基于Segnet网络和迁移学习的全景街区影像变化检测

针对采用传统方法难以提高全景街区影像变化检测精度的问题,提出一种基于Segnet网络和迁移学习的全景街区影像变化检测方法。首先对数据集“TSUNAMI”进行预训练并对训练集进行分类归并;然后采用Segnet网络对全景街区影像进行语义分割,最后对语义分割结果进行差值运算,得到变化差异图并进行精度评价。实验选取两组全景街区影像,采用最大似然法、支持向量机方法(SVM,support vector machine)以及提出方法对这2组数据进行对比实验,第一组得到的精度分别为65.1%、72.1%和81.4%;第二组得到的精度分别为66.5%、70.6%、82.2%。实验结果表明提出的方法具有更高的变化检测精度,可为城市违章排查、灾后重建等提供技术支撑。

Complex street scene change detection based on segnet network and migration learning

The use of multi-temporal panoramic block images is of great significance for monitoring urban development and assisting government decision-making. However, due to the influence of solar rays, ground spectrum and shooting angle during the process of collecting data, it is difficult to obtain high precision by traditional methods. Complex neighborhood changes information. To this end, this paper proposes a method for detecting image change in panoramic blocks based on Segnet and migration learning. Firstly, the data set "TSUNAMI" is pre-trained and the training set is classified and merged. Then, the Segnet network is used to semantically segment the panoramic block image, and the semantic segmentation result is subjected to difference calculation to obtain the change result map and evaluate the accuracy. Experiments were carried out to select two groups of panoramic block images. The maximum likelihood method, the support vector machine method and the method proposed in this paper were used to compare the two groups of data. The accuracy of the first group was 65.1%, 72.1% and 81.4%, respectively. The accuracy of the second group was 66.5%, 70.6%, and 82.2%, respectively. The experimental results show that the proposed method has higher detection accuracy and can provide technical support for urban violation investigation, post-disaster reconstruction, and ancient cultural relics restoration.