基于改进Deeplabv 3+模型的遥感影像地物语义分割方法研究  

The research on semantic segmentation of remote sensing image about ground objects based on improved Deeplabv 3+model

作　　者：南国君 王敏[1] 都海波[1] 谢枫 许水清 NAN Guo-jun;WANG Min;DU Hai-bo;XIE Feng;XU Shui-qing(School of Electrical Engineering and Automation,Hefei University of Technology,Hefei 230009,China;China Energy Engineering Group Anhui Electric Power Design Institute Co.,Ltd.,Hefei 230000,China)

机构地区：[1]合肥工业大学电气与自动化工程学院,合肥230009 [2]中国能源建设集团安徽省电力设计院有限公司,合肥230000

出　　处：《控制与决策》2025年第2期423-431,共9页Control and Decision

基　　金：国家自然科学基金项目(62073113,62003122);安徽省自然科学基金项目(2208085UD15)。

摘　　要：面向电力自动化领域,针对在遥感影像关键地物信息提取过程中,地物类别分布不均衡和不同域场景风格差异较大带来提取效果一般的问题,采用一种改进Deeplabv 3+语义分割网络.首先,在主干网络ResNet 101中使用IBN模块,用于增强模型对风格差异较大的遥感影像的泛化能力,同时为了进一步提高模型的分割精度,在网络中加入SE模块,加强重要的通道信息,缓解信息丢失问题;然后,损失函数使用Dice+Focal的联合损失函数,Dice Loss损失函数可缓解类别分布不均衡对小目标提取的影响,Focal Loss损失函数不仅可以使得模型更关注分类困难的目标,还可以改善Dice Loss造成的网络训练的不稳定.实验结果表明:所提出改进Deeplabv 3+与原Deeplabv 3+模型相比,将F 1-Score提高了7.78%,Intersection over Union提高了5.78%;与其他主流语义分割模型(包括FCN、UNet、SegNet)相比,所提出改进Deeplabv 3+在地物提取中实现了更好的分割精度.Aiming at the field of power automation,this paper considers the problem of general extraction effect caused by the unbalanced distribution of ground object categories and the large differences in scene styles in different domains in the process of extracting key feature information from remote sensing images,and an improved Deeplabv 3+semantic segmentation network is proposed.The instance-batch normalization(IBN)module is used in the backbone network ResNet 101 to enhance the model’s generalization ability for remote sensing images with large differences in styles.The squeeze-and-excitation(SE)module is added to the network to strengthen important channel information,alleviating the problem of information loss.The joint loss function of Dice Focal is adopted as the loss function,and Dice Loss can alleviate the impact of imbalanced category distribution on the extraction of small targets.Focal Loss can not only make the model pay more attention to objects that are difficult to classify,but it can also improve the instability of network training caused by Dice Loss.Experimental results show that in comparison with the original Deeplabv 3 model,the improved Deeplabv 3 improves F 1-Score by 7.78%and Intersection over Union(IoU)by 5.78%.In comparison with other mainstream semantic segmentation models(including FCN,UNet,and SegNet),the improved Deeplabv 3+also achieves better segmentation accuracy in ground objects extraction.

关 键 词：语义分割 Deeplabv 3+ IBN模型 遥感影像 损失函数 地物提取 

分 类 号：TP273[自动化与计算机技术—检测技术与自动化装置]