现代电子技术2025，Vol.48Issue(11)：156-162,7.DOI:10.16652/j.issn.1004-373x.2025.11.024
基于卷积神经网络的内窥镜图像识别及FPGA实现
Endoscopic image recognition based on convolutional neural network and FPGA implementation
摘要
Abstract
In view of the large number of parameters of the existing VGG16 network model and the difficulty of FPGA acceleration,an image recognition circuit system that improves VGG is proposed,so as to improve the accuracy rate and speed of endoscopic lesion identification and reduce the power consumption of detection instruments.The system is applied to endoscopic image lesion recognition for the first time.The software technology is used to optimize the convolutional layer and fully connected layer of the VGG algorithm,and an adaptive average pooling layer is added.An optimized convolution IP core is designed to achieve convolution and maximum pooling FPGA acceleration.The integration of the network layer and the batch normalization(BN)layer will be improved in order to reduce the number of model parameters and reduce FPGA resource consumption effectively.Experimental results show that the average recognition accuracy rate of the improved network model is 95.59%,and the model size is 35.90 MB.In comparison with the accuracy rate of the original network,the accuracy rate of the improved network model is improved by 3.24%,and its number of model parameters is reduced by 92.99%.The FPGA board-level detection time is 0.55 seconds per image,which is 1 509.06 seconds per image and 0.14 seconds per image shorter than the detection time of ARM side and CPU side.After optimization and improvement,the proposed circuit system significantly improves the efficiency and accuracy rate of endoscopic lesion identification and reduces hardware resource consumption effectively.关键词
卷积神经网络/VGG模型/FPGA/内窥镜图像识别/高层次综合工具/软硬协同Key words
CNN/VGG model/FPGA/endoscopic image recognition/high-level synthesis tool/software and hardware synergy分类
电子信息工程引用本文复制引用
刘生山,林金朝,庞宇,王元发,周前能..基于卷积神经网络的内窥镜图像识别及FPGA实现[J].现代电子技术,2025,48(11):156-162,7.基金项目
重庆市自然科学基金项目(cstc2021jcyj-msxmX0590) （cstc2021jcyj-msxmX0590）
重庆教委科学技术研究计划项目(KJQN202300637) （KJQN202300637）
