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

Accurate inspection of rebars in Reinforced Concrete (RC) structures is essential and requires careful counting. Deep learning algorithms utilizing object detection can facilitate this process through Unmanned Aerial Vehicle (UAV) imagery. However, their effectiveness depends on the availability of large, diverse, and well-labelled datasets. This article details the creation of a dataset specifically for counting rebars using deep learning-based object detection methods. The dataset comprises 874 raw images, divided into three subsets: 524 images for training (60 %), 175 for validation (20 %), and 175 for testing (20 %). To enhance the training data, we applied eight augmentation techniques—brightness, contrast, perspective, rotation, scale, shearing, translation, and blurring—exclusively to the training subset. This resulted in nine distinct datasets: one for each augmentation technique and one combining all techniques in augmentation sets. Expert annotators labelled the dataset in VOC XML format. While this research focuses on rebar counting, the raw dataset can be adapted for other tasks, such as estimating rebar diameter or classifying rebar shapes, by providing the necessary annotations.

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