Deep Learning-Based Key Indicator Estimation in Rivers by Leveraging Remote Sensing Image Analysis

Abstract

Estimation of key indicators in rivers was usually conducted with use of monitoring data of Internet of Things. Currently, it has been a more practical demand to extract key indexes in rivers from the perspective of visual remote sensing, rather than data-driven perspective. As consequence, this study aims to explore the deep learning-based key indicators extraction method from remote sensing images of rivers. First of all, a large amount of river-related remote sensing images, including high-resolution satellite images and aerial photographs, are collected. Then, U-Net structure is utilized as the backbone network to realize semantic segmentation via multimodal feature fusion. On this basis, fine-grained vision features are extracted to estimate values of key indicators. Finally, the width and flow velocity of the river are identified and verified. Using deep convolutional neural networks and recurrent neural networks for feature extraction and modeling, the model can infer and estimate relevant information of rivers by learning key features from images. Empirically, we have also carried out some experiments on real-world remote sensing images to evaluate the proposal. The results indicate that our method performs well in extracting key indicators of rivers, and has higher accuracy compared to traditional methods. In addition, we also conduct sensitivity analysis on the model and find that it has certain stability to factors that affect river characteristics, such as geographical environment and climate conditions.