A Multi-Scale Superpixel-Guided Filter Feature Extraction and Selection Approach for Classification of Very-High-Resolution Remotely Sensed Imagery

Abstract

In this article, a novel feature selection-based multi-scale superpixel-based guided filter (FS-MSGF) method for classification of very-high-resolution (VHR) remotely sensed imagery is proposed. Improved from the original guided filter (GF) algorithm used in the classification, the guidance image in the proposed approach is constructed based on the superpixel-level segmentation. By taking into account the object boundaries and the inner-homogeneity, the superpixel-level guidance image leads to the geometrical information of land-cover objects in VHR images being better depicted. High-dimensional multi-scale guided filter (MSGF) features are then generated, where the multi-scale information of those land-cover classes is better modelled. In addition, for improving the computational efficiency without the loss of accuracy, a subset of those MSGF features is then automatically selected by using an unsupervised feature selection method, which contains the most distinctive information in all constructed MSGF features. Quantitative and qualitative classification results obtained on two QuickBird remotely sensed imagery datasets covering the Zurich urban scene are provided and analyzed, which demonstrate that the proposed methods outperform the state-of-the-art reference techniques in terms of higher classification accuracies and higher computational efficiency.

A Novel Feature Fusion Approach for VHR Remote Sensing Image Classification

“…In the intergroup fusion step, by taking into account the distinctive features at different fusion scales, the size of the moving window z and r were set to [3,10] to search more significant feature information. For the compared GF algorithm, the parameter  was manually fixed to 0.02 due to the fact that it has less influence on the classification results.…”

“…Unlike the traditional moderate resolution multispectral images, VHR images are characterized by a higher spatial detail of land objects, so context coherence and spatial patterns are as important as spectral information in the classification process in order to produce an accurate land-cover thematic map. In the literature, many advanced techniques have been proposed to utilize multiple features, especially spectral-spatial features and to improve the classification or detection performance, such as the morphological reconstruction [7], the attribute profiles (AP) [8], the edge-preserving filtering [9], the superpixel segmentation [10], and the convolutional neural networks [11][12][13][14]. Despite their effectiveness in extracting multi-scale spectral-spatial features, many of them do not address the feature fusion problem due to increasing feature complexity and computational cost, which may limit their utilization in practical applications.…”

Knowledge and Geo-Object Based Graph Convolutional Network for Remote Sensing Semantic Segmentation

“…For better modeling of the multi-scale information of land-cover classes in remote sensing images, ref. [ 35 ] integrates high-dimensional multi-scale guided filter (MSGF) features with the superpixel-level guidance image. Ref.…”

Attention-Based Context Aware Network for Semantic Comprehension of Aerial Scenery

“…Ref. [ 45 ] advocated a new method of a multi-scale super-pixel guidance filter based on feature selection. By processing high-dimensional and multi-scale guidance filters, considering the boundary and internal consistency of objects, it can better explain the geometric information of land covering objects in high-definition images.…”