Improving the image quality of machine vision thread detection

In this paper, the disadvantage of lack of effective information in threaded workpiece imaging with industrial cameras is analyzed against the research background of external thread measurement in the vision measurement technique, aiming at solving the problems of poor anti-interference ability and low level of automation in the existing vision detection algorithms. A thread image measurement device is also designed for the experiment. In terms of algorithm, the current thread parameter measurement method based on machine vision is easily disturbed by dust, grease and other factors in the industrial measurement environment, which leads to higher image noises. In view of the above problems and based on the combination of machine vision and deep learning technology, this paper proposes a method of measuring external threads with high automation and certain anti-interference ability. Firstly, this paper adopts the U-Net model, and incorporates it with the Attention Augment mechanism and residual learning module to form AA ResU-Net model, so as to improve the ability of learning the features of the target. In addition, in this paper, defect removal and sub-pixel processing are carried out on the thread edge, which further improves the measurement accuracy and meets the needs of industrial detection.