Abstract:Most of current methods used for machine-vision surface roughness detection are artificially designed indices based on image information or using deep learning.However,the former is a complex computational process,and the latter takes a long time for model training and classification,which is not suitable for online inspection occasions with fast judging.To address this problem,abroad learning based milling surface roughness grade detection method was proposed.The pictures of milling workpiece surface under normal lighting environment were acquired by industrial camera.Then they were input into the constructed broad learning model for training to realize the grade detection of milling surface roughness.The method not only enables feature auto-extraction but also has fast model training,which offers a new strategy for online visual roughness measurement.