Method Research On Printed Character On Circuit Board Based On Machine Learning

With the rapid development of machine learning and deep learning,a new wave of development has been driven in the field of traditional image processing.The research on the application of new machine learning methods in the field of image text recognition has become increasingly active.At the same time,with the development of smart devices,traditional industries are in urgent need of new technological reforms,in order to adapt to faster production rhythms and higher production efficiency.Image text recognition is also an indispensable part of the industrial field.Whether it is automated production lines or logistics,image text recognition technology is needed everywhere.This article takes the image text recognition technology of industrial background as the research topic,mainly studies the image text area recognition,character segmentation and character recognition in the industrial background.The specific contents are as follows:1)The method of accurate area localization of text based on PCB board is studied.In the extraction of candidate character regions,the traditional algorithm uses the maximum stable extremum region algorithm and then directly performs character segmentation.This method is in the background of more noise.The effect is not ideal.In this article,after using the MSER algorithm,the maximal value region is processed morphologically,and a self-designed rule filter is proposed.Most non-text regions are filtered according to the characteristics of the text region,and then the HOG of the training sample is extracted.Features,training samples,using SVM classifiers to achieve accurate location of candidate regions.The accuracy and feasibility of the test cases are statistically analyzed.2)The method of precise segmentation of characters is studied.In order to more accurately segment each character in the text region,a better segmentation accuracy than the traditional method is obtained,combined with the actual features of the sample,based on the vertical projection method.Combining the knowledge of knowledge,the coordinates of the starting position and the end position of the character are obtained more accurately,the divided characters are more accurate,and there is no extra blank area.The accuracy of the test cases and the reasons for segmentation failure samples are statistically analyzed.3)Based on the KNN character recognition method,in order to obtain more excellent features and reduce the number of training samples in the later stage,the extracted features are more intuitive.In this article,a combined feature extraction is proposed.Based on the contour feature,structural features and density features are added and combined to form a new combined feature extraction.Using the acquired features,the samples were trained and the KNN classifier was used for character recognition.Finally,the accuracy of the identification is counted.In summary,this article achieves the image recognition under the industrial background through the above methods and improvements,and achieves good results,confirming the feasibility and practical value of the design of this article.