Publications

Li, GM; Tan, L; Liu, X; Kan, AK (2022). Network for Very-High-Resolution Urban Imagery Classification. PHOTOGRAMMETRIC ENGINEERING AND REMOTE SENSING, 88(6), 399-405.

Abstract
In the process of manual image interpretation, the use of a combination of spectral and spatial features can aid in more accurately classifying urban land coverage. In this study, to simulate this procedure, we use two concurrent convolutional neural networks (CNNs) with two scales of input to represent fields of view corresponding to object detail and the related information among objects. In our approach, the results derived from every convolution process are retained and stacked together at the end of the convolution process. Thus, not only are the spectral and spatial features combined, but all the scales of spatial features are also considered. When applied to very-highresolution remote sensing images, our proposed model with its feature-based CNN achieves a noticeable improvement over other state-of-the-art methods, which helps to assess the urban environment to some extent. In addition, we show that the digital surface model features, either in image form or in numerical characteristic form, can improve the overall accuracy rate of current structures.

DOI:
10.14358/PERS.21-00055R2

ISSN:
2374-8079