Recognition Of Numeric Characters On Ammeter Dial Plate Based On Ga-bp And Structural Features

The automatic recognition system for numeric characters which display on single-phase ammeter dial plate is a new field of the application of computer vision.The system can do the same job to people and then increase the efficiency. In this paper, the design of this system includes image preprocessing and recognition of words, non-whole words and scales.Image preprocessing includes image enhancement,binary transform and image thinning process.To improve the quality of images, we use histogram equalization method,smoothing method and mid-value filter method. During the process of binary transform,we use the gray-scale threshold based on priori knowledge to help choseing the threshold so that we can segment the image well. Projection algorithm was used to field a single character and the segmentation.This algorithm is simple and very practical.We also should thin images to extract there features and reduce calculate amounts.The GA-BP algorithm was introduced into the recognition process of meter, the use of genetic algorithm optimizing weights of BP network can overcome some defects of BP network such as the slow convergence speed and local extremun, the convergence curve of BP network has good consistency as well. Change the array of sample images into a single row as the input of the network, in this way,it can retain the main features which contained in informations of images. use network which has been trained to study prepared images, then get results.The method based on structural features to recognize non-whole words is mainly against the features of non-whole words show on single-phase ammeter dial plate.Using five-line method to extract the features of non-whole words from "0-1" to "9-0" .These features include whether the word contains horizontal line or not, whether the word with or without arc lines,and the difference of these arc lines. With these features,we can recognize different non-whole words effectively. Because of the reliationship between scales and non-whole words,we analyzed two different kinds of non-whole words and derivated results of them.After simulating,we got conclutions as follows:The effect of image preprocessing was better than before, the BP algorithm was improved effectively by using GA-BP algorithm, non-whole word knowledge other than the ideal rate and non-whole words recognition rate was high.