River Detection in Remotely Sensed Imagery Using Gabor Filtering and Path Opening

Abstract

Detecting rivers from remotely sensed imagery is an initial yet important step in space-based river studies. This paper proposes an automatic approach to enhance and detect complete river networks. The main contribution of this work is the characterization of rivers according to their Gaussian-like cross-sections and longitudinal continuity. A Gabor filter was first employed to enhance river cross-sections. Rivers are better discerned from the image background after filtering but they can be easily corrupted owing to significant gray variations along river courses. Path opening, a flexible morphological operator, was then used to lengthen the river channel continuity and suppress noise. Rivers were consistently discerned from the image background after these two-step processes. Finally, a global threshold was automatically determined and applied to create binary river masks. River networks of the Yukon Basin and the Greenland Ice Sheet were successfully detected in two Landsat 8 OLI panchromatic images using the proposed method, yielding a high accuracy (~97.79%), a high true positive rate (~94.33%), and a low false positive rate (~1.76%). Furthermore, experimental tests validated the high capability of the proposed method to preserve river network continuity.

A Natural-Rule-Based-Connection (NRBC) Method for River Network Extraction from High-Resolution Imagery

“…Mason et al [26] linked river centerlines using search triangles at river centerline endpoints that were restricted by terrain information and flow path direction. To detect river network from remotely sensed imagery, a recent method [27] used a Gabor filter to enhance river cross-sections first and then used path opening to lengthen the river channel continuity and suppress noise. These methods were designed to connect river segments represented by one-pixel wide lines.…”

Preface: Remote Sensing of Water Resources

“…In their study, Yang et al [4] successfully characterized and mapped a river network from the noisy image background using their Gaussian-like cross-section and longitudinal continuity. Their proposed methods were able to accurately classify images to a binary river map using automatically determined threshold.…”

Assessing Uncertainty in LULC Classification Accuracy by Using Bootstrap Resampling

“…Remote sensing images have been widely used for earth surface monitoring [1][2][3][4][5][6][7][8], environmental change detection [9][10][11][12][13][14], and water resource management [15][16][17][18][19][20][21][22]. Many of these applications require land-use/land-cover (LULC) classifications derived from multispectral images.…”