;模拟退火算法具有摆脱局部最优点的能力,能抑制遗传算法的早熟现象.因此,本文在BP算法结合遗传算法的同时,再加入模拟退火算法,可以有效地缓解遗传算法的选择压力.     

Wafer fault measurement by coherent optical processor

A microscope coherent optical processor based on the VanderLugt optical correlator is applied to the measurement of registration error in multilayer integrated-circuit wafers. A treatment of the effects of wafer faults on the correlation signal is given. Threshold criteria and fault-induced peak splitting of the correlation signal from reject production samples are exploited to demonstrate the easy and rapid detection of faults in partially processed integrated-circuit wafers.     

Properties of Cr:LiSrAIF(6) crystals for laser operation

We have performed several physical and optical measurements on the Cr:LiSAF (LiSrAlF(6)) laser material that are relevant to its laser performance, including thermal and mechanical properties, water durabilities, and Auger upconversion constants. The expansion coefficient, Young's modulus, fracture toughness, thermal conductivity, and heat capacity are all used to determine an overall thermomechanical figure of merit for the crystal. An investigation of the water durability suggests that the cooling solution should be maintained at pH = 7 to ameliorate problems associated with water dissolution. The Auger constant was found to become much more significant at higher Cr doping, in which excited-state migration leads to a substantial increase in the upconversion rate. We propose a design for a 50-W Cr:LiSAF laser system that is based on a detailed knowledge of all the relevant material parameters.     

Validation of interferometric imaging from a pupil-masking experiment on a solar telescope

We present the results of a pupil-masking experiment that uses the Sun as the source object. The goal of our experiment was a proof-of-concept validation for a Fizeau (image-plane) interferometric beam combination with a complex source that overfilled the field of view. We employed a phase-diversity technique to measure the optical phases required to recover the instantaneous optical transfer function for the masked pupil. We used a Wiener filter to deconvolve the dirty images.     

Line-shape studies for single- and triple-pass Fabry-Perot interferometer systems

To test the model developed in the preceding paper [Appl. Opt. 34, this issue (1994)] regarding the line shape produced by a Fabry-Perot interferometer system in a multipass mode, we have used three Lorentzian line shapes formed in scattering processes and subjected them to single and triple passes with some variation in the pinhole diameter. In most cases we find good agreement with the calculations with the only adjustable parameter being the single-pass contrast C(1). Where differences occur, plausible explanations are offered.     

Fiducial markers for increasing the versatility of optical correlation in the measurement of faults on integrated-circuit wafers

A microscope-coherent optical processor is used for the measurement of the registration errors on integrated-circuit wafers. The measurements are obtained from the optical correlation of wafers with reference wafer patterns by use of matched spatial filters. Previously, the intricate pattern of the active circuit area of wafers has been used in the correlation process, and a new matched spatial filter had to be created for each different integrated circuit. Here, the results of using comparatively plain fiducial markers on a wafer for the registration-error measurement are presented, and these show that the measurements can be made independent of the design of the integrated circuit while maintaining the advantages and accuracy of the optical correlation technique.     

Model-based inversion of speckle interferometer fringe patterns

Micrometer-scale rigid-body translations are determined fromelectronic speckle interferometric fringe patterns. An iterativeminimum error procedure employs the relative fringe order of pickedpositions of fringe maxima and minima within a single interferogram tocalculate the displacement field directly. The method does notcalculate the displacement at a single point but relies on theassumption that the character, but not the magnitudes or directions, ofthe displacements over the viewing area of the interferogram isknown. That is, a model of the displacements exists. Onperfect, noise-free forward modeled fringe patterns calculated for an 8.0-mum displacement, the phase error is less than 2 x10(-6) fringe orders (1.3 x 10(-5) rad)and probably results only from numerical noise in the inversion. Onreal fringe patterns obtained in electronic speckle interferometricexperiments, mean phase errors are generally less than 5 x10(-5) fringe orders (3.2 x 10(-4)rad), suggesting that the technique is robust despite errorsresulting from speckle noise, lack of accuracy in positioning ofexperimental components, and image-distortion corrections.