基于多特征差异的随机森林遥感影像变化检测

Change Detection of Remote Sensing Image Based on Random Forest with Multi-feature Differences

为充分发挥遥感影像中多种特征的优势及不同时相影像对象之间特征差异优势,本文利用预测精度高性能稳定的随机森林算法,提出一种基于特征差异的面向对象变化检测方法。首先,基于变化向量分析法对影像进行像元级变化检测,并多尺度分割检测结果;然后,提取每个对象在前后时相影像上的光谱、纹理特征及特征差值作为随机森林的输入数据,在像素级检测结果基础上选择分类样本构建分类模型;最后,利用训练好的分类模型提取最终的变化区域。实验结果表明,该方法能有效地利用对象特征差值提高变化检测精度。

In order to give full play to the advantages of multiple features in remote sensing images and the differences of features among different image objects,an object-oriented change detection method based on feature differences is proposed by using the random forest algorithm with high prediction accuracy and stable performance.Firstly,the change vector analysis is used to detect the pixel level change of the image,and the multi-scale segmentation results are obtained.Then it extracts the spectrum,texture features and feature difference of each object on the front and back temporal images as the input data of random forest and selects the classification samples to build the classification model based on the pixel level detection results.Finally,the trained classification model is used to extract the final change region.The experimental results show that this method can effectively improve the change detection accuracy by using the difference of object features.