Intelligent Mapping Study Of Urban Forests From High-Resolution Remotely Sensed Imagery Using Object-Based Method Coupled Deep Learning

The classification of urban land use information and the extraction of urban forests is closely related to the construction of smart cities.With the rapid development of modern remote sensing technologies,a huge amount of very high spatial resolution(VHSR)remotely sensed imagery is now commercially available,opening new opportunities to classify urban land use information and extract urban forests at a very detailed level.However,the improvement of spatial resolution enriches the details and features inside the remote sensing images,which will lead to the phenomenon of the “same spectrum but different objects” and “same object with different spectra” issues.Meanwhile,the fragmented distribution of urban land use types and the complex structure of urban forests also bring about a variety of challenges for urban land use mapping and the extraction of urban forests.Recently,the application of deep learning techniques,especially deep convolutional neural networks(DCNN),in the intelligent mapping of VHSR remote sensing images is the hot spot in domestic and foreign research.Considering the characteristics of urban land use categories and urban forest from VFSR images,this study proposes the U-Net-Dense Net-Coupled Network(UDN)based on the DCNN algorithm.By combining the idea of object-based multiresolution segmentation with UDN algorithm,a novel ObjectBased U-Net-Dense Net-Coupled Network(OUDN)algorithm for VFSR image classification is proposed.The proposed OUDN has three parts: the first part involves the coupling of the improved Unet and Dense Net architectures;then,the network is trained according to the labeled data sets,and the land use information in the study area is classified;the final part fuses the object boundary information obtained by object-based multiresolution segmentation into the classification layer,and a voting method is applied to optimize the classification results.In this paper,the classification features are divided into three groups: the original spectral features(Spe);the spectral features combined with vegetation index(Spe-Index);and the spectral features combined with texture features based on graylevel co-occurrence matrix(GLCM).Based on these three groups of classification features,this paper studies the classification performance of four deep learning algorithms,including improved U-net(U),improved Dense Net(D),UDN and OUDN.The results show that(1)the classification results of the OUDN algorithm are better than those of U-net and Dense Net,and the average classification accuracy based on three groups of classification features is 92.9%,an increase in approximately 3%;(2)for UDN and OUDN algorithms,the urban forest extraction accuracies are higher than those of U and D algorithms based different classification features,and the OUDN effectively alleviates the classification error caused by the fragmentation of urban distribution by combining object-based multiresolution segmentation features,making the overall accuracy(OA)of urban land use classification and the extraction accuracy of urban forests superior to those of the UDN algorithm;(3)based on the Spe-Texture(the spectral features combined with the texture features),the OA of the OUDN in the extraction of urban land use categories can reach93.8%,thereby the algorithm achieved the accurate discrimination of different land use types,especially urban forests(99.7%).Therefore,this study provides a reference for classification algorithm and feature setting for the mapping of urban land use information from VHSR imagery.