Problem Statement [Extraction of River Networks from Satellite Images using Image Processing & Deep Learning Techniques]
Paper 01: A Survey: Extraction of River Networks from Satellite Images (Research Survey for Technical Gaps) Abstract:
- River network extraction is crucial to keep track of the water resources. Various methods have been implemented in times series to yield profound and incisive outputs and are still being developed and combined with predefined available methods. We have carried out a structured survey on these methods and have presented them with their outputs. There are numerous Web sites available for data set collection. Some generalized methods are available like image processing, using predefined models, or developing user-defined algorithms. For image processing, various segmentation methods are available out of which clustering- and threshold-based segmentations are mostly used. Predefined models such as CNN, ResUNet, YOLO, Faster CNN, and MSCFF are available. These algorithms can be used for the extraction of river networks but might not yield higher accuracies. Hence, this paper concentrates mainly on approaches for the extraction of river networks from satellite images.
Link: https://link.springer.com/chapter/10.1007/978-981-19-5224-1_49
Paper 02: Extraction of River Networks from Satellite Images using Image Processing & Deep Learning Techniques (Implementation and Improvement of Survey Targeted Technical Gaps) Abstract:
- River networks are widely observed and scrutinized for various purposes, which incorporate determining the terrestrial positions of water bodies, examining the gauge levels of the river, predicting river flows, and conserving sustainable energy resources as a consequence of Global warming. Extraction of these River networks on digital imagery systems is executed by various segmentation and machine learning model integration. In this paper, distinct datasets are used from Kaggle and Google Earth Engine, Segmentation methods such as Image segmentation, gray scaling, enhancement, global thresholding, and Deep Learning UNet Architecture are integrated with contemplation of extracting river networks from satellite images which result in achieving 80.98 % dice score for the developed UNet Model. Hence, these developed techniques can further be used for river extraction from satellite images. And can be applied to various semantic segmentation detection datasets.
Link: https://ieeexplore.ieee.org/document/10088330