Reasearch On The Number Recognizing System For The Electricity Meter Based On Digital Image Processing

With the development of information society and the further advance of computer technology, as part of automatic identifying technique, digital image processing has been extensively applied to many fields. At the same time, the management means of every walk of life is changing from artificialization to automation or semi- automation.At present, electric energy is the energy most commonly used in the world therefore, in order to ensure the quality of power supply, it is urgent to employ efficient and accurate meter reading for power secter. Contraposing to the various defects of the traditional manual meter reading, an automatic meter reading technology is brought forward, and digit character recognition is a very important link, which is the content of this paper.In this paper, number recognition system of the electricity meter based on digital image processing technology is proposed. The main results of research are as follows:In image preprocessing, comparison of a wide range of threshold segmentation methods, the Ostu method which can automatically obtain the best threshold value according to the image quality is used in gray level image segmentation, the method gets good effect.In this paper, one of the most important and difficult step is number area location and segmentation for the system design. The accuracy of that directly affects the accuracy rate of single characters segmentation and number recognition. The paper uses projection method to find the rough location of the number area, then combines edge detection with mathematical morphology to accomplish the exact location of the number area and partial updating.Owing to the image acquisition which has inevitable tilt, this paper deals with the tilt correction approach which based on edge detection and Hough transform. That can effectively realize image tilt correction.In number recognition, Voxel-based method is employed to extract the features of the sample after contrasting test, and BP neural network model is proposed to accomplish the number recognition. In the pilot phase of the BP neural network, the number of the hidden layer neuron is obtained which is most acceptable to come up to the expected error and the expected output; one training algorithm of BP neural network is adopted which can obtain the best training effect and the highest recognition rate after attempting four training algorithms. Finally the recognition rate is up to 80%, the number recognition system comes off satisfactorily.The results from the present studies suggest that the number recognition of electric meter can be completed by BP neural network, at one time, the recognition method of this studies is worth the design of automatic meter reading system learning.