摘要
针对皮肤病灶区域存在多样性、边缘模糊和毛发噪声等情况,提出了一种皮肤病灶分割算法(BEDU-Net)。首先,融合密集块连接和高效通道注意力模块以捕捉皮肤病变图像中的多尺度信息;其次,在编码器最后一层使用空洞空间金字塔结构来提升网络的感受野,从而更好地捕捉皮肤病灶区域的边缘信息;最后,使用带有双向循环特征增强残差模块进行跳跃连接,细化了皮肤病灶图像的边缘的同时增强了网络的抗干扰能力。此算法在ISIC2018和PH2两个数据集上进行了实验,其准确率分别为95.4%和94.8%,特异性分别为0.979和0.971,灵敏度分别为87.1%和87.9%,F_(1)评分分别为91.2%和91.4%,通过与U-Net、BCDU-Net、BUSU-Net、MCGU-Net的实验对比表明,此算法具有更好的分割效果。
A skin lesion segmentation algorithm(BEDU Net)is proposed to address the diversity,edge blur,and hair noise in skin lesion areas.Firstly,dense block connections and efficient channel attention modules are fused to capture multi-scale information in skin lesion images;Secondly,in the last layer of the encoder,the hollow space pyramid structure is used to enhance the receptive field of the network,so as to better capture the edge information of the skin lesions;Finally,a residual module with bidirectional cyclic feature enhancement is used for skip connections,which refined the edges of the skin lesion image while enhancing the network's anti-interference ability.This algorithm is tested on two datasets,ISIC2018 and PH2,with the accuracy of 95.4%and 94.8%,specificity of 0.979 and 0.971,sensitivity of 87.1%and 87.9%,F_(1) score of 91.2%and 91.4%,respectively.Experimental comparisons with U-Net,BCDU-Net,BUSU-Net,and MCGU-Net show that this algorithm has better segmentation performance.
作者
吕义付
张乾
徐艳
LÜ Yifu;ZHANG Qian;XU Yan(School of Data Science and Information Engineering,Guizhou Minzu University,Guiyang 550025,China;Key Laboratory of Pattern Recognition and Intelligent Systems of Guizhou,Guizhou Minzu University,Guiyang 550025,China;Academic Affairs Office,Guizhou Minzu University,Guiyang 550025,China)
出处
《智能计算机与应用》
2023年第8期73-79,87,共8页
Intelligent Computer and Applications
基金
贵州民族大学校级科研项目(GZMUZK[2021]YB23)。
关键词
皮肤病灶图像
高效通道注意力机制
双向循环网络
跳跃连接
深度学习
image of skin lesions
efficient channel attention mechanism
bicircular network
jump connection
deep learning
