Pyramidal Image Segmentation Based on U-Net for Automatic Multiscale Crater Extraction

Hong, Zhonghua; Zhai, Fan; Zhou, Ruyan; Pan, Haiyan; Zhang, Yun; Han, Yanling; Wang, Jing; Yang, Sheng; Jin, Yanmin

doi:10.18494/sam3564

Cited by 3 publications

(1 citation statement)

References 14 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Therefore, many scholars applied deep learning methods to change detection tasks and achieved satisfactory results because deep learning has significantly better generalization abilities than traditional methods. (7,8) During the last few years, many neural networks have emerged and have been applied to the challenge of the detection of changes in images by remote sensing; convolutional neural networks (CNNs) are the most commonly used. Especially with the development of a fully convolutional network (FCN) that has the ability to make dense predictions, FCN and its adaptations soon became suitable choices for the task of change detection.…”

Section: Introductionmentioning

confidence: 99%

Change Detection for High-resolution Remote Sensing Images Based on a UNet-like Siamese-structured Transformer Network

Chen¹,

Chen²,

Liu³

et al. 2023

Sensors and Materials

View full text Add to dashboard Cite

Change detection using high-resolution remote sensing images provides crucial information for geospatial monitoring, which is of great importance as urbanization continues. However, current deep learning models for change detection tasks are mostly based on convolutional neural networks (CNNs), from which it is difficult to extract global information owing to the locality of convolution operations. In this paper, we propose a deep learning model, Siam-Swin-UNet (SSUNet), for remote sensing change detection. SSUNet is designed following the classic UNet-like encoder-decoder framework but has three major innovations: (1) The encoder and decoder are pure transformer-based and hierarchically structured, which avoids the locality problem of CNN but retains the capability of hierarchical representation. (2) The encoder incorporates the Siamese structure, which can process bi-temporal remote sensing images in parallel, and to which is added a fusion module to properly fuse the feature maps extracted from the Siamese structure. (3) The backbone of the SSUNet is Swin Transformer V2 blocks, which can be more stable in further applications of the model, such as transfer learning or scaling up of the model capacity. We experimented with the proposed SSUNet on the LEVIR-CD dataset, along with CNN-based models such as UNet, UNet++, FC-Siam-Conc, and FC-Siam-Diff. The results showed our model outperformed the CNN-based models by a large margin based on evaluation metrics including precision, recall, F1-score, and overall accuracy (OA). Moreover, we conducted ablation studies to further prove the effectiveness of the Siamese structure and the choice of the backbone. The proposed SSUNet has great potential for use in remote sensing change detection tasks.

show abstract

Section: Introductionmentioning

confidence: 99%

Change Detection for High-resolution Remote Sensing Images Based on a UNet-like Siamese-structured Transformer Network

Chen¹,

Chen²,

Liu³

et al. 2023

Sensors and Materials

View full text Add to dashboard Cite

show abstract

Automated Multi-class Crater Segmentation in Mars Orbital Images

Malvi,

Shah,

Chandarana

et al. 2023

Proceedings of the 6th ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery

View full text Add to dashboard Cite

MC-UNet: Martian Crater Segmentation at Semantic and Instance Levels Using U-Net-Based Convolutional Neural Network

Chen

Mathiopoulos

et al. 2023

Remote Sensing

View full text Add to dashboard Cite

Crater recognition on Mars is of paramount importance for many space science applications, such as accurate planetary surface age dating and geological mapping. Such recognition is achieved by means of various image-processing techniques employing traditional CNNs (convolutional neural networks), which typically suffer from slow convergence and relatively low accuracy. In this paper, we propose a novel CNN, referred to as MC-UNet (Martian Crater U-Net), wherein classical U-Net is employed as the backbone for accurate identification of Martian craters at semantic and instance levels from thermal-emission-imaging-system (THEMIS) daytime infrared images. Compared with classical U-Net, the depth of the layers of MC-UNet is expanded to six, while the maximum number of channels is decreased to one-fourth, thereby making the proposed CNN-based architecture computationally efficient while maintaining a high recognition rate of impact craters on Mars. For enhancing the operation of MC-UNet, we adopt average pooling and embed channel attention into the skip-connection process between the encoder and decoder layers at the same network depth so that large-sized Martian craters can be more accurately recognized. The proposed MC-UNet is adequately trained using 2∼32 km radii Martian craters from THEMIS daytime infrared annotated images. For the predicted Martian crater rim pixels, template matching is subsequently used to recognize Martian craters at the instance level. The experimental results indicate that MC-UNet has the potential to recognize Martian craters with a maximum radius of 31.28 km (136 pixels) with a recall of 0.7916 and F1-score of 0.8355. The promising performance shows that the proposed MC-UNet is on par with or even better than other classical CNN architectures, such as U-Net and Crater U-Net.

show abstract

Pyramidal Image Segmentation Based on U-Net for Automatic Multiscale Crater Extraction

Cited by 3 publications

References 14 publications

Change Detection for High-resolution Remote Sensing Images Based on a UNet-like Siamese-structured Transformer Network

Change Detection for High-resolution Remote Sensing Images Based on a UNet-like Siamese-structured Transformer Network

Automated Multi-class Crater Segmentation in Mars Orbital Images

MC-UNet: Martian Crater Segmentation at Semantic and Instance Levels Using U-Net-Based Convolutional Neural Network

Contact Info

Product

Resources

About