Community Content
This article has been written and edited by the research team. It has not passed through the quality control procedures applied to content produced by the Research Outreach team.

Recognition of urban water bodies using deep learning with multi-source and multi-temporal high spatial resolution remote sensing imagery

We use the excellent “self-learning ability” of deep learning to construct a modified structure, employing the method of Mask-RCNN, which integrates bottom-up and top-down processes for water recognition. Compared with traditional approaches, our method is completely data-driven without requiring prior knowledge — a novel technical procedure for water body recognition with practical engineering applications.

Data from High Spatial Resolution Remote Sensing Imagery (HSRRSI) provides detailed information required for the recognition of surface water bodies, including the texture, geometric structure and spatial distribution of these liquid masses. The comprehensive information gathered means that the internal components of surface water bodies can be represented, and the relationship between adjacent objects are better reflected. In the context of the “Big Data Era”, we have witnessed significant improvements in recognition methods using geometric processing remote sensing data, e.g. Geographic Object-Based Image Analysis (GEOBIA). However, these methods focus mainly on bottom-up classifications of visual features to semantic categories, ignoring top-down feedback capable of optimising recognition results, and thus cannot meet the needs of current large-scale applications. Due to the powerful feature extraction and representation capability of deep learning, using a special Convolutional Neural Network (CNN) structure-based region proposal generation and object detection integrated framework greatly promotes the performance of object detection for HSRRSI; providing a new model for remote sensing water body recognition. We use the excellent “self-learning ability” of deep learning to construct a modified structure of Mask-RCNN, which integrates bottom-up and top-down processes for water recognition. Compared with traditional approaches, our method is completely data-driven without requiring prior knowledge, a novel technical procedure for water body recognition with practical engineering applications. Experimental results indicate that the method produces accurate recognition results for multi-source and multi-temporal water bodies – and can effectively avoid confusing shadows and other ground features with water bodies.

Aiming to address the problems of poor automation, low efficiency and overgeneralisation inherent in routine remote sensing image water extraction, we built a region based solution network structure based on Mask R-CNN; which is suitable for HSRRSI water recognition, object-based recognition and automatic extraction of water bodies. This method is completely data-driven and does not require prior knowledge. It improves the speed of water extraction, provides technical support for the real-time monitoring of surface water, and provides data support for water related research.

As shown in Fig.1, featuring the WorldView-3 images and GF-2 images of Tongzhou District of Beijing, we applied the modified Mask-RCNN model (trained with four different datasets) to conduct the experiments for object recognition of water bodies. The results of the experiment show that the proposed method successfully improves water recognition in four types of remote sensing imagery. For both the remote sensing of large water bodies (such as lakes and rivers) and small water bodies (such as paddy fields and small tributaries) improved results were seen, regarding the range extracted for the water bodies. Remarkably, the cross-validation of the four-image training model shows accurate results and strong generalisation ability – even if the training area is different from the test area. In addition to the extraction of water data, this method can also be applied to the accurate extraction of vegetation, desert, wetland and other thematic information – areas certainly worth further study in the future.


  1. Zhu C, Luo J, Shen Z. River Linear Water Adaptive Auto-extraction on Remote Sensing Image Aided by DEM[J]. Acta Geodaetica et Cartographica Sinica, 2013, 42(2): 27–283.

  2. Sheng Y W and Xiao Q G. 1994. Water-body Identification in Cloud Contaminated NOAA/AVHRR Image. Environmental Remote Sensing, 9(4): 247–255.

  3. Zhou C H, Du YY and Luo J C. 1996. A Description Model Based on Knowledge for Automatically Recognizing Water from NOAA/AVHRR. Journal of Natural Disasters, 5(3): 100–108.

  4. Du YY and Zhou C H. 1998. Automatically Extracting Remote Sensing Information for Water Bodies. Journal of Remote Sensing, 2(4): 264–269.

  5. Yang C J and Xu M. 1998. Study on the Water-body Extraction Methods of Remote Sensing Information Mechanism. Geographic research, 17: 86–89.

  6. Du J K, Huang Y S, Feng X Z and Wang Z L. 2001. Study on Water Bodies Extraction and Classification from SPOT Image. Journal of Remote Sensing, 5(3): 214–219.

  7. He Z Y, Zhang X C, Huang Z C and Jiang H X. 2004. A Water Extraction Technique Based on High-spatial Remote Sensing Images. Journal of Zhejiang University, 31(6): 701–707. DOI:10.1007/s11769-003-0057-9

  8. Liu H and Jezek K C. 2004. Automated Extraction of Coastline from Satellite Imagery by Integrating Canny Edge Detection and Locally Adaptive Thresholding Methods. International Journal of Remote Sensing, 25(5): 937–958. DOI:10.1080/0143116031000139890

  9. Chunqi Z, Congsheng Z, Zhengzheng L. Study on Terrestrial Water Information Identified Based on the Analysis of Spectral Signature and Ratio Index. Geo-Information Science, 2008,10(5): 663–669.

  10. Jian-Cheng L, Yong-Wei S, Zhan-Feng S, et al. Automatic and High-precise Extraction for Water Information from Multispectral Images with the Step-by-Step Iterative Transformation Mechanism. Journal of Remote Sensing, 2009, 13(4): 604–615. DOI: 10.3321/j.issn:1007-4619.2009.04.005

Written By

Liu Jianhua
Beijing University of Civil Engineering and Architecture

Contact Details


School of Geomatics and Urban Spatial Information
Beijing University of Civil Engineering and Architecture

Want to read more articles like this?

Sign up to our mailing list and read about the topics that matter to you the most.
Sign Up!

Leave a Reply

Your email address will not be published. Required fields are marked *

Thank you for expressing interest in joining our mailing list and community. Below you can select how you’d like us to interact with you and we’ll keep you updated with our latest content.

You can change your preferences or unsubscribe by clicking the unsubscribe link in the footer of any email you receive from us, or by contacting us at at any time and if you have any questions about how we handle your data, please review our privacy agreement.

Would you like to learn more about our services?

We use MailChimp as our marketing automation platform. By clicking below to submit this form, you acknowledge that the information you provide will be transferred to MailChimp for processing in accordance with their Privacy Policy and Terms.

Subscribe to our FREE PUBLICATION