The Research And Realization Of Handwriting Number Recognition Based On ARM9

Intelligent technology had been developping rapidly in recent years. The product became more powerful, and it showed a trend of capability integration. As the most simple and directive method of man-machine conversation, handwriting recognition technique played an important role in intelligent device. Compared with traditional CISC processor, the new RISC processor ARM, can compute faster with lower power consume, so it was widely used in embedded applications in recent years. This paper would give a research about how to implement the functionality of handwriting numbers recognition, based on ARM9 hardware platform.(1) By analyzing the development of handwriting recognition technique and the applications of ARM serial processor, this paper proved the feasibility of implementing the handwriting number recognition functionality on the platform.(2) With the core of SamSung S3C2410 processor, this paper expatiated on the structure and specialty of the system'hardware, and the theory and functions of hardware blocks.Based on the research of the Linux'structure, this paper configured and built the ARM-Linux core, installed the YAFFS file system which fit the NANDFLASH management, configured and debugged the LCD and Touch Screen, and explanted the Linux system to the ARM hardware platform.(3)With discussion on the improvement of image recognition technique, this paper expatiated the common used methods of image pre-processing in pattern recognition, Analysed the algorithms used in the blocks of image pre-processing, and then used Qt, which was a platform spanning programming tool, to implement the image pretreatment functionality.(4)By analysing the classical BP Neural Network'training method and their strong and weak points, this paper used the improved training arithmetic in view of disadvantages of traditional BP Neural Network, reduced the chance of failure in the Neural Network'training and the time cost at it. With the powerful data processing and fast programming capability of Matlab, this paper used Neural Network tool box to implement the BP Neural Network block, and then used the image data files generated from the pretreatment block to train the network. After training, this paper used character image samples to test the Neural Network, gave the test result and proved the correctness of the network's function.(5)With the benefit of Matlab's combined programming capability, this paper configured the block's params, exported the well trained BP network to a block of C code, and did some adaptive modification according to the features of Qt and Linux. According to the requirement of the issue, this paper configured the Touch Screen in the Qt environment, made a program block to convert human actions on Touch Screen into image files. In the last, the paper implant the neural network block into the main program block, and downloaded the program to the ARM-Linux platform.The theory and approach fit more complex character recognition applications, so that it could be a reference. By improving the network according to the requirement, adding more networks and reducing the scale of a single network, it could implement the more complex handwriting recognition functionality.