Mechanical Assembly Parts Recognition Technology Based on Yolov5 Target Detection Algorithm

Aiming at the problem of automatic identification of mechanical assembly parts in industrial production lines, this paper proposes a common mechanical assembly part nut, screw and washer recognition technology based on Yolov5 target detection algorithm. In the data preparation stage, the training set and verification set of three categories of nuts, bolts and gaskets are formed by collecting pictures of nuts, bolts and gaskets on the Internet and take pictures of different angles. In order to improve the adaptability of the data set to the Yolov5 target detection algorithm, the photos of the data set are subjected to random exposure, Gaussian noise, and image rotation. The Labelimg tool in Python is used to label the data set to form a TXT file containing the location information of the label prediction box. Based on Python (3.8) and Pycharm compiler, the training and verification network of Ylolv5 target detection algorithm is constructed. The TXT file is input to extract and learn the features of nuts, bolts and gaskets. Through 300 rounds of iteration, a more accurate recognition and classification model is formed. The training results show that the training location loss, training loss of confidence loss and training classification loss gradually converge with the increase of iteration rounds, which are stable at about 5%, 2 % and 1.8 % respectively. The recall rate and the average accuracy of the mean value at the confidence level of 50 % are about 88 % and 86 % respectively.