Computer Engineering and Applications ›› 2022, Vol. 58 ›› Issue (11): 193-199.DOI: 10.3778/j.issn.1002-8331.2010-0233

• Graphics and Image Processing • Previous Articles     Next Articles

Research on Optic Disc and Macula Fovea Simultaneous Location and Detection Method on FPGA

ZHANG Wei, ZHOU Hua, LIU Yuhong, ZHANG Rongfen   

  1. College of Big Data and Information Engineering, Guizhou University, Guiyang 550025, China
  • Online:2022-06-01 Published:2022-06-01

视盘与黄斑同时定位检测的FPGA方法研究

张卫,周骅,刘宇红,张荣芬   

  1. 贵州大学 大数据与信息工程学院,贵阳 550025

Abstract: In the automatic analysis of retinal images, the location of optic disc and macula fovea is a prerequisite for computer aided diagnosis or screening of diabetic retinopathy. In this paper, a novel algorithm for locating the optic disc and macula fovea in retinal images based on YOLOv4-tiny is proposed and transplanted to field programmable gate array (FPGA) platform. Compared with the traditional method, the present work can not only locate the position of the disc and macula in retinal images simultaneously, quickly and precisely, but also be the first attempt to realize 38 layers medium-sized neural network through using high level synthesis(HLS) language and time division multiplexing technology. The COCO data set and 381 retinal images from the recognized Kaggle-Diabetic Retinopathy Detection competition are used in the training experiment. From the experimental results, the mean average precision(mAP) of optic disc and macular fovea localization after transplanting to FPGA platform is 96.11%, and it only takes 150.445?ms to detect an image on the FPGA system, which presents a good clinical application prospect in related fields.

Key words: diabetic retinopathy, optic disc, macula fovea, YOLOv4-tiny, field programmable gate array(FPGA)

摘要: 在眼底图像自动分析中,视盘与黄斑的定位是实现利用计算机辅助诊断或筛查糖尿病视网膜病变的先决条件。提出一种实现眼底图像中视盘与黄斑同时定位检测的新方法,使用YOLOv4-tiny算法定位检测,将该算法移植到现场可编程逻辑门阵列(field programmable gate array,FPGA)。与传统方法相比,该方法不仅可以快速准确地同时定位眼底图像中视盘和黄斑的位置,而且也是利用高层综合(high level synthesis,HLS)语言和时分复用技术实现38层中型神经网络的首次尝试。实验采用公认的COCO数据集和Kaggle-Diabetic Retinopathy Detection竞赛中的381幅眼底图像对算法进行训练,将训练后的算法移植到FPGA平台后视盘和黄斑定位的平均正确率(mean average precision,mAP)为96.11%,检测一张图片只需要150.445?ms,在相关领域具有良好的临床应用前景。

关键词: 糖尿病视网膜病变(DR), 黄斑, 视盘, YOLOv4-tiny, 现场可编程逻辑门阵列(FPGA)