基于机器学习的多源遥感影像融合土地利用分类研究

为了快速获取准确的城市土地利用信息,提高西南地区遥感影像城市土地利用分类信息提取的精度,探讨了当前快速发展的机器学习技术在该领域中的分类实验.选用昆明市主城区作为研究区域,以Landsat8与Sentinel-1A影像为原始数据,使用GS变换法对影像进行融合,使用卷积神经网络(Convolutional Neural Network,CNN)和BP神经网络(Back Propagation Network)2种分类算法对融合前后的遥感影像进行土地利用分类信息提取,对分类结果进行分析.研究结果表明:基于Landsat8和Sentinel-1A的融合影像数据的卷积神经网络分类算法具有最好的分类效果,其总体分类精度和Kappa系数分别为85.8091%,0.8124,认为基于多源遥感影像融合的卷积神经网络分类方法是获取准确的城市土地利用分类信息的一种可行的方法,可以为高原地区城市的土地利用分类提取研究参考.

On Land Use Classification by Means of Machine Learning Based on Multi-source Remote Sensing Image Fusion

In order to obtain accurate urban land use information quickly and improve the precision of land use classification information in high altitude areas by remote sensing image, the study deals with the exploration of the application of a rapidly developing technology, machine learning, in such fields. The main urban area of Kunming City was chosen as the case area in the research, taking Landsat8 and Sentinel-1A remote sensing image as the original data. Then the convolution neural network and BP neural network was used to extract the land use classification information of the remote sensing images before and after the fusion. After that the classification results were analyzed. Finally the results show that the classification method of convolutional neural network classification based on fused image data of Land sat 8 and Sentinel-1A had the best classification results, those overall classification accuracy and the Kappa coefficient reached 85.8091% and 0.8124. Therefore the classification method of convolutional neural network based on multi-source remote sensing image fusion is feasible to obtain accurate urban land use classification information, which provides a reference for the research of land use classification in urban areas of high altitude.