融合遗传算法和BP神经网络的光斑定位方法

Positioning algorithm for laser spot center based on BP neural network and genetic algorithm

针对振动环境中传统光斑中心定位算法存在的处理时间长、精度低等问题,本文提出一种基于遗传算法优化BP神经网络的光斑定位方法。使用BP神经网络对光斑位置进行预测,并通过遗传算法对神经网络进行优化。构建BP神经网络模型,将使用质心、形心、高斯拟合等方法求出的光斑中心位置以及形心法求出的光斑半径作为输入,对光斑真实中心位置进行预测。并使用遗传算法优化神经网络的权值和阈值,以增强预测效果。实验过程中,通过对光学系统外加干扰模拟振动环境,采集数据用于神经网络训练和算法验证。实验结果表明,优化前后的标定测试迭代次数分别为55和29,平均误差分别为0.81像素和0.45像素。由本文结果可知,在遗传算法的优化下,神经网络算法的迭代速度和预测精度均有所提高。

Aming at the problems of long processing time and low accuracy of the traditional laser spot center positioning algorithm used in a vibrating environment.We proposed a laser spot center positioning method based on a genetic algorithm optimized BP neural network.A BP neural network was applied to predict the spot center position and a genetic algorithm was applied to optimize the neural network.Based on the BP neural network,the gray weighted centroid method,centroid method,Gaussian fitting method were used to obtain the spot center position,and the centroid method was used to obtain the radius of laser spot,on the above basis,we predicted the actual center position of the spot.Genetic algorithms were used to optimize the weights and thresholds of neural networks to improve prediction accuracy.An experimental platform is es-tablished to simulate the vibration environment by applying perturbations to the optical system and the data is collected to train neural network and verify the algorithm.The experimental results show that the number of calibration test iterations before and after optimization is 55 and 29,and the average errors are 0.81 pixels and 0.45 pixels,respectively.Under the optimization of the genetic algorithm,the iteration speed and predic-tion accuracy of the neural network algorithm is improved.