面向胃息肉检测的深度学习神经网络优化  

Deep Learning Neural Network Optimization for Gastric Polyp Detection

作　　者：金洪杨 董晓淦 魏青彪 刘景达 岳龙旺[1] JIN Hong-yang;DONG Xiao-gan;WEI Qing-biao;LIU Jing-da;YUE Long-wang(School of Mechanical and Electrical Engineering,Henan University of Technology,Zhengzhou 450001,China;School of Energy and Intelligence Engineering,Henan University of Animal husbandry and Economy,Zhengzhou 450001,China)

机构地区：[1]河南工业大学机电工程学院,郑州450001 [2]河南牧业经济学院能源与智能工程学院,郑州450001

出　　处：《科学技术与工程》2023年第15期6506-6512,共7页Science Technology and Engineering

基　　金：国家自然科学基金(51541508);河南省自然科学基金(182300410286)。

摘　　要：胃镜检查是发现胃息肉的主要方法。传统的人工检查方式存在准确率低,易漏诊、误诊的情况。提出了一种基于深度学习的YOLOv5-SE胃息肉检测网络。该网络在目标检测算法YOLOv5的基础上进行了改进,引入注意力机制,将SE Block加入到主干网络的最后一层,增强网络的特征提取能力。改进后的YOLOv5-SE胃息肉检测网络的平均精度均值(mean average precision, mAP)达到了94.5%,相比原网络提高了3.1%,推理速度达到67 f/s(帧/秒),在满足实时性要求下较好地完成了胃息肉检测的要求。YOLOv5-SE胃息肉检测网络具有在实时性、自动检测的精度和速度等方面有一定提升,对促进胃息肉的自动检测有重要意义。Gastroscopy is the main method of finding gastric polyps.The traditional artificial inspection method has problems,such as low accuracy,easy missed diagnosis,misdiagnosis and so on.A gastric polyp detection network YOLOv5-SE based on deep learning was presented,which was improved on the basis of object detection algorithm YOLOv5 by introducing attention mechanism and SE Block to the last layer of the backbone network was added to enhance the feature extraction ability of the network.The experimental results shown that the improved YOLOv5-SE gastric polyp detection network reached mean average precision(mAP)94.5%,which had been increased by 3.1%compared to the original network,and inference speed reaches 67 f/s.The YOLOV5-SE gastric polyp detection network has good real-time performance,high automatic detection accuracy and speed,which is of great significance to promote the automatic detection of gastric polyps.

关 键 词：胃息肉检测 深度学习 神经网络 YOLOv5 SE-Block 

分 类 号：TP242.3[自动化与计算机技术—检测技术与自动化装置]