Surface roughness measurement based on laser speckle and neural network

In order to realize non-contact and rapid measurement of surface roughness, a new surface roughness measurement method based on laser speckle and radial basis function neural network is proposed. The validity of the measurement method is verified by experiment, and several main influencing factors are analyzed. By utilizing image processing technique, four feature vectors are extracted from gathered speckle images. The four feature vectors that are nearly correlative to surface roughness include contrast, dark region ratio, gray distribution and binary feature. As neural network has characteristics such as automatically organizing, automatically studying and memory capability etc, these four above feature vectors are taken as inputs of the radial basis function neural network to realize the surface roughness measurement. A number of samples are used to train the neural network, and the trained neural network measured 4 flat-grinding specimens with different roughness values. The results indicate that the measurement method can measure surface roughness in a classifying way. The method can measure surface roughness not-contact and rapidly in high-precision. And the analysis of influencing factors is helpful to in-depth research.