A survey of methods incorporating spatial information in image classification and spectral unmixing

Abstract

Over the past decade, the incorporation of spatial information has drawn increasing attention in multispectral and hyperspectral data analysis. In particular, the property of spatial autocorrelation among pixels has shown great potential for improving understanding of remotely sensed imagery. In this paper, we provide a comprehensive review of the state-of-the-art techniques in incorporating spatial information in image classification and spectral unmixing. For image classification, spatial information is accounted for in the stages of pre-classification, sample selection, classifiers, post-classification, and accuracy assessment. With regards to spectral unmixing, spatial information is discussed in the context of endmember extraction, selection of endmember combinations, and abundance estimation. Finally, a perspective on future research directions for advancing spatial-spectral methods is offered.

Applicability of multifractal features as global characteristics of WorldView-2 panchromatic satellite images

“…Aleksandrowicz et al [2016] used Hölder exponents to characterise texture at pixel neighborhood level. Incorporation of such textural (spatial) features into classification of multispectral and hiperspectral images constitutes one of actual research fields in remote sensing [Wang et al, 2016].…”

VPRS-Based Regional Decision Fusion of CNN and MRF Classifications for Very Fine Resolution Remotely Sensed Images

“…This is mainly because the MLP-MRF utilizes the spectral information in the classification process without fully exploiting the abundant spatial information appearing in the VFSR imagery (e.g. texture, geometry or spatial arrangement) [57]. Such deficiencies often lead to unsatisfactory classification performance in classes with spectrally mixed but spatially distinctive characteristics (e.g., the confusion and misclassification between Trees and Grassland or Low Vegetation that are spectrally similar, the severe salt and pepper effects on railway with linear textures, etc.…”

Developments in Landsat Land Cover Classification Methods: A Review

“…In pixel-based classification, it is assumed that a pixel is made up of one homogenous land cover type; however, many pixels record more than one land cover types [63,66,67]. Considering Landsat's ground resolution of between 60 and 30 m, a number of land cover classes can constitute a single pixel.…”

“…LSMA is designed to work with a fixed number of endmembers while MESMA can be used on pixels with different numbers of endmembers [23,77]. The major challenge for SMA is the errors in the final allocation of fractional endmembers resulting from spectral variability and similarity during the selection of endmembers [66,67].…”

“…This method was developed in the early 1980s (Figure 2) and has been applied extensively on Landsat land cover classification [63,66,67]. The output of SMA is represented as a fraction of each land cover type called endmembers [74].…”