Research On Missing Detection Of Bolts And Nuts In Sheet Metal Based On Computer Vision

With the continuous improvement of production technology,automotive sheet metal parts not only have been improved output increasingly,but also made their products more and more high.This has led to the detection of a problem with the sheet metal product,which has become a small probability event in a huge amount of data.The traditional manual detection method not only wastes manpower,consumes a lot of energy and money,but also makes it difficult to make a detailed and comprehensive inspection of all the large-volume sheet metal parts.In summary,the traditional manual detection method has many problems,which may lead to extremely serious security risks.With the continuous development of computer vision,it provides a new way of thinking for industrial automation detection methods.In this paper,the convolutional neural network technology is used to study the bolt and nut detection of automotive sheet metal parts.The specific research contents are as follows:First,the picture data of the sheet metal part is marked,and after the labeling,the data sample is expanded by using a small amount of data provided by the enterprise to expand the data sample.In order to expand the difference between the samples,the subject also carried out some image transformation on the basis of the sampled data for further sample expansion,and prepared data for the subsequent training model.Secondly,the problem that the image itself is large and the detection target is small may lead to large calculations and inaccurate models.This paper designs and implements a classification network that initially determines the image debris to classify the foreground and background of the sheet metal parts,and uses it to initially identify the foreground area where the bolt nut is located.This is equivalent to the subsequent target detection model.Add a layer of attention mechanism to facilitate the subsequent use of the target detection model to detect accurate foreground position coordinates.Finally,the subject uses the discriminant result of the preliminary recognition of the foreground region to input the image whose discrimination result is the foreground into the target detection model.Position the coordinates of the foreground and detect the specific location of the foreground coordinates in the image.The final conclusion is obtained by comparing the predicted coordinate position with the coordinates of the position of the bolt nut on the standard sheet metal part.