江西科学2025，Vol.43Issue(5)：857-862,6.DOI:10.13990/j.issn1001-3679.2025.05.006
改进瓶颈与Lite-CSA块的城市植被语义分割提取研究
Enhanced Semantic Segmentation and Urban Vegetation Extraction Using Bottleneck and Lite-CSA Blocks
摘要
Abstract
To address common challenges in vegetation extraction such as semantic informa-tion loss,poor computing performance,and suboptimal target attention capability,this study proposes an improved GSCBottleneck module based on ResNetV1c(50).The goal is to enhance computational efficiency while stabilizing semantic information extraction.The effectiveness of the module was validated through experiments conducted on both the Vaihin-gen data set provided by ISPRS and Zhanggong central area of Ganzhou city.Additionally,a lightweight structure of Lite-CSA block was embedded at the low-level feature extraction and transmission stage to enhance feature attention and improve model's ability to recognize vegetation within complex backgrounds and small-scale targets.Experimental results show that models that bottleneck with GSCBottleneck and Lite-CSA blocks achieved significant improvements across multiple metrics,including vegetation extraction accuracy(Acc),in-tersection over ratio(IoU)and Fscore.This method not only improves the precision of vege-tation extraction,but also optimizes the computational efficiency,which provides strong support for the application of deep learning in urban vegetation monitoring.In addition,the experiment also verified the effectiveness and generalization ability of the method in the Dee-pLabV3+model,showing good scalability and practical value.关键词
深度学习/语义分割/植被提取/注意力机制Key words
deep learning/semantic segmentation/vegetation extraction/attention mecha-nism分类
信息技术与安全科学引用本文复制引用
胡继匀,兰小机..改进瓶颈与Lite-CSA块的城市植被语义分割提取研究[J].江西科学,2025,43(5):857-862,6.基金项目
国家自然科学基金项目(41561085). （41561085）
