基于FPGA的快速带钢表面缺陷检测系统设计

Design of fast steel surface defective detection system based on FPGA

为使带钢表面缺陷检测系统满足实时检测的要求,提出以VGG-19作为主干网络,完成现场可编程门阵列(field programmable gate array,FPGA)对带钢表面进行快速检测与识别系统的设计。本系统基于ZYNQ-7000平台进行软硬协同设计:1)在PL端通过加速数据流方法对卷积进行并行化的设计,实现硬件加速,并且在PL端用高级语言对整个网络进行调度,控制与使用PL端加速的各个IP核。2)通过定点数据量化的方式,在数据精度几乎损失很少的情况下,大大减少了FPGA片上资源的使用,从而实现算法的加速。最终实验结果表明,本算法与CPU相比,速度提升了6倍,CPU平台与FPGA平台功耗比为12.6,GPU平台与FPGA平台功耗比为38.2,更适合嵌入式平台上的应用。

In order to make the detection system of steel surface defect meet the requirements of real-time detection,the VGG-19 was proposed of the main network to complete the design for the rapid detection and identification system of the steel surface by field programmable gate array(FPGA).This system conducts software and hard collaborative design based on the ZYNQ-7000 platform:1)The hardware acceleration was achieved through the design of convolution parallelization by accelerating the data flow design on the PL side.High-level language was applied to schedule the whole network,control and use the each accelerated IP core in the PL side.2)By means of fixed-point data quantization,the use of FPGA chip resources was greatly reduced with little loss of data accuracy,so as to realize the acceleration of the algorithm.The final experimental results show that compared with the CPU,the speed of this algorithm has increased by 6 times.The power consumption ratio between the CPU platform and the FPGA platform is 12.6,and the power consumption ratio between the GPU platform and the FPGA platform is 38.2,which is more suitable for applications on the embedded platform.