基于改进CNN的弱边缘超声图像分割方法  

Segmentation Method for Weak Edge Ultrasound Images Based on Improved CNN

作　　者：朱彦华[1] ZHU Yanhua(Equipment Department,the First Affiliated Hospital of Guangdong Pharmaceutical University,Guangzhou 510090,China)

机构地区：[1]广东药科大学附属第一医院设备科,广州510090

出　　处：《吉林大学学报(信息科学版)》2024年第6期1018-1024,共7页Journal of Jilin University(Information Science Edition)

基　　金：广东省经济与信息化委员会、广东省财政厅共同编制的广东省工业和信息化专项资金“互联网+”应用基金资助项目(粤经信电软函[2017]74号)。

摘　　要：为解决弱边缘超声图像分割难度大的问题,提出基于改进CNN(Convolutional Neural Networks)的弱边缘超声图像分割方法。该方法首先利用平稳小波变换去除图像中的噪声,并通过加权最小二乘滤波器强化图像边缘细节,然后将改进卷积注意力模块添加到残差网络模型中提取图像特征,最后通过优化损失函数提高图像的分割精度。实验结果表明,所提方法对超声图像的弱边缘细节处理效果好,可提高对医学超声图像的分割精度。To solve the problem of difficulty in segmentation of weak edge ultrasound images,an improved CNN(Convolutional Neural Networks)based weak edge ultrasound image segmentation method is proposed.The method first uses stationary wavelet transform to remove the noise in the image,and then uses weighted least square filter to enhance the image edge details.Then,an improved convolutional attention module is added to the residual network model to extract image features.Finally,the image segmentation accuracy is improved by optimizing the loss function.The experimental results show that the proposed method has good performance in processing weak edge details of ultrasound images and can improve the segmentation accuracy of medical ultrasound images.

关 键 词：超声图像分割 图像预处理 卷积神经网络 平稳小波变换 加权最小二乘滤波器 

分 类 号：TN911.73[电子电信—通信与信息系统]