传感技术学报2025，Vol.38Issue(10)：1784-1792,9.DOI:10.3969/j.issn.1004-1699.2025.10.008
基于超轻量实时分割网络的皮肤病变图像分割方法
Skin Lesion Image Segmentation Method Based on Ultra-Lightweight Real-Time Segmentation Network
摘要
Abstract
Skin lesion image segmentation is crucial in diagnosing skin lesions.However,deep learning-based segmentation models typi-cally have high computational costs and slow inference speeds,making them difficult to be be deployed on dermatoscopic devices with limited computing power.To address this issue,an ultra-lightweight real-time segmentation network,named ULRTS-Net,is proposed.First,depthwise separable convolutions are adopted to replace standard convolutions,and a relatively lightweight encoder-decoder network architecture is designed to reduce model complexity and computational load.Second,multi-level semantic feature fusion modules are added at skip connections to effectively bridge the semantic gap between shallow and deep features.Additionally,a multi-scale feature fusion module is proposed to enhance the model's ability to learn contextual information.Finally,spatial and channel atten-tion modules are introduced to focus on important features.Experiments show that ULRTS-Net achieves JI scores of 85.78%and 89.95%on the ISIC2016 and PH2 datasets,respectively,with only 0.407M model parameters and 1.51 GFLOPs.Compared to other methods,ULRTS-Net achieves fast and accurate segmentation with low computational costs,demonstrating its effectiveness.关键词
皮肤病变分割/超轻量实时分割/特征融合/注意力机制Key words
skin lesion segmentation/ultra-lightweight real-time segmentation/feature fusion/attention mechanism分类
信息技术与安全科学引用本文复制引用
梅厦锦,巫笠平,张文新,马玉良..基于超轻量实时分割网络的皮肤病变图像分割方法[J].传感技术学报,2025,38(10):1784-1792,9.基金项目
国家自然科学基金项目(62071161) （62071161）
浙江省教育厅科研项目(Y202351775) （Y202351775）
