基于U-Net多尺度和多维度特征融合的皮肤病变分割方法

吉林大学学报(理学版) ›› 2021, Vol. 59 ›› Issue (1): 123-127.

基于U-Net多尺度和多维度特征融合的皮肤病变分割方法

王雪

吉林农业科技学院网络信息中心, 吉林吉林 132101

收稿日期:2020-06-22 出版日期:2021-01-26 发布日期:2021-01-26
通讯作者: 王雪 E-mail:wx@jlnku.edu.cn

Skin Lesion Segmentation Method Based on U-Net with Multi-scale and Multi-dimensional Feature Fusion

WANG Xue

Center of Network Information, Jilin Agricultural Science and Technology University, Jilin 132101, Jilin Province, China

Received:2020-06-22 Online:2021-01-26 Published:2021-01-26

摘要/Abstract

摘要： 针对皮肤病变区域尺度不同和形状不规则, 传统U-Net网络方法缺乏从不同尺度分析目标的鲁棒性, 并在提取图像高层语义特征时丢失部分空间上下文信息而影响后续分割精度等问题, 提出一种基于U-Net多尺度和多维度特征融合的医学图像分割方法. 首先, 用空洞卷积融合不同尺度的空间上下文信息; 其次, 用通道上下文信息融合模块提取特征图各通道间的权重信息; 最后, 将特征图中的多尺度和多维度信息进行融合, 以保留更多的空间上下文信息. 实验结果表明, 该方法在皮肤病变数据集上对皮肤病变区域进行分割的分割效果较好.

关键词: 医学图像分割, U-Net网络, 多尺度和多维度特征融合, 皮肤病变分割

Abstract: In view of the skin lesions with different scales and irregular shapes, the traditional U-Net method lacked robustness to analyze targets from different scales, and lost some spatial context information when extracting high-level semantic features of the image, which affected the accuracy of subsequent segmentation, the author proposed a medical image segmentation method based on U-Net with multi-scale and multi-dimensional feature fusion. Firstly, the spatial context information from different scales was fused by atrous convolution. Secondly, the weight information of each channel of the feature map was extracted by the channel context information fusion module. Finally, the multi-scale and multi-dimensional information in the feature map was fused to preserve more spatial context information. Experimental results show that the proposed method can segment the skin lesion on the skin lesion dataset, and the segmentation effect is good.

Key words: medical image segmentation, U-Net network, multi-scale and multi-dimensional feature fusion, skin lesion segmentation

中图分类号:

TP391

王雪. 基于U-Net多尺度和多维度特征融合的皮肤病变分割方法[J]. 吉林大学学报(理学版), 2021, 59(1): 123-127.

WANG Xue. Skin Lesion Segmentation Method Based on U-Net with Multi-scale and Multi-dimensional Feature Fusion[J]. Journal of Jilin University Science Edition, 2021, 59(1): 123-127.