Trademark Recognition Method Based On BP Neural Network

Today we are in the era which the information technology develops rapidly, the image digital technology has played a important role in internet industry, and the image recognition technology is the core of the digital technologies. The pattern recognition technology used the computer to handle the images, so as to identify the different images. The Image recognition technology has been widely used in traffic? health care?military and so on. This technology not only can get the information of the picture rapidly, but also can change the world. Now the accuracy, speed and portability have become the focus of attention.Now ann-bp algorithm is the most classical neural network algorithm, it adjusts the weights and thresholds by the error which communicates in opposite direction. When the error is constant, the training is over. FPGA uses high speed CMOS process. It has the high integration, high speed and the operation is very convenient. In order to improve the accuracy and speed, we could use the FPGA in image processing.The trademark recognition is unlike traditional barcode recognition or two-dimensional code recognition, it identifies the trademark directly to get the messages. The recognition system which i designed can be widely used in sales, tour, procurement and so on, it's very convenient. The article focused on designing the BP neural network. I constantly improves the arithmetic, finally i get the optimal solution. The used the ann-bp as the system's foundation. It used EP2C35F672C6 as the main control chip and simulated in Quartus. Because of the hardware accelerator, the recognition speed is highly improved. I also improved the eigenvalue arithmetic:at first, the original image divided into blocks, then i got the and reduced the noise. I used the invariant moment to extract the eigenvalues. At last i did the tset. The result showed that the amount of calculation is smal1?the velocity is high, the recognition rate is higher than 95%.