Ghost引导UNet++的高分遥感影像变化检测  

Ghost-guided UNet++for high-resolution remote sensing image change detection

作　　者：王鑫[1] 李莹莹 张香梁 Wang Xin;Li Yingying;Zhang Xiangliang(School of Computer and Information,Hohai University,Nanjing 211100,China)

机构地区：[1]河海大学计算机与信息学院,南京211100

出　　处：《中国图象图形学报》2024年第5期1460-1478,共19页Journal of Image and Graphics

基　　金：国家自然科学基金项目(51979085);江苏省“六大人才高峰”项目(XYDXX-007);江苏政府留学奖学金项目(20194296)。

摘　　要：目的 随着遥感观测技术的飞速发展,遥感影像的分辨率越来越高,如何从高分遥感影像中有效提取具有鉴别性的特征进行地物变化检测成为一个具有挑战性的问题。卷积神经网络广泛应用于计算机视觉领域,但面向遥感影像变化检测时仍存在图像语义或位置信息的丢失及网络参数量过大等缺陷,导致检测性能受限。为此,提出一种新型GUNet++(Ghost-UNet++)网络,用于遥感影像的精准变化检测。方法 首先,为了提取双时相遥感影像更具判别性的深度特征,设计具有多分支架构的高分辨率网络HRNet替换传统UNet++的主干网;其次,采用UNet++解码结构进行差异判别时,引入鬼影(Ghost)模块代替传统卷积模块以降低网络参量,并设计密集跳跃连接进一步加强信息传输,以减少深层位置信息的丢失;最后,设计一个集成注意力模块,将网络的多个语义层次特征进行聚合和细化,抑制语义和位置信息的丢失,进一步增强特征表征能力用于最终的精准变化检测。结果 在LEVIR-CD(LEVIR change detection data set)和Google Data Set两个公开数据集上进行实验,结果表明本文算法变化检测精度高达99.62%和99.16%,且网络参数量仅为1.93 M,与现有主流变化检测方法相比优势明显。结论 提出方法综合考虑了遥感图像中语义和位置信息对变化检测性能的影响,具有良好的特征抽取和表征能力,因此变化检测的精度和效率比现有同类方法更高。Objective With the rapid development of remote sensing observation technology,the resolutions of remote sens⁃ing images(RSIs)are increasing.Thus,how to extract discriminative features effectively from high-resolution RSIs for ground-object change detection has become a challenging problem.The existing RSI change detection methods can be divided into two categories:methods based on conventional image processing approaches and methods based on deep learn⁃ing(DL)theory.The former extracts low-level or mid-level features from RSIs for change detection,making it easy to implement and have high detection efficiency.However,the increasing resolution of RSIs result in the images having rich ground objects and complex background clutter;thus,the low-or mid-level features can hardly meet the demand of precise change detection. In recent years, DL has been introduced into the field of high-resolution RSI change detection because ofits powerful feature extraction capability. Various methods based on convolutional neural networks (CNNs) have been pro⁃posed for RSI change detection. Compared with conventional image processing methods, CNNs can extract high-levelsemantic information for high-resolution RSIs, which is beneficial to precise detection. Although CNNs have greatly raisedthe accuracy of change detection, they always involve numerous parameters and have high computational complexity. Toraise the efficiency of change detection, many scholars have proposed to perform parameter pruning on pretrained models ordesign simple network structures. However, these strategies lead to the loss of some crucial image information, includingsemantics and location information, thus reducing the detection accuracy. Therefore, this study proposes a novel GhostUNet++ (GUNet++) network for precise RSI change detection to address the problems. Method First, a high-resolutionnetwork called HRNet, which has a multibranch architecture, is designed to replace the traditional UNet++ backbone andthus extract additional discriminative deep f

关 键 词：高分辨率遥感影像 变化检测 深度学习(DL) 鬼影模块 集成注意力 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]