MBGPIN: Multi-Branch Generative Prior Integration Network for Super-Resolution Satellite Imagery

Remote Sensing, Journal Year: 2025, Volume and Issue: 17(5), P. 805 - 805

Published: Feb. 25, 2025

Achieving super-resolution with satellite images is a critical task for enhancing the utility of remote sensing data across various applications, including urban planning, disaster management, and environmental monitoring. Traditional interpolation methods often fail to recover fine details, while deep-learning-based approaches, convolutional neural networks (CNNs) generative adversarial (GANs), have significantly advanced performance. Recent studies explored large-scale models, such as Transformer-based architectures diffusion demonstrating improved texture realism generalization diverse datasets. However, these frequently high computational costs require extensive datasets training, making real-world deployment challenging. We propose multi-branch prior integration network (MBGPIN) address limitations. This novel framework integrates multiscale feature extraction, hybrid attention mechanisms, priors derived from pretrained VQGAN models. The dual-pathway architecture MBGPIN includes extraction pathway spatial features external guidance, dynamically fused using an adaptive fusion (AGPF) module. Extensive experiments on benchmark UC Merced, NWPU-RESISC45, RSSCN7 demonstrate that achieves superior performance compared state-of-the-art methods, delivers higher peak signal-to-noise ratio (PSNR) structural similarity index measure (SSIM) scores preserving high-frequency details complex textures. model also significant efficiency, reduced floating point operations (FLOPs) faster inference times, it scalable applications.

Language: Английский

---

A 3D medical image segmentation network based on gated attention blocks and dual-scale cross-attention mechanism

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: Feb. 20, 2025

In the field of multi-organ 3D medical image segmentation, Convolutional Neural Networks (CNNs) are limited to extracting local feature information, while Transformer-based architectures suffer from high computational complexity and inadequate extraction spatial channel layer information. Moreover, large number varying sizes organs be segmented result in suboptimal model robustness segmentation outcomes. To address these challenges, this paper introduces a novel network architecture, DS-UNETR++, specifically designed for segmentation. The proposed features dual-branch encoding mechanism that categorizes images into coarse-grained fine-grained types before processing them through blocks. Each block comprises downsampling Gated Shared Weighted Pairwise Attention (G-SWPA) submodule, which dynamically adjusts influence attention on extraction. Additionally, Dual-Scale Cross-Attention Module (G-DSCAM) is incorporated at bottleneck stage. This module employs dimensionality reduction techniques cross-coarse-grained features, using gating balance ratio two thereby achieving effective multi-scale fusion. Finally, comprehensive evaluations were conducted four public datasets. Experimental results demonstrate DS-UNETR++ achieves good performance, highlighting effectiveness significance method offering new insights various organ tasks.

Language: Английский

---

Lightweight image super-resolution for IoT devices using deep residual feature distillation network

Knowledge-Based Systems, Journal Year: 2023, Volume and Issue: 285, P. 111343 - 111343

Published: Dec. 28, 2023

The 5th industrial revolution is characterized by an extensive interconnection of embedded devices, which offer a range services, including the monitoring their environments. Images captured from remote cameras require enhancements for effective analysis. Despite recent progress in single-image super-resolution techniques yielding impressive results through deep convolutional neural networks, complexity these advanced models renders them impractical use on miniaturized Internet Things (IoT) primarily due to limited computational capabilities and memory constraints. Furthermore, rapid evolution IoT devices necessitates efficient image techniques, while existing methods, based are too resource-intensive this gap highlights need more suitable solution. In study, we introduce lightweight, model specifically designed devices. This incorporates novel residual feature distillation block (DRFDB), leverages depthwise-separable convolution (DCB) extraction. focus reducing demands without compromising quality. proposed DCB extracts coarse features given input as calculation units, using two operations, depthwise pointwise convolutions. These operations able significantly reduce number parameters floating-point maintaining PSNR value higher than 90% threshold. We modify multi-kernel (MKDCB) fine-tune model. experiments, conduct various standard datasets such DIV2K, Set5, Set14, Urban100, Manga109 demonstrating that our outperforms methods terms both quality efficiency. shows improved performance metrics like PSNR, requiring fewer less usage, making it highly applications. study presents breakthrough balancing high-quality reconstruction with resources

Language: Английский

---

Enhancing the Super-Resolution of Medical Images: Introducing the Deep Residual Feature Distillation Channel Attention Network for Optimized Performance and Efficiency

Bioengineering, Journal Year: 2023, Volume and Issue: 10(11), P. 1332 - 1332

Published: Nov. 19, 2023

In the advancement of medical image super-resolution (SR), Deep Residual Feature Distillation Channel Attention Network (DRFDCAN) marks a significant step forward. This work presents DRFDCAN, model that innovates traditional SR approaches by introducing channel attention block is tailored for high-frequency features-crucial nuanced details in diagnostics-while streamlining network structure enhanced computational efficiency. DRFDCAN's architecture adopts residual-within-residual design to facilitate faster inference and reduce memory demands without compromising integrity reconstruction. strategy, combined with an innovative feature extraction method emphasizes utility initial layer features, allows improved clarity particularly effective optimizing peak signal-to-noise ratio (PSNR). The proposed redefines efficiency models, outperforming established frameworks like RFDN improving compactness accelerating inference. meticulous crafting extractor effectively captures edge texture information exemplifies model's capacity render detailed images, necessary accurate analysis. implications this study are two-fold: it viable solution deploying technology real-time applications, sets precedent future models address delicate balance between high-fidelity paramount applications where images can significantly influence diagnostic outcomes. DRFDCAN thus stands as transformative contribution field super-resolution.

Language: Английский

---

Enhancing Medical Image Denoising with Innovative Teacher–Student Model-Based Approaches for Precision Diagnostics

Sensors, Journal Year: 2023, Volume and Issue: 23(23), P. 9502 - 9502

Published: Nov. 29, 2023

The realm of medical imaging is a critical frontier in precision diagnostics, where the clarity image paramount. Despite advancements technology, noise remains pervasive challenge that can obscure crucial details and impede accurate diagnoses. Addressing this, we introduce novel teacher–student network model leverages potency our bespoke NoiseContextNet Block to discern mitigate with unprecedented precision. This innovation coupled an iterative pruning technique aimed at refining for heightened computational efficiency without compromising fidelity denoising. We substantiate superiority effectiveness approach through comprehensive suite experiments, showcasing significant qualitative enhancements across multitude modalities. visual results from vast array tests firmly establish method’s dominance producing clearer, more reliable images diagnostic purposes, thereby setting new benchmark

Language: Английский

---

AI-driven 3D bioprinting for regenerative medicine: From bench to bedside

Bioactive Materials, Journal Year: 2024, Volume and Issue: 45, P. 201 - 230

Published: Nov. 23, 2024

Language: Английский

---

Lightweight Deep Learning Framework for Accurate Detection of Sports-Related Bone Fractures

Diagnostics, Journal Year: 2025, Volume and Issue: 15(3), P. 271 - 271

Published: Jan. 23, 2025

Background/Objectives: Sports-related bone fractures are a common challenge in sports medicine, requiring accurate and timely diagnosis to prevent long-term complications enable effective treatment. Conventional diagnostic methods often rely on manual interpretation, which is prone errors inefficiencies, particularly for subtle localized fractures. This study aims develop lightweight efficient deep learning-based framework improve the accuracy computational efficiency of fracture detection, tailored needs medicine. Methods: We proposed novel detection based DenseNet121 architecture, incorporating modifications initial convolutional block final layers optimized feature extraction. Additionally, Canny edge detector was integrated enhance model ability detect structural discontinuities. A custom-curated dataset radiographic images focused sports-related used, with preprocessing techniques such as contrast enhancement, normalization, data augmentation applied ensure robust performance. The evaluated against state-of-the-art using metrics accuracy, recall, precision, complexity. Results: achieved 90.3%, surpassing benchmarks like ResNet-50, VGG-16, EfficientNet-B0. It demonstrated superior sensitivity (recall: 0.89) specificity (precision: 0.875) while maintaining lowest complexity (FLOPs: 0.54 G, Params: 14.78 M). These results highlight its suitability real-time clinical deployment. Conclusions: offers scalable, accurate, solution addressing critical challenges By enabling rapid reliable diagnostics, it has potential workflows outcomes athletes. Future work will focus expanding applications other imaging modalities types.

Language: Английский

---

Modified U-Net with attention gate for enhanced automated brain tumor segmentation

Neural Computing and Applications, Journal Year: 2025, Volume and Issue: unknown

Published: Jan. 2, 2025

Language: Английский

---

MSFM-UNET: enhancing medical image segmentation with multi-scale and multi-view frequency fusion

Pattern Analysis and Applications, Journal Year: 2025, Volume and Issue: 28(1)

Published: Jan. 7, 2025

Language: Английский

---

A multi-scale enhanced large-kernel attention transformer network for lightweight image super-resolution

Signal Image and Video Processing, Journal Year: 2025, Volume and Issue: 19(3)

Published: Jan. 17, 2025

Language: Английский

---