Research On Neural Network Recognition Based On ARM Embedded System

In the world of today, the production and management in the enterprises cannot live without informational technology. To a certain extent, the degree of informationization of an enterprise is concerned with its survival and development. However, many enterprises merely use all kinds of informational management software to carry on business, so that it could become optimization. But the method of inputting some basic data is still remain in an artificial stage, which greatly affects the facticity and efficiency of the data, thus it could further affects the efficiency of the management software. How to combine the informational management software which is already worked, making the basic data input change into automation input, it is an urgent problem for the enterprises to solve.In the modern enterprises, during the product management and information processing, the basic data is the only serial number for identifying the enterprises'production. The majority of the products'serial numbers marked begin with digit. How to realize serial number identification by using intellectual technology, and develop a kind of identified equipment that is portable and convenient, it is a value and significant research. Based on this mentality, I read a lot of domestic and foreign literatures. After that, I put forward a proposal of neural networks digital image recognition based on ARM + Linux technique frame.In this article, the recognition system is applied in the ARM intellectual protable devices. This design divides into four parts; they are hardware platforms, operating systems, image preprocessing, neural networks recognition. The hardware platforms choose ARM9 processor S3C2440 chip; the operating systems select embedded Linux system. The plan of ARM + Linux technique frame could guarantee the procedure developed more easily and changeable, while meets the need of identified devices microminiaturization. In view of identified object, the part of image preprocessing uses Image format converting, Image Graying, Image enhancing, Image Segmentation and Image normalization to settle the procedure. Based on image preprocessing, it uses BP network to realize digit image identification. Concerning BP network, it makes appropriate settings on element number of hidden layer cells, momentum, target accuracy, performance function, etc., in order to make the system achieve the best identification results. After the experimental tests, this system identified collected digital image within anticipated effect.