摘要
由于传统方法对肺结节检测准确率低、速度慢,故提出一种改进YOLOv3模型检测算法,利用聚类算法对数据集进行聚类分析,并获得新的Anchor size,在原有的基础网络上进行更新改进并且调整YOLOv3算法的模型使其适应肺结节的检测任务,提高检测效率。另外,利用K-means分离前景和背景完成CT图像的预分割,对预分割结果使用腐蚀、膨胀等形态学操作提取出肺实质。改进后的YOLOv3算法在LUAN16数据集上做实验。实验结果表明,改进的YOLOv3算法对于肺结节的检测得到的mAP值达到0.932,其性能明显优于传统YOLOv3算法且具有可行性。
Due to the low accuracy and slow detection speed of the traditional methods for lung nodule,an improved YOLOv3 model detection algorithm is proposed.The data set is clustered by clustering algorithm to obtain a new anchor size.The model of YOLOv3 algorithm is updated and improved on the original basic network to adapt to the detection task of lung nodules.In addition,K-means is used to separate foreground and background to complete the pre-segmentation of CT(computed tomography)images,and morphological operations of erosion and swelling are used to extract lung parenchyma from the pre-segmentation results.The improved YOLOv3 algorithm is tested on the dataset LUAN16.The experimental results show that the mAP value of lung nodule detection by the improved YOLOv3 algorithm reaches 0.932.It can be seen that the performance of the improved algorithm is obviously better than that of the traditional YOLOv3 algorithm and it is feasible.
作者
杨友良
张建舒
陈波
YANG Youliang;ZHANG Jianshu;CHEN Bo(College of Electrical Engineering,North China University of Science and Technology,Tangshan 063210,China)
出处
《现代电子技术》
2021年第23期44-47,共4页
Modern Electronics Technique
基金
河北省自然科学基金项目(F2019209443)
河北省教育厅科技计划项目(QN2018039)。
