Explore the vast range of titles available.

Sedimentary facies modeling is a critical approach for understanding geological phenomena, yet the strong heterogeneity of reservoir systems poses a serious challenge for their refined characterization. In this study, we innovatively propose an interpretable attention-guided generative adversarial network framework with dual-domain learning, which achieves precise sedimentary facies modeling under the constraints of well facies and soft probability data. Specifically, we first effectively extract and preserve prior information of sedimentary facies models from both spatial and frequency domain perspectives. Then, during simulation, to enhance the capability of the network model for finely characterizing complex heterogeneous models, cross-spatial attention mechanisms are designed to effectively capture short-range and long-range dependencies between multi-scale pattern features. Additionally, through systematic feature map visualization analysis, we elucidate the processes of conditional fitting and complex sedimentary facies model reconstruction, intuitively demonstrating the functional mechanisms of each module. Finally, systematic experiments are conducted on multiple datasets to validate the effectiveness of the proposed method. The results demonstrate that the generated sedimentary facies models exhibit high consistency with training datasets in terms of visual realism and statistical indicators. Quantitative comparisons reveal remarkable performance of the method, achieving low Wasserstein distance (0.09), Kernel Inception Distance (0.0017) and Kernel Maximum Mean Discrepancy (0.21). These findings further confirm the high realism of the generated realizations regarding pattern features. This study offers a reliable and practical method for geological reservoir modeling, thereby advancing quantitative, precise geological research with broad application prospects.

Due to the attenuation and scattering of underwater light and water conditions, underwater images usually suffer from degradation problems such as color distortion and detail blurring, which seriously affect underwater engineering and research tasks. Previous underwater image enhancement (UIE) methods primarily focus on spatial domain enhancement, neglecting the crucial role of frequency-domain information. This bias leads to distortion of low-frequency background colors and loss of high-frequency texture details. In contrast, frequency-domain analysis effectively separates and enhances components of different frequencies, fundamentally improving the image’s authenticity and detail representation. To compensate for these shortcomings, we propose a frequency-domain attention-guided generative adversarial network (GAN). Our approach contains the following key components: first, we design a frequency-domain attention-guided module (FDAG) for guiding the network to learn key frequency-domain information in the image. Second, to address the problem of high-frequency information loss during feature propagation from the encoder to the decoder, we propose a Frequency Restoration Block (FRB). This unit contains a set of filters that can co-emphasize the medium and high frequencies of the input signal. Finally, we propose the global spatial self-calibrating convolution block (GSSC), which effectively combines global information, local features, and spatial attention to further refine important details in the image. Experiments on benchmark synthetic and real underwater image datasets demonstrate our method. It achieves an average improvement of 2% in PSNR and 2.77% in SSIM compared to the best-performing baseline. Additionally, it exhibits superior visual quality in terms of color restoration and detail preservation. Extensive ablation studies and comparative analyses further validate the effectiveness and robustness of our approach.

Objective To develop high-quality synthetic CT (sCT) generation method from low-dose cone-beam CT (CBCT) images by using attention-guided generative adversarial networks (AGGAN) and apply these images to dose calculations in radiotherapy. Methods The CBCT/planning CT images of 170 patients undergoing thoracic radiotherapy were used for training and testing. The CBCT images were scanned under a fast protocol with 50% less clinical projection frames compared with standard chest M20 protocol. Training with aligned paired images was performed using conditional adversarial networks (so-called pix2pix), and training with unpaired images was carried out with cycle-consistent adversarial networks (cycleGAN) and AGGAN, through which sCT images were generated. The image quality and Hounsfield unit (HU) value of the sCT images generated by the three neural networks were compared. The treatment plan was designed on CT and copied to sCT images to calculated dose distribution. Results The image quality of sCT images by all the three methods are significantly improved compared with original CBCT images. The AGGAN achieves the best image quality in the testing patients with the smallest mean absolute error (MAE, 43.5 ± 6.69), largest structural similarity (SSIM, 93.7 ± 3.88) and peak signal-to-noise ratio (PSNR, 29.5 ± 2.36). The sCT images generated by all the three methods showed superior dose calculation accuracy with higher gamma passing rates compared with original CBCT image. The AGGAN offered the highest gamma passing rates (91.4 ± 3.26) under the strictest criteria of 1 mm/1% compared with other methods. In the phantom study, the sCT images generated by AGGAN demonstrated the best image quality and the highest dose calculation accuracy. Conclusions High-quality sCT images were generated from low-dose thoracic CBCT images by using the proposed AGGAN through unpaired CBCT and CT images. The dose distribution could be calculated accurately based on sCT images in radiotherapy.

This paper proposes a single-image translation method based on attention guidance to solve the problem of poor image quality in current single-image translation. The model uses a multi-scale pyramid architecture. First, the input image is downsampled, and then the downsampled image is input into the attention-guided generator to complete the translation of an image from the source domain X to the target domain Y. We introduce an attention module and a Scale-Add (SA) module, which can stabilize the training process of GAN and effectively improve the image quality. The attention module can retain the contour and detail of the object. In addition, the Scale-Add (SA) module can adjust the style of the image and add some low-scale detail information. Through extensive experimental verification and comparison with several baseline methods on benchmark datasets, we verify the effectiveness of the proposed framework.