A review of deep learning methods for semantic segmentation of remote sensing imagery

Yuan, Xiaohui; Shi, Jinhong; Gu, Lichuan

doi:10.1016/j.eswa.2020.114417

Cited by 403 publications

(211 citation statements)

References 80 publications

Supporting

Mentioning

209

Contrasting

Unclassified

Order By: Relevance

Section: Introductionmentioning

confidence: 99%

“…Fully Convolutional Neural Networks are a particular type of Convolutional Neural Networks (CNN) that allow obtaining a class prediction for each pixel in an image [25,26,[35][36][37][38][39]. These algorithms are capable of identifying patterns at different scales to produce classifications [32,37]. Thus, CNN architectures are nowadays among the most widely applied algorithms for classification tasks [20], and particularly, the U-net is one of the most popular algorithms in LULC studies [31,32,40] due to its capability of summarizing patterns in both the spectral and spatial domain (for additional details see Section 2.3).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Land Use Land Cover Classification with U-Net: Advantages of Combining Sentinel-1 and Sentinel-2 Imagery

et al. 2021

View full text Add to dashboard Cite

The U-net is nowadays among the most popular deep learning algorithms for land use/land cover (LULC) mapping; nevertheless, it has rarely been used with synthetic aperture radar (SAR) and multispectral (MS) imagery. On the other hand, the discrimination between plantations and forests in LULC maps has been emphasized, especially for tropical areas, due to their differences in biodiversity and ecosystem services provision. In this study, we trained a U-net using different imagery inputs from Sentinel-1 and Sentinel-2 satellites, MS, SAR and a combination of both (MS + SAR); while a random forests algorithm (RF) with the MS + SAR input was also trained to evaluate the difference in algorithm selection. The classification system included ten classes, including old-growth and secondary forests, as well as old-growth and young plantations. The most accurate results were obtained with the MS + SAR U-net, where the highest overall accuracy (0.76) and average F1-score (0.58) were achieved. Although MS + SAR and MS U-nets gave similar results for almost all of the classes, for old-growth plantations and secondary forest, the addition of the SAR band caused an F1-score increment of 0.08–0.11 (0.62 vs. 0.54 and 0.45 vs. 0.34, respectively). Consecutively, in comparison with the MS + SAR RF, the MS + SAR U-net obtained higher F1-scores for almost all the classes. Our results show that using the U-net with a combined input of SAR and MS images enabled a higher F1-score and accuracy for a detailed LULC map, in comparison with other evaluated methods.

show abstract

Section: Introductionmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

Land Use Land Cover Classification with U-Net: Advantages of Combining Sentinel-1 and Sentinel-2 Imagery

et al. 2021

View full text Add to dashboard Cite

show abstract

“…Cheng et al [13] provided a comprehensive review on the image scene classification task and described the details of autoencoder-based, convolutional neural network (CNN)-based, and generative adversarial network (GAN)-based image scene classification methods. Yuan et al [14] conducted a complete review for the semantic segmentation of remotely sensed images. In particular, they explained the CNN architectures used in semantic segmentation such as U-Net, SegNet, and DeepLab.…”

Section: Introductionmentioning

confidence: 99%

Automatic Road Extraction from Historical Maps Using Deep Learning Techniques: A Regional Case Study of Turkey in a German World War II Map

Ekim

Sertel

Kabadayı

2021

IJGI

View full text Add to dashboard Cite

Scanned historical maps are available from different sources in various scales and contents. Automatic geographical feature extraction from these historical maps is an essential task to derive valuable spatial information on the characteristics and distribution of transportation infrastructures and settlements and to conduct quantitative and geometrical analysis. In this research, we used the Deutsche Heereskarte 1:200,000 Türkei (DHK 200 Turkey) maps as the base geoinformation source to construct the past transportation networks using the deep learning approach. Five different road types were digitized and labeled to be used as inputs for the proposed deep learning-based segmentation approach. We adapted U-Net++ and ResneXt50_32 × 4d architectures to produce multi-class segmentation masks and perform feature extraction to determine various road types accurately. We achieved remarkable results, with 98.73% overall accuracy, 41.99% intersection of union, and 46.61% F1 score values. The proposed method can be implemented in DHK maps of different countries to automatically extract different road types and used for transfer learning of different historical maps.

show abstract

“…In recent years, deep learning (DL) has been widely used in the field of computer vision. Image segmentation using DL to predict classes on a pixel level, which is regarded as semantic segmentation [14]. Hoeser et al [15] reviewed 261 studies in the Earth observation field which used Convolutional Neural Networks (CNNs) for image segmentation, among which 62% were encoder-decoder models and 54% were related to the U-Net design.…”

Section: Introductionmentioning

confidence: 99%

Remote Sensing Image Change Detection Using Superpixel Cosegmentation

2021

View full text Add to dashboard Cite

The application of cosegmentation in remote sensing image change detection can effectively overcome the salt and pepper phenomenon and generate multitemporal changing objects with consistent boundaries. Cosegmentation considers the image information, such as spectrum and texture, and mines the spatial neighborhood information between pixels. However, each pixel in the minimum cut/maximum flow algorithm for cosegmentation change detection is regarded as a node in the network flow diagram. This condition leads to a direct correlation between computation times and the number of nodes and edges in the diagram. It requires a large amount of computation and consumes excessive time for change detection of large areas. A superpixel segmentation method is combined into cosegmentation to solve this shortcoming. Simple linear iterative clustering is adopted to group pixels by using the similarity of features among pixels. Two-phase superpixels are overlaid to form the multitemporal consistent superpixel segmentation. Each superpixel block is regarded as a node for cosegmentation change detection, so as to reduce the number of nodes in the network flow diagram constructed by minimum cut/maximum flow. In this study, the Chinese GF-1 and Landsat satellite images are taken as examples, the overall accuracy of the change detection results is above 0.80, and the calculation time is only one-fifth of the original.

show abstract

A review of deep learning methods for semantic segmentation of remote sensing imagery

Cited by 403 publications

References 80 publications

Land Use Land Cover Classification with U-Net: Advantages of Combining Sentinel-1 and Sentinel-2 Imagery

Land Use Land Cover Classification with U-Net: Advantages of Combining Sentinel-1 and Sentinel-2 Imagery

Automatic Road Extraction from Historical Maps Using Deep Learning Techniques: A Regional Case Study of Turkey in a German World War II Map

Remote Sensing Image Change Detection Using Superpixel Cosegmentation

Contact Info

Product

Resources

About