PMNet: A Multi-branch and Multi-scale Fusion Convolutional Neural Network for Water Body Extraction of High-resolution Remote Sensing Images

Liu, Qi; Zhang, Ziwen; Liu, Xiaodong; Zhang, Yonghong; Du, Zihao

Abstract

Automatic extraction of water body information from high-resolution remote sensing images is one of the core tasks of remote sensing image interpretation. Since the complex multi-scale characteristics of high-resolution remote sensing images, it is difficult for traditional water body extraction methods and some basic deep convolutional neural networks to consider the global spatial relationships of the target objects. As a result, the extracted water body information often has insufficient integrity, rough shape, and blurred edges. Although some advanced deep learning networks have good results, these models are often too heavy to be deployed on some low-cost edge devices. In contrast, some lightweight deep learning network models are less resource intensive, but the extraction results on high-resolution remote sensing images are often unsatisfactory. Therefore, in order to trade-off the lightweight and accuracy, this paper proposes the PMNet which uses a pipelined multi-step aggregation method to acquire contextual information at different scales to learn richer image features with fewer parameters. Meanwhile, both horizontal and vertical dimensions of spatial information are considered in combination to capture the remote dependencies between pixels. The mode of combining multiple decoding branches is used in the decoding stage instead of the traditional single decoding branch, which can better learn the multi-scale nature of water objects. The model achieves an F1-score of 90.22 and 88.57 on Landsat-8 and GID remote sensing image datasets with lower model complexity, which is better than other semantic segmentation models.

Citation

Liu, Q., Zhang, Z., Liu, X., Zhang, Y., & Du, Z. (in press). PMNet: A Multi-branch and Multi-scale Fusion Convolutional Neural Network for Water Body Extraction of High-resolution Remote Sensing Images. Intelligent Automation and Soft Computing,

• Journal Article Type: Article
• Acceptance Date: Jun 28, 2023
• Deposit Date: Dec 13, 2023
• Print ISSN: 1079-8587
• Electronic ISSN: 2326-005X
• Publisher: Tech Science Press
• Peer Reviewed: Peer Reviewed
• Keywords: Artificial intelligence, water extraction, semantic segmentation, multi-branch, remote sensing image
• Public URL: http://researchrepository.napier.ac.uk/Output/3422274
• Publisher URL: https://techscience.com/journal/iasc