Water Areas Segmentation from Remote Sensing Images Using a Separable Residual SegNet Network

Abstract

Changes on lakes and rivers are of great significance for the study of global climate change. Accurate segmentation of lakes and rivers is critical to the study of their changes. However, traditional water area segmentation methods almost all share the following deficiencies: high computational requirements, poor generalization performance, and low extraction accuracy. In recent years, semantic segmentation algorithms based on deep learning have been emerging. Addressing problems associated to a very large number of parameters, low accuracy, and network degradation during training process, this paper proposes a separable residual SegNet (SR-SegNet) to perform the water area segmentation using remote sensing images. On the one hand, without compromising the ability of feature extraction, the problem of network degradation is alleviated by adding modified residual blocks into the encoder, the number of parameters is limited by introducing depthwise separable convolutions, and the ability of feature extraction is improved by using dilated convolutions to expand the receptive field. On the other hand, SR-SegNet removes the convolution layers with relatively more convolution kernels in the encoding stage, and uses the cascading method to fuse the low-level and high-level features of the image. As a result, the whole network can obtain more spatial information. Experimental results show that the proposed method exhibits significant improvements over several traditional methods, including FCN, DeconvNet, and SegNet.

Integrating Gate and Attention Modules for High-Resolution Image Semantic Segmentation

“…The average size of the tiles is 2494 ×2064 pixels with a spatial resolution of 9 cm. Similar to previous studies, we use 16 tiles for training (ID: 1, 3, 5, 7,11,13,15,17,21,23,26,28,30,32,34,37) and the rest 17 tiles are for testing [52], [53].…”

“…It has a wide range of applications that can be roughly divided into two categories. One is to label a single category, such as road extraction [8], [9], building segmentation [10], [11], ship detection [12], cloud segmentation [13], [14], and water area segmentation [15]. The other is to label multiple categories all together [16]- [20].…”

Dual attention deep fusion semantic segmentation networks of large-scale satellite remote-sensing images

“…It can achieve excellent results in image segmentation with fewer samples. Afterwards, variant SegNet-based semantic segmentation architectures were developed to address goal-specific tasks (Weng, Xu, and Xia et al 2020;Wang, Wang, and Zhang et al 2017).…”

DFFAN: Dual Function Feature Aggregation Network for Semantic Segmentation of Land Cover

“…In recent years, with the improvement of computer hardware and the increasing demand for image processing in practical work, deep learning (DL) has made great progress in the field of security [8], handwritten digit recognition [9], human action recognition [10], financial trading [11], remote image processing [12][13][14][15][16][17], and others [18][19][20][21][22]. According to the study of Kussul et al [23] in processing land cover remote sensing images, deep learning algorithms are significantly better than machine learning algorithms such as the SVM.…”