基于俯观相似度的遥感图像变化目标检测  

Change Object Detection for Remote Sensing Image Based on Top-view Similarity

作　　者：王建社 

机构地区：[1]合肥讯飞数码科技有限公司人工智能研究院,安徽合肥230088

出　　处：《工业控制计算机》2024年第6期21-23,26,共4页Industrial Control Computer

摘　　要：提出了一种基于俯观相似度的光学遥感图像变化目标检测方法。对输入的两幅同一区域不同时段拍摄的光学遥感图像进行配准后,分别进行目标检测,提取出所有感兴趣的目标区域。采用残差网络提取目标的俯观特征,提升了特征的鲁棒性。通过计算目标间的俯观相似度,利用距离判据和相似度判据判断目标是否发生变化,实现变化目标检测。实验结果表明,提出的方法较现有的方法具有更好的性能表现,平均F1可以达到0.837。In this paper,a change object detection method based on top-view similarity for optical remote sensing images is proposed.After registering the two optical remote sensing images taken in different periods in the same area,the object detection is carried out respectively to extract all the target areas of interest.The residual network is used to extract the top-view features of objects,which improves the robustness of the features.The change of objects is measured by calculating the top-view similarity between objects,and the distance criterion and similarity criterion are proposed to judge whether the object has changed,so as to realize the change object detection.The experimental results show that the proposed method has better performance than the existing methods,and the average F1 can reach 0.837.

关 键 词：光学遥感图像 感兴趣目标 变化目标检测 俯观相似度 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]