Research And Application Of Automatic Recognition Of Dial Instrument Based On Deep Learning

With the continuous development of science and technology,industrial informationization and digitalization are constantly improving.As an important tool for data measurement,data monitoring and data collection,the pointer instrument plays a huge role in industrial production and life,and has the function of maintaining normal production and life.Significance.However,there are still a large number of pointer-type instrument monitoring data that need to be manually collected and input.This data collection method not only consumes a lot of manpower and material resources,but also affects the timeliness of data and causes certain data errors.At the same time,some pointer instruments have a bad working environment and cannot be manually collected and transmitted.Therefore,how to make efficient and accurate automatic data entry for pointer instruments becomes more important.With the reduction of image acquisition cost and image acquisition quality in recent years,image recognition technology is becoming more and more mature.Therefore,this thesis adopts image processing based method to carry out automatic recognition of pointer type representation.Based on the related techniques of image processing and deep learning,this thesis proposes and designs an automatic pointer recognition algorithm based on deep learning and morphology based on the research and learning of automatic pointer recognition algorithms.The pointer-type automatic identification algorithm designed in this thesis mainly includes three core contents:detection and extraction of instrument panel and instrument digital,positioning and fitting of instrument panel pointer,identification of instrument panel number and determination of instrument representation number.In the detection and extraction of instrument panel and instrument digital,this thesis combines the knowledge of deep learning,proposes and designs a convolutional neural network model MASKR2CNN for effective information extraction of natural scene instruments,and constructs a corresponding training data set to train the model.And testing to achieve image segmentation and effective information extraction for dashboards in natural scenes;in the positioning and fitting of dashboard pointers,this thesis uses Ostu threshold segmentation method and probability for the design of pointer-type instrument effective information feature extraction method.The Hough line method fits and locates the pointer;in the last part of the core content,KNN uses the KNN to digitally identify the digital area of the meter and uses the distance method to determine the final number.According to the experimental results,the accuracy of the identification of the relative error within 5%is 81.64%.