In order to solve the problems about slow speed and low accuracy on the online roughness recognition of the grinding workpiece, an online measurement system for surface roughness was developed based on DSP. In this system, the surface scattered images based on the light scattering principle were captured by an industrial camera, then these images were preprocessed and their feature parameters were extracted by the DSP chip. Finally, these images with different surface roughness were classified by the multi-class support vector machine model. Experimental results show that it takes about 0.5 s for the entire identification process and the recognition rate can be up to 96% or more on this hardware platform, so this designed system can effectively identify the level of the surface roughness and realize the online testing of surface roughness.