Embedded Software And Hardware Cooperative Image Processing Technology And Its Application In Money Counters

Image processing technology has been widely used in various fields since its birth,but with the development of society,embedded systems,such as the A-type money counter system,require higher and higher recognition accuracy,interface throughput and processing speed.Therefore,it is of great value to the research of image technology with high recognition accuracy,high throughput rate and high processing speed.In this paper,based on the LeNet-5 image classification architecture,design the FPGA-based convolution calculation module for the high accuracy of the image classification system.Then use the on-chip memory for data caching.For the high throughput requirements of the acquisition interface,first design the ADC configuration,ADC control,CIS interface control,and then design the AXI4-Lite interface,AXI4-Stream interface.For high-speed image processing requirements,first design an OpenCV image processing program,then design an image processing program on the FPGA.The innovation points of the work of the thesis is reflected in:(1)The program implementation completes the convolutional layer,the pooling layer,and the fully connected layer based on the LeNet-5 framework.Accelerate convolutional layer operations on FPGAs using SDSoC.Through the ZedBoard platform,the accuracy rate reaches 98.61%at a speed of 1.37ms per image.(2)Based on verilog language program,CIS control module,ADC configuration module,AXI4-Stream interface module and AXI4-Lite interface module are designed.Line buffer and DMA transmission are used in data transmission.Through the ZedBoard platform,the design achieve throughput of up to 376.36Mbps.(3)Based on OpenCV library and combined with OpenMP and NEON coprocessors,image contour extraction,angle correction and character segmentation are implemented on software.Acceleration of Image Processing on Hardware Using SDSoC Tools.Implemented on the ZedBoard platform,the image processing speed is up to 18 banknotes per second.Finally,based on design of this paper,ZedBoard is used as the hardware platform to build the image processing system of Class A banknote counter.The function of each module is verified by the actual test.