SegRivWidth: River Width Measurement Using Deep Learning

Abstract

A precise river width measurement is crucial for various river modeling, habitat evaluation, flood risk analysis, and other hydrological environmental, and technical applications. Conventional techniques is time-consuming and error-prone, such as direct field measurements. Nowdays fast and accurate riven identification and its width measurement is highely needed to save human life during flud and other natural disaster. Recent deep learning technology can greately be applied to identify and measure width of river automatic, fast and accurate from a remote place. The proposed deep learning based method is executed in two steps, identification of river and river width measurement. Deep learning based segmentation is used to identify river from remote sensing image. The accuracy of the semantic segmentation to identify river depends on rich spatial data and the resolution of the remote sensing images. In this work proposed SegRivWidth algorithm for automatic river width measurement from segmented images.  The obtained results are compared with the ground truth river width and found better accuracy. The obtained results are also compared with the existing methods in terms of Average Absolute Error (AAE) and Root Mean Square Error (RMSE). The proposed SegRivWidth has an RMSE of 4.76 m and an AAE error of 2.16 m for the river width measurement.