A Transformer-Based Approach for Efficient Geometric Feature Extraction from Vector Shape Data DOI Creative Commons
Longfei Cui, Xinyu Niu, Haizhong Qian

et al.

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(5), P. 2383 - 2383

Published: Feb. 23, 2025

The extraction of shape features from vector elements is essential in cartography and geographic information science, supporting a range intelligent processing tasks. Traditional methods rely on different machine learning algorithms tailored to specific types line polygon elements, limiting their general applicability. This study introduces novel approach called “Pre-Trained Shape Feature Representations Transformers (PSRT)”, which utilizes transformer encoders designed with three self-supervised pre-training tasks: coordinate masking prediction, offset correction, sequence rearrangement. enables the applicable both generating high-dimensional embedded feature vectors. These vectors facilitate downstream tasks like classification, pattern recognition, cartographic generalization. Our experimental results show that PSRT can extract effectively without needing labeled samples adaptable various features. Compared pre-training, enhances training efficiency by over five times improves accuracy 5–10% such as element matching classification. innovative offers more unified, efficient solution for data across applications.

Language: Английский

Development of an Autonomous Chess Robot System Using Computer Vision and Deep Learning DOI Creative Commons

Truong Duc Phuc,

Bui Cao Son

Results in Engineering, Journal Year: 2025, Volume and Issue: unknown, P. 104091 - 104091

Published: Jan. 1, 2025

Language: Английский

Citations

1
Heavy Equipment Detection on Construction Sites Using You Only Look Once (YOLO-Version 10) with Transformer Architectures DOI Creative Commons

Ikchul Eum,

Jae-Jun Kim,

Seunghyeon Wang

et al.

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(5), P. 2320 - 2320

Published: Feb. 21, 2025

Monitoring heavy equipment in real time is crucial for ensuring safety and operational efficiency at construction sites, yet achieving both high detection accuracy fast inference remains challenging under diverse environmental conditions. Although previous studies have attempted to improve speed, their findings often lack generalizability, partly due inconsistent datasets the need more advanced techniques. In response, this study proposes an enhanced object method that integrates transformer-based backbone networks into You Only Look Once (YOLO-version 10) framework. Evaluations conducted on a large-scale dataset of construction-site images demonstrate notable improvements detecting varying sizes. Comparisons with other detectors confirm proposed model not only achieves higher but also maintains competitive processing making it suitable real-time deployment. Additionally, made available broader experimentation development. These underscore method’s potential strengthen on-site by providing reliable efficient complex work environments, while acknowledging areas further refinement.

Language: Английский

Citations

1
A text dataset of fire door defects for pre-delivery inspections of apartments during the construction stage DOI Creative Commons
Seunghyeon Wang, Sungkon Moon,

Ikchul Eum

et al.

Data in Brief, Journal Year: 2025, Volume and Issue: unknown, P. 111536 - 111536

Published: April 1, 2025

Defect classification from text descriptions written by inspectors during the construction stage can be highly beneficial, offering advantages such as cost savings and improved reputation of apartment complexes allowing early identification resolution issues. Combining automated methods with textual data facilitate rapid diagnosis faults. To develop methods, this research constructed a dataset real-world collected three complexes. This study classifies fire door defects into eight categories: frame gap, closer adjustment defect, contamination, dent, scratch, sealing components, mechanical operation others. The level detail in ensures comprehensive understanding main contributions to field are twofold. First, it represents unique based on defect descriptions, which is currently non-existent domain. Second, dataset's expert labeling adds significant value ensuring accurate fault classification. We hope will encourage development robust techniques suitable for applications providing reliable benchmark.

Language: Английский

Citations

0
Mobile robot for leaf disease detection and precise spraying: Convolutional neural networks integration and path planning DOI Creative Commons

Youssef Bouhaja,

Hatim Bamoumen,

Israe Derdak

et al.

Scientific African, Journal Year: 2025, Volume and Issue: unknown, P. e02717 - e02717

Published: April 1, 2025

Language: Английский

Citations

0
Deuterium-deuterium fusion charged particle detection using CR-39 and Deep Learning Model DOI Creative Commons
Yuxing Wang, Allan Xi Chen,

Matthew Salazar

et al.

Radiation Measurements, Journal Year: 2025, Volume and Issue: unknown, P. 107444 - 107444

Published: May 1, 2025

Citations

0
A Transformer-Based Approach for Efficient Geometric Feature Extraction from Vector Shape Data DOI Creative Commons
Longfei Cui, Xinyu Niu, Haizhong Qian

et al.

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(5), P. 2383 - 2383

Published: Feb. 23, 2025

The extraction of shape features from vector elements is essential in cartography and geographic information science, supporting a range intelligent processing tasks. Traditional methods rely on different machine learning algorithms tailored to specific types line polygon elements, limiting their general applicability. This study introduces novel approach called “Pre-Trained Shape Feature Representations Transformers (PSRT)”, which utilizes transformer encoders designed with three self-supervised pre-training tasks: coordinate masking prediction, offset correction, sequence rearrangement. enables the applicable both generating high-dimensional embedded feature vectors. These vectors facilitate downstream tasks like classification, pattern recognition, cartographic generalization. Our experimental results show that PSRT can extract effectively without needing labeled samples adaptable various features. Compared pre-training, enhances training efficiency by over five times improves accuracy 5–10% such as element matching classification. innovative offers more unified, efficient solution for data across applications.

Language: Английский

Citations

0