Research On Recognition Metihod Of Digital In Strument Based On Extreme Learning Machine

With the rapid development of modern social economy and science and technology,computer vision is becoming more and more widely used in face recognition,industrial product appearance measurement,product sorting,industrial instrument recognition and other fields,in order to replace the traditional human eye vision recognition.It significantly shortens the time of human visual recognition,and improves the efficiency of industrial production and people's daily life.In actual industrial production,various instruments,such as barometer,electric meter,noise measuring instrument,balance instrument,all have a common feature: these instruments are usually located in the extreme environment that can not be long lived,or in remote areas where the cost of using artificial reading watches is high.If the manual reading table is used,it will not only work efficiently,but also cost a lot.At the same time,it will also cause uncontrollable errors.Therefore,it is of great practical significance to replace human eyes with computer vision.In view of the above situation,this thesis mainly completes the application and Realization of computer vision in instrument recognition.In this thesis,a method of digital instrument recognition based on extreme learning machine is proposed.Firstly,the thesis preprocessing the image of digital instrument by digital image processing technology(using weighted mean method to grayscale processing,using angle correction method for Image tilt correction,and using 3D block matching algorithm." The image denoising method of BM3D(Block-Matching 3D)algorithm and histogram equalization are used to enhance the image,and then the digital instrument dial is extracted according to the statistical features,geometric features and morphological features of the digital meter dial,and the position,length and height of the character string of the digital instrument are expressed in accordance with the position,the length and the height of the digital instrument character string.The interest area is located;then the string of the region of interest is segmented,and the single character is separated from the string.Then themethod of Softmax regression model based on the three layer feedforward neural network and the limit learning machine are used as the training and training classification algorithm for character recognition,thus the instrument character is identified.The reading number.The breakthrough and innovation in this thesis can be summed up as follows three points:1.The method of image denoising based on BM3 D is applied to this thesis.Through experiments,it is proved that the image processed by this method can greatly reduce the interference of noise to the image,thus improving the accuracy of the final recognition.2.A method of locating the region of interest in image characters is designed: the vertex of a corner of the first character of the string and the vertex of a corner of the instrument digital display edge forms the position relation between a point and the origin in a rectangular coordinate system,and then accurately locate the position of the character by the ratio of the length to the width of the single character and the width ratio of the single character.Its size and size.3.The method of Softmax regression model based on three layer feedforward neural network and the method of limit learning machine ELM(Extreme Learning Machine)are used as recognition methods respectively,and the advantages of ELM algorithm in image processing and recognition are proved by experiments.The ELM method has higher recognition than the Softmax regression model method.Rate(Softmax regression model method: 92.69%;ELM method: 99.07%),faster recognition speed(Softmax regression model method: 0.546s;ELM method:0.0468s).The recognition algorithm is tested through experiments,and the recognition algorithm used in this thesis can accurately and effectively identify digital instruments.