Road Extraction from UAV Images Using a Deep ResDCLnet Architecture

Boonpook, Wuttichai; Tan, Yumin; Bai, Bingxin; Xu, Bo

doi:10.1080/07038992.2021.1913046

Cited by 9 publications

(8 citation statements)

References 22 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…1) Experimental results on Massachusetts Road Dataset: On the test set of the Massachusetts roads dataset, we performed ten existing deep-learning-based road extraction techniques. To verify the impact of designed AA-ResUNet on the road extraction, several mainstream existing segmentation networks including UNet [10], SegNet [37], DeepLabV3+ [16], D-LinkNet [17], GL-DenseUNet [15], DA-RoadNet [18], ResDCLNet [19], ResUNet [27], D-ResUNet [20], and SE-ResUNet [21] were used to assess the performance of the proposed method. In Fig.…”

Section: Resultsmentioning

confidence: 99%

“…2) Experimental results on DeepGlobe Road Dataset: The experimental results of the proposed technique show its effectiveness of the method. We compared the AA-ResUNet with existing methods UNet [10], SegNet [37], DeepLabV3+ [16], D-LinkNet [17], GL-DenseUNet [15], DA-RoadNet [18], ResDCLNet [19], ResUNet [27], D-ResUNet [20], and SE-ResUNet [21]. The training, validating, and testing of image construction, as explained in Section IV-A2 was used for the experimentation.…”

Section: Resultsmentioning

confidence: 99%

“…Note that these four modalities can be easily inserted into various baseline networks. To ensure a fair experimental comparison, we conduct the ablation experiment on an additional UAV road dataset, which is originally used in [42]. The UAV road dataset contains RGB and labelled images of size 480 × 480 × 3, respectively.…”

Section: Resultsmentioning

confidence: 99%

“…This network requires to fine-tune integrity for detecting the complex road structure. Boonpook et al [19] have developed deep residual deconvolution network (Deep ResDCLnet) for road extraction. Although, this network required postprocessing methods to reduce segmentation error.…”

Section: Related Workmentioning

confidence: 99%

“…The bigger the circle, the faster the speed (proposed method marked as a red point). Results of existing segmentation methods, including UNet [10], Seg-Net [37], DeepLabV3+ [16], D-LinkNet [17], GL-DenseUNet [15], DA-RoadNet [18], ResDCLNet [19], ResUNet [27], D-ResUNet [20], SE-ResUNet [21], and our AA-ResUNet. time-consuming approach.…”

Section: Introductionmentioning

confidence: 99%

See 4 more Smart Citations

An Attention Augmented Convolution based Improved Residual UNet for Road Extraction

Patil¹

2022

Preprint

View full text Add to dashboard Cite

<p>Recently remote sensing images have become more popular due to improved image quality and resolution. These images have been shown to be a valuable data source for road extraction applications like intelligent transportation systems, road maintenance, and road map making. In recent decades, the use of highly significant deep learning in automatic road extraction from these images has been a hot research area. However, fully automated and highly accurate road extractions from remote sensing images remain a challenge due to topology differences, complicated image backgrounds, and complex contexts. This paper proposes novel attention augmented convolution-based residual UNet architecture (AA-ResUNet) for road extraction, which adopts powerful features of self-attention mechanism and advantageous properties of residual UNet structure. The self-attention mechanism uses attention augmented convolutional operation to capture long-range global information; however, traditional convolution has a fundamental disadvantage: it only performs on local information. Therefore, we use the attention augmented convolutional layer as an alternative to standard convolution layers to obtain more discriminant feature representations. It allows to develop a network with fewer parameters. We also adopt improved residual units in standard ResUNet to the speedup training process and enhance the segmentation accuracy of the network. Experimental results on Massachusetts, DeepGlob Challenge, and UAV Road Dataset show that the AA-ResUNet performs well in road extraction, with Intersection over Union (IoU) (94.27%), lower trainable parameters (1.20 M), and inference time (1.14 sec). Comparative results on the proposed method have proven the supremacy or compatibility in road extraction with ten recently established deep learning approaches.</p>

show abstract