Due to the problem of single factor and no benchmark in the design of image characteristic index, the measurement results of machine vision-based grinding surface roughness measurement are greatly influenced by the light source brightness, and the accuracy is limited. To address this problem, a method based on the combination of a full-reference (FR) image quality algorithm visual saliency-induced index (VSI) and a backward propagation (BP) neural network for grinding surface roughness measurement is proposed. First, the VSI is applied to characterize grinding surface roughness. Its performance is then compared with the mainstream color image-based indices. Meanwhile, the ability of anti-interference to light source is analyzed for the VSI. Furthermore, in order to improve the prediction accuracy, the BP neural network is selected and applied to construct the roughness prediction model. The experimental results show that the VSI has significant advantages in the abilities of measurement accuracy and anti-interference to light source brightness levels. BP can significantly improve the prediction accuracy for VSI.

No abstract