Research On Detection And Recognition Technology Of Auto Brake Pads Spray Code Based On Deep Neural Network

With the development of science and technology,the intelligent upgradation has become a necessity for manufacturing industry in China.How to apply the advanced technology theory to the actual production line has become the focus and hot spot.In the production process of auto brake pads,the traditional manual spray code detection has many weaknesses,such as intensive labor intensity,high cost of personnel,low detection accuracy and etc.Advanced technologies like machine vision and machine learning,however,are able to address these issues successfully.Therefore,this paper studies the spray code detection and recognition of auto brake pads emphasizing on machine vision and machine learning.The main work is as follows:(1)This paper studies the method of character location based on connected component.This paper firstly concludes that the length-width ratio of the code characters is fixed and the code characters are usually distributed along a straight line by analyzing the characteristics of the spray code characters on brake pads,and then proposes an improved MSER algorithm which could locate spray code characters by screening the maximally stable extremal regions in terms of geometric and positional features in character region.(2)In this paper,the method of spray code characters localization based on deep learning is studied.After analyzing the weaknesses and strengthens of EAST algorithm for text detection in natural scenes,a character localization algorithm for auto brake pads based on the improved EAST algorithm is proposed.It is improved in four parts including network structure,label generation,loss function construction and candidate area postprocessing,so as to improve its detection ability for long text area and complete the localization of the spray code characters.(3)The method of character recognition based on character segmentation is studied.By analyzing the image features of spray code characters,and aiming at the characteristics of internal discontinuity and close distance between characters,an image binarization algorithm based on edge elimination and an improved projection algorithm are proposed to complete the character segmentation of a line of spray code characters,resulting in single spray code characters.The convolution neural network is constructed to recognize the single characters.(4)This paper studies the end-to-end spray code character recognition method based on deep learning.Aiming at the problem that the recognition result depends on the segmentation quality in the segmentation based character recognition method,an end-toend spray code character recognition method is proposed.By using convolutional neural network and recurrent neural network to extract and learn the characteristics of character sequence,a line of characters can be recognized without segmentation.(5)The proposed algorithms are verified in an automobile brake pad manufacturer.By collecting the spray code images of the brake pads,the corresponding character localization data set and character recognition data set are made.In the data set experiments,the detection and recognition effects of different algorithms are compared,and the advantages and disadvantages of each algorithm are analyzed.Through experiments,the feasibility and reliability of the proposed algorithm in the application of spray code detection and recognition of automobile brake pads are verified.