基于人工神经网络的函数逼近法计量精度的提高

提出一种基于BP神经网络的提高激光干涉计量精度的数据处理方法，并将该方法与常用最小二乘法作二次曲线拟合进行比较。结果表明，用神经网络逼近函数方法，不但解决了光学干涉计量中像素尺寸以及图像采集卡空间量化误差对测量精度的影响，而且解决了目标函数多极值问题，从而干涉条纹可获得高于光敏像素级定位精度。最后将干涉条纹的光强精确地拟合出来，讨论曲线拟合误差，表明该算法具有很好的鲁棒性和自适应能力。

Improvement of the Metric Precision Based on Artificial Neural Network

A novel data processing method to improve the interference metric precision of laser beams based on BP neural network is presented. The method is compared with the conventional least square method to fit curves. Using the function approximation of neural network, result shows that the influence on the precision of measurement from errors in quantitating the size of image element and on the image-gathering cards is removed, and that the multi-extremum of object function is resolved. The location precision of interference stripes is also increased, and the intensity distribution of the stripes is fitted accurately. In the end we discuss the error of curve fitting. Neural network is shown to be very well in possession of robustness and adaptability.