A Convolution Neural Network for Parts Recognition Using Data Augmentation

In order to solve the problem of classifying different kinds of parts automatically, a simplified convolutional neural network was designed. Firstly, image data of the parts that were going to be classified were collected, and were divided into training dataset and testing dataset. Then, a simplified convolutional neural network that was suitable for part recognition was designed. A data augmentation method was raised to augment both train data and test data. The result of the prediction was got by voting. The accuracy was highly improved after the data augmentation was used. The method in this paper classified 29 different sizes of screws, nuts and washers successfully, and validates that data augmentation can improve the performance of the network strongly when the dataset is not big enough.