变照度下的视觉测量系统误差建模

针对二维视觉在线测量工件时，照度变化因素导致测量误差的问题，提出基于遗传算法优化的最小二乘支持向量机（GA-LSSVM），建立照度误差模型的方法. 分析视觉测量系统的误差来源，通过最小二乘法分析照度影响下的误差规律. 利用照度变化误差实验，获得照度和测量系统的误差数据，分别训练GA-LSSVM、支持向量机（SVM）以及BP神经网络，建立照度和测量系统误差模型，对系统测量误差进行预测. 结果表明：在变照度测量误差预测模型中，GA-LSSVM模型、SVM模型及BP神经网络模型的预测精度分别为94.90%、90.23%及80.60%. 这表明遗传算法优化的最小二乘支持向量机建立的变照度误差模型，在拟合和预测精度上优于传统的BP神经网络.

Modeling error of visual measurement system under changing illuminance

A modeling method of the illuminance error was proposed based on genetic algorithm (GA) to optimize the least squares support vector machine (LSSVM) aiming at the problem that illuminance changes led to the error of measurement for online measuring two-dimensional visual artifacts. Vision measurement system error sources were analyzed, and illuminance under the influence was analyzed by the least squares method of error patterns. The illuminance and the error of the measurement system data were obtained by using the illumination change error of experiment. The illuminance and measurement system error model was established by training the GA - LSSVM, support vector machine (SVM) and BP neural network respectively to forecast the system measurement error. Results show that the prediction accuracy of GA-LSSVM model, SVM model and BP neural network model is 94.90%, 90.23% and 80.60% respectively in the prediction model of variable illuminance measurement error. The error model of variable illuminance that is established by the least square support vector machine and optimized by genetic algorithm is superior to the traditional BP neural network in terms of fitting and prediction accuracy.