Convolutional neural network-related applications have now been developed on a variety of platforms, including CPUs, GPUs, and others,but most of them sacrificed energy consumption to achieve good performance. Therefore, in recent years, more and more hot spots have shifted to how to achieve related applications such as low energy consumption research methods. In order to make the steel surface defect detection system meet the requirements of real -time detection, this article proposes that VGG16 is used as the main network to complete the design of the FPGA's rapid detection and identification system for the FPGA. This system conducts software and hard-hard design based on the ZYNQ-7000 platform (1) The design of parallelization of convolution on the PL side The hardware acceleration is achieved by accelerating the data flow design. Each IP core accelerated with the PL end. (2) In the method of quantifying the data, in the case of almost little loss of data accuracy, the use of resources on FPGA films has been greatly reduced to achieve acceleration of algorithms. The final experimental results show that compared with the CPU, this algorithm has increased by 6 times. The power consumption ratio of the CPU platform and the FPGA platform is 12.6, and the power consumption ratio of the GPU platform and the FPGA platform is 38.2, which is more suitable for applications on the embedded platform.