摘要
针对传统的变化检测模型单一且大多针对二维影像,并没有考虑三维信息的问题。研究并提出一种融合DOM(数字正射影像图)与DSM(数字表面模型)的三维变化检测算法,采用DC2GAN(Deep Convolutional Genrative Adversarial Networks)融合DOM与DSM数据,生成三维融合影像数据,并通过改进的Unet网络对其训练,得到变化检测模型,实现对三维变化类特征的学习。在worldview2数据上取得了90%的准确率、0.6KAPPA系数。实验结果表明,该方法在拥有较高变化检测准确率的同时还具有良好的泛化能力,能够用于实际工程。
The traditional change detection model is single and mostly for two-dimensional images, and does not consider the problem of three-dimensional information. Research and propose a three-dimensional change detection algorithm that combines DOM(digital orthophoto map) and DSM(digital surface model).DSM is extracted through a semi-global matching algorithm, and DC2GAN(Deep Convolutional Genrative Adversarial Networks) is used to fuse DOM and DSM data, generate 3D fusion image data. And through the improved Unet network training, the change detection model is obtained, and the learning of three-dimensional change characteristics is realized. Achieved a 90% accuracy rate and 0.6KAPPA coefficient on worldview2 data. The experimental results show that this method not only has high change detection accuracy rate, but also has good generalization ability, and can be used in practical engineering.
作者
杨晓冬
严华
Yang Xiaodong;Yan Hua(Beijing Daoda TianJi Science and Technology Co.,Ltd.,Beijing 100000,China)
出处
《科学技术创新》
2022年第28期145-149,共5页
Scientific and Technological Innovation
