Engineering Drawing Recognition Technology Based On Convolutional Neural Network

Image recognition technology more and more research is applied to the practice in all aspects of life,but the direction of the engineering drawings image recognition research is not much,traditional way of recognition is to use reverse Neural Network(Back Propagation Neural Network,the BPNN)to implement the information extraction on the drawings,but due to the types of engineering drawings,generally consists of lines with a specific element symbols,diversity,high complexity and so on,and have no fixed characteristics as to identify the target is extracted,so difficult to identify,the application effect is low.Therefore,this paper conducts research and practice on the image recognition and classification of engineering drawings,and conducts comparative experiments by building a convolutional neural network Convolutional Neural Networks(CNN)model and a reverse neural network model.The purpose of this thesis is to realize the application of computer to identify and classify engineering drawings instead of the staff.In today's engineering enterprise,the existing engineering drawings scanner is suitable for the office automation,the number of engineering drawings,still need the artificial recognition in the computer file types classified operation,needs a lot of work,it takes a long time,and the working process of the very boring,workers are easily fatigue,thus reduce the work efficiency.Therefore,it is of practical significance for computer image recognition to replace manual recognition of engineering drawings,which technically supports the informatization of enterprise office and promotes the paperless office of industrial enterprises.As the previous method is not ideal for the recognition and classification of engineering drawings,the effect can be further improved.In this paper,CNN is used to classify the electrical drawings,mechanical engineering drawings and literal images in the small-scale data set and limited computing power application scenarios.The main tasks of this paper are listed below:1.Data enhancement technology is adopted to expand the image data set through rotation transformation,random clipping,affine translation and increasing noise.The more data,the stronger the model generalization ability.2.Train the image recognition model of engineering drawings according to BPNN.Firstly,the BP neural network model is obtained by reducing and then feeding the image into the neural network in batches.Then,the BP neural network model is obtained by using iterative training and optimizing the weight and bias parameters.Finally,the accuracy of the model is tested.3.This paper proposes a lightweight CNN for training,and uses the algorithm to optimize and improve the model.On the basis of improving the image recognition degree,it reduces the parameters to be optimized,the calculation amount and the calculation time.The BPNN model trained in this paper is compared with the CNN models.According to the comprehensive performance analysis,the lightweight CNN model is more suitable for the image recognition of engineering drawings than other network models,and the accuracy in the test set reaches 98.7%.