Engineering Drawing Recognition Model with Convolutional Neural Network

We proposed a convolutional neural network architecture that achieves the new state of the art for classification and detection in the engineering drawings data sets. The main hallmark of this architecture is the algorithm has higher accuracy and faster rapidity for the recognition compared with the traditional algorithm. The data sets include three categories: the electrical engineering drawings, the mechanical engineering drawings and the text drawings. To meet the requirements of training pictures in experimental model, we adopted some data enhancement techniques to expand the data set, such as rotation transformation, random cutting and salt and pepper noise. By a carefully crafted design, we constructed a convolutional neural network with moderate depth while keeping the model classification accuracy of engineering drawings is more than 98%.