Sharpened Cosine Similarity U-Net for Deforestation Mapping DOI
Ali Jamali, Swalpa Kumar Roy, Avik Bhattacharya

и другие.

Опубликована: Дек. 10, 2023

Forests cover about 30% of the Earth's surface, having a significant global impact on climate and atmosphere. A change (i.e., forest loss) in world's forests has been brought by factors, such as rising population, increased urbanization, environmental pollution due to economic activities. Consequently, loss mapping monitoring are vital. Convolutional Neural Networks (CNNs) among most utilized segmentation algorithms for deforestation mapping. However, CNNs may be more prone model variance, over-sensitivity, lack generalizability. Thus, new concepts, Cosine Similarity can investigated an alternative approach current extensively CNNs. this study, we develop propose SCS-UNet precise utilizing satellite imagery Sentinel-2 South America. The results illustrated that proposed algorithm exhibited least training time complexity compared other implemented models, UNet, Attention R2UNet, ResUNet, Swin UNet+++, TransUNet, TransUNet++, while resulting comparable statistical U-Net model.

Язык: Английский

DA-TransUNet: integrating spatial and channel dual attention with transformer U-net for medical image segmentation DOI Creative Commons
Guanqun Sun,

Yizhi Pan,

Weikun Kong

и другие.

Frontiers in Bioengineering and Biotechnology, Год журнала: 2024, Номер 12

Опубликована: Май 16, 2024

Accurate medical image segmentation is critical for disease quantification and treatment evaluation. While traditional U-Net architectures their transformer-integrated variants excel in automated tasks. Existing models also struggle with parameter efficiency computational complexity, often due to the extensive use of Transformers. However, they lack ability harness image’s intrinsic position channel features. Research employing Dual Attention mechanisms have not been specifically optimized high-detail demands images. To address these issues, this study proposes a novel deep framework, called DA-TransUNet, aiming integrate Transformer dual attention block (DA-Block) into U-shaped architecture. Also, DA-TransUNet tailored requirements images, optimizes intermittent channels (DA) employs DA each skip-connection effectively filter out irrelevant information. This integration significantly enhances model’s capability extract features, thereby improving performance segmentation. validated tasks, consistently outperforming state-of-the-art techniques across 5 datasets. In summary, has made significant strides segmentation, offering new insights existing techniques. It strengthens model from perspective advancing development high-precision diagnosis. The codes parameters our will be publicly available at https://github.com/SUN-1024/DA-TransUnet .

Язык: Английский

Процитировано

67
Residual wave vision U-Net for flood mapping using dual polarization Sentinel-1 SAR imagery DOI Creative Commons
Ali Jamali, Swalpa Kumar Roy, Leila Hashemi-Beni

и другие.

International Journal of Applied Earth Observation and Geoinformation, Год журнала: 2024, Номер 127, С. 103662 - 103662

Опубликована: Янв. 21, 2024

The increasing severity, duration, and frequency of destructive floods can be attributed to shifts in climate, infrastructure, land use, population demographics. Obtaining precise timely data about the extent floodwaters is crucial for effective emergency preparedness mitigation efforts. Deep convolutional neural networks (CNNs) have shown astonishing effectiveness various remote sensing applications, including flood mapping. One key limitations CNNs that they only predict whether a desired feature will appear an image, not where it recognized. To address this limitation, incorporation self-attention mechanisms deployed vision transformers (ViTs) particularly effective. However, modules ViTs are complex computationally expensive, require wealth ground attain their full capability image classification/segmentation. Thus, paper, we develop Residual Wave Vision U-Net (WVResU-Net), deep learning segmentation architecture utilizes advanced Multi-Layer Perceptrons (MLPs) ResU-Net accurate reliable mapping using Sentinel-1 SAR's dual polarization data. Results showed significant superiority developed WVResU-Net algorithms over several well-known CNN ViT models, Swin U-Net, U-Net+++, Attention R2U-Net, ResU-Net, TransU-Net TransU-Net++. For example, accuracy TransU-Net++, SwinU-Net, TransU-Net, was significantly improved by approximately 5, 12, 13, 16, 19, 23 percentage points, respectively terms recall obtained with value 69.67%. code made publicly available at https://github.com/aj1365/RWVUNet.

Язык: Английский

Процитировано

21
Deep object segmentation and classification networks for building damage detection using the xBD dataset DOI Creative Commons
Zongze Zhao, Fenglei Wang, Shiyu Chen

и другие.

International Journal of Digital Earth, Год журнала: 2024, Номер 17(1)

Опубликована: Янв. 8, 2024

Deep learning has been extensively utilized in the assessment of building damage after disasters. However, field segmentation faces challenges, such as misjudged regions, high network complexity, and long running times. Hence, this paper proposes a two-stage called Efficient Channel Attention Depthwise Separable Convolutional Neural Network (ECADS-CNN). It aims to quickly detect types disaster buildings. object deep classification networks were integrated into unified detection network. In study, efficient channel attention (ECA) module was used enhance performance semantic segmentation, depthwise separable (DS) added dimension upscaling process. Finally, untrained dataset images test robustness proposed model by comparing evaluation results each disaster. The experiments involve testing total five common models, indicate that ECADS-CNN improves speed 7.4% overall F1 score 5.2% compared with baseline model. comprehensive is better than mainstream models.

Язык: Английский

Процитировано

7
Enhanced-HisSegNet: Improved SAR Image Flood Segmentation with Learnable Histogram Layers and Active Contour Model DOI
Maryam Asadi, Soroush Sarabi, Marjan Kordani

и другие.

IEEE Geoscience and Remote Sensing Letters, Год журнала: 2025, Номер 22, С. 1 - 5

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

1
Category attention guided network for semantic segmentation of Fine-Resolution remote sensing images DOI Creative Commons
Shunli Wang, Qingwu Hu, Shaohua Wang

и другие.

International Journal of Applied Earth Observation and Geoinformation, Год журнала: 2024, Номер 127, С. 103661 - 103661

Опубликована: Янв. 20, 2024

Язык: Английский

Процитировано

5
CTBANet: Convolution transformers and bidirectional attention for medical image segmentation DOI Creative Commons

Sha Luo,

Li Pan,

Yuanming Jian

и другие.

Alexandria Engineering Journal, Год журнала: 2024, Номер 88, С. 133 - 143

Опубликована: Янв. 16, 2024

In the last few years, Transformer has revolutionized area of medical image segmentation. Several similar studies have used UNet architecture to combine convolutional neural networks with transformers. However, these approaches fail account for speed at which segmentation occurs and ability extract features within Transformer. They consider fact that changing shape feature maps in a subtle way can be rapid extraction local global information. To solve above problems, CTBANet (Convolutional Bidirectional Attention Based Medical Image Segmentation) is proposed, two prominent components, CTblock Combined module) BAblock (Bidirectional Attentionblock). integrates strengths CNNs Transformers, enabling it spatial details data. order improve accuracy model, multi-scale pyramid pooling embedded into PAM, named APAM (Asymmetric PAM), strip convolution CAM, ACAM CAM). critical issue field, experimental results benchmarks show our model obviously more accurate faster than other methods segmenting images.

Язык: Английский

Процитировано

3
Vision transformers and multi-sensor Earth observation DOI
Shakti Sharma,

Lalita Chaudhari

Elsevier eBooks, Год журнала: 2025, Номер unknown, С. 201 - 210

Опубликована: Янв. 1, 2025

Язык: Английский

Процитировано

0
An Agricultural Leaf Disease Segmentation Model Applying Multi-Scale Coordinate Attention Mechanism DOI

Renyuan Gu,

Liqun Liu

Applied Soft Computing, Год журнала: 2025, Номер unknown, С. 112904 - 112904

Опубликована: Фев. 1, 2025

Язык: Английский

Процитировано

0
A deep learning-based micro-CT image analysis pipeline for nondestructive quantification of the maize kernel internal structure DOI Creative Commons
Juan Wang, Si Yang,

Chuanyu Wang

и другие.

Plant Phenomics, Год журнала: 2025, Номер unknown, С. 100022 - 100022

Опубликована: Фев. 1, 2025

Язык: Английский

Процитировано

0
S2-IFNet: A spatial-semantic information fusion network integrated with boundary feature enhancement for forest land extraction from Sentinel-2 data DOI

Junyang Xie,

Mengyao Zhang,

Hao Wu

и другие.

International Journal of Applied Earth Observation and Geoinformation, Год журнала: 2025, Номер 139, С. 104505 - 104505

Опубликована: Апрель 4, 2025

Язык: Английский

Процитировано

0