ERN: Edge Loss Reinforced Semantic Segmentation Network for Remote Sensing Images

Abstract

Abstract:The semantic segmentation of remote sensing images faces two major challenges: high inter-class similarity and interference from ubiquitous shadows. In order to address these issues, we develop a novel edge loss reinforced semantic segmentation network (ERN) that leverages the spatial boundary context to reduce the semantic ambiguity. The main contributions of this paper are as follows: (1) we propose a novel end-to-end semantic segmentation network for remote sensing, which involves multiple weighted edge supervisions to retain spatial boundary information; (2) the main representations of the network are shared between the edge loss reinforced structures and semantic segmentation, which means that the ERN simultaneously achieves semantic segmentation and edge detection without significantly increasing the model complexity; and (3) we explore and discuss different ERN schemes to guide the design of future networks. Extensive experimental results on two remote sensing datasets demonstrate the effectiveness of our approach both in quantitative and qualitative evaluation. Specifically, the semantic segmentation performance in shadow-affected regions is significantly improved.

Urban Land Cover Classification of High-Resolution Aerial Imagery Using a Relation-Enhanced Multiscale Convolutional Network

“…Table 1 shows the classification results in terms of the per-class accuracy, the mean F 1 score, the mIoU, and the overall accuracy (OA) of experiments on the Vaihingen dataset. As a comparison, we also show the results recently published by SVL-boosting + CRF [54], RF + dCRF [55], RotEqNet [56], HSNet [19], and ENR [34] with a model of the same architecture and size as ours. It is demonstrated that REMSNet outperforms other methods in terms of the mean F 1 score, mIoU, and overall accuracy.…”

“…To explicitly represent class boundaries, Marmanis et al [33] added boundary detection to the SegNet to further refine classification results. Liu et al [34] introduced multiple weighted edge supervisions based on the encoder-decoder framework to leverage the spatial boundary context and reduce the semantic ambiguity. Some work also uses multi-source data fusion data to reduce ambiguous boundary.…”

“…Six categories were defined in the dataset including impervious surfaces, building, low vegetation, tree, car, and clutter/background, as shown in Figure 4. We selected five images (image IDs: 11,15,28,30,34) to evaluate our network; the remaining images were used for training, following the procedure in [19,34,[54][55][56]. To reduce the impact of uncertain border definitions, another version of the ground truth with eroded boundaries is provided, on which the accuracy is measured [19,34,[54][55][56].…”

Improved Winter Wheat Spatial Distribution Extraction Using A Convolutional Neural Network and Partly Connected Conditional Random Field

“…Based on previous studies [51][52][53]58,59], approximately 80% of the pixel-by-pixel classification results generated by CNN models are credible. Therefore, only approximately 20% of the pixel classification results require post-processing.…”

“…RefineNet [52], form the basis for the rapidly developing field of remote sensing image segmentation. Although the use of CNNs can significantly improve the accuracy of remote sensing image segmentation, errors remain common near object edges owing to the inherent characteristics of the convolution operation [49][50][51]53]. Thus, convolution must be combined with other post-processing techniques to improve the accuracy of the results [51,54,55].…”