基于通道层剪枝的YOLOv4-MobileNetv3 FPGA加速策略 翻译

Based on Channel-Level Pruning, an FPGA Acceleration Strategy for YOLOv4-MobileNetv3

摘要：本文提出一种基于通道层剪枝的FPGA加速策略，该策略应用于目标检测算法YOLOv4-MobileNetv3。通过对MobileNetv3中的卷积层进行剪枝，减少了算法中的参数数量和计算量，从而提高了算法的运行速度和效率。同时，利用FPGA的并行计算能力，对剪枝后的模型进行加速处理，大幅度缩短了算法的运行时间。实验结果表明，该加速策略能够在保持模型精度的同时，实现了较高的加速比。

关键词：通道层剪枝；FPGA加速；目标检测；YOLOv4-MobileNetv3

Abstract: This paper proposes an FPGA acceleration strategy based on channel-level pruning, which is applied to the object detection algorithm YOLOv4-MobileNetv3. By pruning the convolutional layers in MobileNetv3, the number of parameters and computing power in the algorithm are reduced, thus improving the running speed and efficiency of the algorithm. At the same time, the parallel computing power of the FPGA is used to accelerate the pruned model, greatly reducing the running time of the algorithm. Experimental results show that this acceleration strategy can achieve a high acceleration ratio while maintaining the model accuracy.

Keywords: Channel-level pruning; FPGA acceleration; object detection; YOLOv4-MobileNetv