Utilizing GCN-Based Deep Learning for Road Extraction from Remote Sensing Images

Abstract

The technology of road extraction serves as a crucial foundation for urban intelligent renewal and green sustainable development. Its outcomes can optimize transportation network planning, reduce resource waste, and enhance urban resilience. Deep learning-based approaches have demonstrated outstanding performance in road extraction, particularly excelling in complex scenarios. However, extracting roads from remote sensing data remains challenging due to several factors that limit accuracy: (1) Roads often share similar visual features with the background, such as rooftops and parking lots, leading to ambiguous inter-class distinctions; (2) Roads in complex environments, such as those occluded by shadows or trees, are difficult to detect. To address these issues, this paper proposes an improved model based on Graph Convolutional Networks (GCNs), named FR-SGCN (Hierarchical Depth-wise Separable Graph Convolutional Network Incorporating Graph Reasoning and Attention Mechanisms). The model is designed to enhance the precision and robustness of road extraction through intelligent techniques, thereby supporting precise planning of green infrastructure. First, high-dimensional features are extracted using ResNeXt, whose grouped convolution structure balances parameter efficiency and feature representation capability, significantly enhancing the expressiveness of the data. These high-dimensional features are then segmented, and enhanced channel and spatial features are obtained via attention mechanisms, effectively mitigating background interference and intra-class ambiguity. Subsequently, a hybrid adjacency matrix construction method is proposed, based on gradient operators and graph reasoning. This method integrates similarity and gradient information and employs graph convolution to capture the global contextual relationships among features. To validate the effectiveness of FR-SGCN, we conducted comparative experiments using 12 different methods on both a self-built dataset and a public dataset. The proposed model achieved the highest F1 score on both datasets. Visualization results from the experiments demonstrate that the model effectively extracts occluded roads and reduces the risk of redundant construction caused by data errors during urban renewal. This provides reliable technical support for smart cities and sustainable development.

Keywords: depthwise separable convolution; gradient operator; graph convolution; graph reasoning; road extraction; smart cities.

Figure 1

Sample images from the Langfang City dataset, Hebei Province. (a) Industrial area, (b) Residential area, (c) Rural area, (d) Densely populated residential area.

Figure 2

Samples of the DeepGlobe dataset.Both images (a,c) have obvious tree occlusions, and images (b,d) have the problem of similar backgrounds.

Figure 3

FR-SGCN model architecture.

Figure 4

Schematic diagram of the feature classification module.

Figure 5

Schematic diagram of the AM: (a) Illustration of applying the AM to spatial features post-feature separation. (b) Illustration of applying the AM to channel features post-feature separation.

Figure 6

The process of generating a new adjacency matrix for spatial features: the upper section depicts the adjacency matrix creation, while the lower section demonstrates the similarity matrix generation.

Figure 7

The procedure for forming a new adjacency matrix for channel features: the upper part shows the adjacency matrix production, and the lower part shows the similarity matrix development.

Figure 8

(a–c) Visualization comparison of different backbone architectures in FR-SGCN. Comparative experiments using different main architectures were conducted on the Langfang City dataset. Red boxes highlight the differences in road extraction results across architectures in complex backgrounds.

Figure 9

(a–c) Visual comparison of FR-SGCN and other DL models using the Langfang city road dataset. Performance Comparison Under Partial Occlusion and Low—Contrast Scenarios. Red boxes highlight differences in road extraction results across models under partial occlusion. Green circles demarcate variations in road extraction performance among different models in low-contrast scenarios where backgrounds resemble road surfaces.

Figure 10

(a–c) Visual comparison between FR-SGCN and other advanced deep learning models using test data from the Langfang city road dataset. Performance Comparison Under Partial Occlusion and Low—Contrast Scenarios. The red boxes highlight the differences in road extraction results among various models under certain occlusion conditions.

Figure 11

(a–d) Visual comparison of FR-SGCN and other deep learning models using test data from the DeepGlobe dataset. Performance Comparison Under Partial Occlusion and Low—Contrast Scenarios. Red boxes highlight differences in road extraction results across models under partial occlusion.

Figure 12

(a–c) Visual results of the ablation experiments for the proposed method using the Langfang city road dataset. Performance Comparison Under Partial Occlusion and Low—Contrast Scenarios. Red boxes highlight the impact of incorporating individual modules on extraction results in complex backgrounds.

Figure 13

(a) Comparison of loss curves and (b) comparison of accuracy curves for the two loss functions used during training with FR-SGCN. The two loss functions are the Dice loss and the proposed hybrid loss.