A Novel Semantic Segmentation Method for Remote Sensing Images Through Adaptive Scale-Based Convolution Neural Network DOI Creative Commons

Jing Zhang,

Bin Li, Jun Li

et al.

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 96074 - 96084

Published: Jan. 1, 2024

In semantic segmentation tasks for remote sensing images, effective feature extraction acts as the most important foundation. As a result, this paper proposes novel method images through adaptive scale-based convolution neural network. Firstly, it uses an encoder to extract features from and utilizes attention mechanisms control information flow in This is expected reduce impact between different scales. Then, updating scale weights established, scale-adaptive convolutional network constructed. The upsampling unit improved increase resolution original image level, order better identify smaller targets. For large-sized problems, pyramid-like processing used segment multiple scales, results are finally merged. Besides, we also make some experiments on ISPRS Potsdam dataset, UC Merced DeepGlobe performance evaluation. research shows that maximum pixel accuracy of proposed increased 86.18%, average value task up 63.72, fastest running speed 9.16FPS. other words, study has more accurate stability.

Language: Английский

GTC: GNN-Transformer co-contrastive learning for self-supervised heterogeneous graph representation DOI
Yundong Sun, Dongjie Zhu, Yansong Wang

et al.

Neural Networks, Journal Year: 2024, Volume and Issue: 181, P. 106645 - 106645

Published: Aug. 16, 2024

Language: Английский

Citations

6
Illustration image style transfer method design based on improved cyclic consistent adversarial network DOI Creative Commons
Xiaojun Wang, Jing Jiang

PLoS ONE, Journal Year: 2025, Volume and Issue: 20(1), P. e0313113 - e0313113

Published: Jan. 14, 2025

To improve the expressiveness and realism of illustration images, experiment innovatively combines attention mechanism with cycle consistency adversarial network proposes an efficient style transfer method for images. The model comprehensively utilizes image restoration capabilities network, introduces improved module, which can adaptively highlight key visual elements in illustration, thereby maintaining artistic integrity during process. Through a series quantitative qualitative experiments, high-quality is achieved, especially while retaining original features illustration. results show that when running on Monet2photo dataset, system iterates to 72 times, loss function value research approaches target 0.00. On Horse2zebra as sample size increases, has smallest FID value, 40.00 infinitely. With change peak signal-to-noise ratio, accuracy algorithm been greater than 95.00%. Practical application found color obtained by more gorgeous line are obvious. above all achieved satisfactory task terms retention details.

Language: Английский

Citations

0
A multi-objective dynamic detection model in autonomous driving based on an improved YOLOv8 DOI
Chaoran Li,

Yinghui Zhu,

Min Zheng

et al.

Alexandria Engineering Journal, Journal Year: 2025, Volume and Issue: 122, P. 453 - 464

Published: March 18, 2025

Language: Английский

Citations

0
Learning to Generate Urban Design Images From the Conditional Latent Diffusion Model DOI Creative Commons

Xiaotang Cui,

Xiao Feng,

Siwen Sun

et al.

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 89135 - 89143

Published: Jan. 1, 2024

With the rapid process of computer vision and deep learning, image synthesis models, such as latent diffusion have exhibited remarkable performances in producing high-quality realistic results. However, achieving precise layout control through adjusting text prompts solely proves to be challenging for model. Therefore, we organize conditional network instruct model towards generating satisfactory Besides, direct training from scratch or fine-tuning on new datasets is non-trivial due massive parameters. To tackle troublesome problem, implement low-rank adaptation strategy The decomposes 2-dimensional matrices into 1-dimensional vectors, which can decrease number parameters greatly accelerate synthesize images, collect urban design images pinterest generate homologous prompts. We intend make this dataset publicly available further research development field. Both qualitative quantitative evaluations demonstrate effectiveness capacity our framework.

Language: Английский

Citations

2
An Efficient Low Complex-Functional Link Artificial Neural Network-Based Framework for Uneven Light Image Thresholding DOI Creative Commons
Tapaswini Pattnaik, P. Kanungo, Prabodh Kumar Sahoo

et al.

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 118315 - 118338

Published: Jan. 1, 2024

Language: Английский

Citations

2
Deep learning-based aquatic plant recognition technique and natural ecological aesthetics conservation DOI
Y. F. Bai,

Xiaomei Bai

Crop Protection, Journal Year: 2024, Volume and Issue: 184, P. 106765 - 106765

Published: May 31, 2024

Language: Английский

Citations

1
BSM-YOLO: A Dynamic Sparse Attention-Based Approach for Mousehole Detection DOI Creative Commons
Tianshuo Xie, Xiaoling Luo, Xin Pan

et al.

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 78787 - 78798

Published: Jan. 1, 2024

In recent years, the proliferation of mousehole in grasslands has exacerbated desertification and compromised grassland productivity, posing potential threats to human safety. Consequently, identification forecasting mouse-hole dynamics for effective infestation control have emerged as pressing concerns. Manual detection is labor-intensive time-consuming, hindering comprehensive spatial understanding. Moreover, prevailing models lack robust feature extraction small targets like mousehole, resulting suboptimal recognition capabilities diminished accuracy. Addressing these challenges, we propose an enhanced one-stage model BSM-YOLO based onYOLOv5 architecture. Firstly, integrates a BiFormer module leveraging Bi-Level Routing Attention capture both global local features within images. Subsequently, incorporation Shuffle mechanisms enhances learning dependencies intricate relationships. Lastly, adoption MPDIoU loss function accurately delineates bounding box characteristics, mitigating redundant generation expediting convergence. our experimental framework, curated dataset comprising 2397 images train model. Results indicate that achieves average accuracy 94.5%, representing 5.4% enhancement over baseline YOLOv5s Additionally, demonstrates 8.7 f/s improvement speed. Furthermore, ablation experiments confirm efficacy each refinement incorporated into

Language: Английский

Citations

1
A Novel Semantic Segmentation Method for Remote Sensing Images Through Adaptive Scale-Based Convolution Neural Network DOI Creative Commons

Jing Zhang,

Bin Li, Jun Li

et al.

IEEE Access, Journal Year: 2024, Volume and Issue: 12, P. 96074 - 96084

Published: Jan. 1, 2024

In semantic segmentation tasks for remote sensing images, effective feature extraction acts as the most important foundation. As a result, this paper proposes novel method images through adaptive scale-based convolution neural network. Firstly, it uses an encoder to extract features from and utilizes attention mechanisms control information flow in This is expected reduce impact between different scales. Then, updating scale weights established, scale-adaptive convolutional network constructed. The upsampling unit improved increase resolution original image level, order better identify smaller targets. For large-sized problems, pyramid-like processing used segment multiple scales, results are finally merged. Besides, we also make some experiments on ISPRS Potsdam dataset, UC Merced DeepGlobe performance evaluation. research shows that maximum pixel accuracy of proposed increased 86.18%, average value task up 63.72, fastest running speed 9.16FPS. other words, study has more accurate stability.

Language: Английский

Citations

0