Symbol Recognition Based On Improved Neural Network

At present, digital display instrument is widely implicated in industrial environment,because it has a lot of advantages, such as high precision, easy to read, can be set and so on.Many types of industrial equipment using seven segments display its runtime parameters orstates, such as to show the current temperature, rotating speed. With the rapid developmentof computer technology, digital instrumentation automation identification technology hasbeen widely applied. Especially in some industrial control systems, which in order toachieve unattended operation, automatic control or centralized control function, digitalinstrumentation automation recognition is essential.This study is the character recognition problem of the digital instrumentation, using BPnetwork as a means of identification. Usually, the BP network needs a large number oftraining samples, for its learning, in order to achieve certain identification andgeneralization ability. However, the acquisitions of a large number of training samples needto spend a lot of manpower and material resources. Therefore, most of the neural network istrained in the relatively small sample size situations, which is also known as the smallsample problem.For small sample problem, the research hotspot in recent years is the virtual sampletechnique. The technique according to the original sample, using reasonable methods,generates a certain number of reasonable virtual samples, to expand the original training set.This paper studies the digital instrumentation using seven segment displays, therefore,based on the seven segment code characteristic, put forward a kind of virtual samplegenerating method, and according to the method of generating a partial training samples.Proved by experiments, the virtual sample generated by this method, is used to train the BPnetwork, and can effectively provide the network identification and generalization ability.