[1]于攀琳,吴旭*,张凌云,等.增强特征提取和解释的遥感图像语义分割模型[J].计算机技术与发展,2025,(07):8-15.[doi:10.20165/j.cnki.ISSN1673-629X.2025.0039]
　YU Pan-lin,WU Xu*,ZHANG Ling-yun,et al.A Semantic Segmentation Model for Remote Sensing Images with Enhanced Feature Extraction and Interpretation[J].,2025,(07):8-15.[doi:10.20165/j.cnki.ISSN1673-629X.2025.0039]
点击复制

增强特征提取和解释的遥感图像语义分割模型()

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:
期数:
2025年07期
页码:
8-15
栏目:
媒体计算
出版日期:
2025-07-10

文章信息/Info

Title:
A Semantic Segmentation Model for Remote Sensing Images with Enhanced Feature Extraction and Interpretation
文章编号:
1673-629X(2025)07-0008-08
作者:
于攀琳; 吴旭*; 张凌云; 刘子涵
成都理工大学 计算机与网络安全学院,四川 成都 610059
Author(s):
YU Pan-lin; WU Xu*; ZHANG Ling-yun; LIU Zi-han
School of Computer Science and Cyber Security,Chengdu University of Technology,Chengdu 610059,China
关键词:
遥感图像; 语义分割; DeepLab V3+; EfficientNet; 转置卷积; 损失函数
Keywords:
remote sensing image; semantic segmentation; DeepLab V3+; EfficientNet; transposed convolution; loss function
分类号:
TP753
DOI:
10.20165/j.cnki.ISSN1673-629X.2025.0039
摘要:
DeepLab V3+是一种具有 Encoder-Decoder 结构的语义分割模型,因其逐像素分类的特点适用于处理遥感图像的土地覆盖分类问题。 然而,下采样过程导致的特征图损失,会使连续大尺度地物的内部出现不连续无标签空洞区域,并且双线性插值算法会丢失分割边缘细节。 针对上述问题,该文提出了一种基于 DeepLab V3+改进的 V3plus-EN-TC 模型。将骨干网络替换为特征提取能力更强的 EfficientNet,引入 SE 模块和倒置残差链接,增强 Encoder 对通道信息和多尺度空间信息的感知与提取能力;融合三个层次的特征,并且采用转置卷积和双线性插值结合的上采样方法,提高 Decoder 的特征解释能力,抑制空洞区域的出现,提高边缘精度;利用 DiceFocal 联合损失函数,解决样本分布不平衡问题,并且进一步聚焦混合像元。 改进模型 V3plus-EN-TC 在预处理后的遥感数据集 GID 上,较 FCN、U-Net、SegNet、PSPNet、CBAM-DeepLab V3+、CRF-DeepLab V3+等模型,空洞区域显著减少,模型精度提升。 改进模型的平均交并比、 F1 分数、平均像素精度分别达到了 84. 74% 、88. 39% 、86. 64% 。
Abstract:
DeepLab V3+ is a semantic segmentation model with an Encoder-Decoder structure. Due to its characteristic of per-pixel clas-sification,it is suitable for dealing with the land cover classification problem of remote sensing images. However,the loss of feature maps caused by the downsampling process will lead to the appearance of discontinuous unlabeled void areas inside continuous large - scale ground objects,and the bilinear interpolation algorithm will lose the details of segmentation edges. In response to the above problems,we propose a V3plus-EN-TC model improved based on DeepLab V3+. The backbone network is replaced with EfficientNet,which has a stronger feature extraction ability. The SE module and inverted residual connections are introduced to enhance the Encoder’s ability to perceive and extract channel information and multi - scale spatial information. Features at three levels are fused, and an upsampling method combining transposed convolution and bilinear interpolation is adopted to improve the feature interpretation ability of the Decoder,suppress the appearance of void areas,and improve the edge accuracy. The DiceFocal combined loss function is utilized to solve the problem of unbalanced sample distribution and further focus on mixed pixels. On the preprocessed remote sensing dataset GID,compared with models such as FCN,U-Net,SegNet,PSPNet,CBAM-DeepLab V3 +,and CRF-DeepLab V3 +,the improved model V3plus-EN-TC has significantly fewer void areas and improved model accuracy. The mean intersection over union, F1 score,and mean pixel accuracy of the improved model reach 84. 74% ,88. 39% ,and 86. 64% respectively.

相似文献/References:

[1]蒋璐璐 王适 王宝成 李慧敏 李鑫慧.一种改进的标记分水岭遥感图像分割方法[J].计算机技术与发展,2010,(01):36.
　JIANG Lu-lu,WANG Shi,WANG Bao-cheng,et al.Segmentation of Remote Sensing Image Based on an Improved Labeling Watershed Algorithm[J].,2010,(07):36.
[2]赵艳飞 高清维 卢一相.基于多尺度Retinex算法的遥感图像增强[J].计算机技术与发展,2008,(02):70.
　ZHAO Yan-fei,GAO Qing-wei,LU Yi-xiang.Remote Sensing Image Enhancement Based on Multi- Scale Retinex Algorithm[J].,2008,(07):70.
[3]陈昌生 许文波 赵健.基于ArcSDE的卫星遥感影像数据库研究[J].计算机技术与发展,2010,(07):80.
　CHEN Chang-sheng,XU Wen-bo,ZHAO Jian.Research on Remote Sensing Image Database Based on ArcSDE[J].,2010,(07):80.
[4]李晓飞 马大玮 范小麟 胡焰智.基于数学形态学的遥感图像多感兴趣区域提取[J].计算机技术与发展,2007,(12):56.
　LI Xiao-fei,MA Da-wei,FAN Xiao-lin,et al.Extraction of ROIs Based on Mathematical Morphology toward Applications on Remote Sense Image Processing[J].,2007,(07):56.
[5]董敏[],阎镇.基于CUDA的航天遥感图像实时快视系统的研究[J].计算机技术与发展,2014,24(06):32.
　DONG Min[][][],YAN Zhen[].Research on Aerospace Remote Sensing Images Real-time Quick-view System Based on CUDA[J].,2014,24(07):32.
[6]张亚亚,张立民,刘小伟,等. 基于语义网的遥感图像分类[J].计算机技术与发展,2015,25(05):218.
　ZHANG Ya-ya,ZHANG Li-min,LIU Xiao-wei,et al. Remote Sensing Image Classification Based on Semantic-net[J].,2015,25(07):218.
[7]黄涛[][],薛丰昌[],钱洪亮[],等. 基于 NSCT 和自适应模糊阈值遥感图像去噪算法[J].计算机技术与发展,2016,26(01):65.
　HUANG Tao[][],XUE Feng-chang[],QIAN Hong-liang[],et al. Remote Sensing Image Denoising Algorithm Based on NSCT and Adaptive Fuzzy Threshold[J].,2016,26(07):65.
[8]龚林松,李士进.基于改进的SLIC和OTSU的遥感影像水体提取[J].计算机技术与发展,2019,29(01):145.[doi:10. 3969 / j. issn. 1673-629X. 2019. 01. 030]
　GONG Lin-song,LI Shi-jin.Water Information Extraction from Remote Sensing Imagery Based onImproved SLIC and OTSU[J].,2019,29(07):145.[doi:10. 3969 / j. issn. 1673-629X. 2019. 01. 030]
[9]朱小明.基于多光谱遥感图像信息的水质污染监测研究[J].计算机技术与发展,2018,28(11):52.[doi:10.3969/ j. issn.1673-629X.2018.11.012]
　ZHU Xiao-ming.Research on Water Quality Monitoring Based on Multi-spectral Remote Sensing Imagery[J].,2018,28(07):52.[doi:10.3969/ j. issn.1673-629X.2018.11.012]
[10]岳钦崟,李 虎.基于 MPI 并行遥感典型地类提取应用[J].计算机技术与发展,2020,30(07):104.[doi:10. 3969 / j. issn. 1673-629X. 2020. 07. 023]
　YUE Qin-yin,LI Hu.Application of MPI-based Parallel Remote Sensing for Typical Ground Classification Extraction[J].,2020,30(07):104.[doi:10. 3969 / j. issn. 1673-629X. 2020. 07. 023]
[11]李松宇.基于 HED—UNet 遥感图像建筑物语义分割[J].计算机技术与发展,2022,32(S2):58.[doi:10. 3969 / j. issn. 1673-629X. 2022. S2. 010]
　LI Song-yu.Semantic Segmentation of Buildings in Remote Sensing Images Based on HED-UNet[J].,2022,32(07):58.[doi:10. 3969 / j. issn. 1673-629X. 2022. S2. 010]
[12]王学伟.基于递归特征与边缘解耦的遥感图像语义分割[J].计算机技术与发展,2025,(03):40.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0325]
　WANG Xue-wei.Semantic Segmentation of Remote Sensing Images Based on Recursive Features and Edge Decoupling[J].,2025,(07):40.[doi:10.20165/j.cnki.ISSN1673-629X.2024.0325]

更新日期/Last Update: 2025-07-10