Article
Engineering, Environmental
Demetris Demetriou, Pavlos Mavromatidis, Ponsian M. Robert, Harris Papadopoulos, Michael F. Petrou, Demetris Nicolaides
Summary: This study evaluates the performance of deep-learning models in real-time localization and classification of Construction and Demolition Waste (CDW). Different detector architectures (SSD, YOLO, Faster-RCNN) along with various backbone feature extractors (ResNet, MobileNetV2, efficientDet) were considered. The latest version of YOLO series, YoloV7, achieved the highest accuracy (mAP50:95% & AP;70%) with the fastest inference speed (<30 ms) and was capable of dealing with heavily stacked and adhered CDW samples. It was observed that apart from YoloV7, Faster-RCNN models showed the least mAP fluctuations over the testing datasets and remained the most robust.
Article
Green & Sustainable Science & Technology
Qinghui Zhou, Haoshi Liu, Yuhang Qiu, Wuchao Zheng
Summary: This study proposes an object detection method based on an improved YOLOv5 model to enhance the accuracy of sorting construction waste. A construction waste image sample set is established by collecting on-site construction waste images and preprocessed using the random brightness method. The improved YOLOv5 model, which includes the convolutional block attention module (CBAM), simplified SPPF (SimSPPF), and multi-scale detection, is trained, validated, and tested using the established construction waste image dataset and compared with other conventional models. The results show that the improved YOLOv5 model achieves a mean average precision (mAP) of 0.9480 on the test dataset, outperforming other conventional models in object detection and verifying the accuracy and availability of the proposed method.
Article
Robotics
Ryan Luke Johns, Martin Wermelinger, Ruben Mascaro, Dominic Jud, Ilmar Hurkxkens, Lauren Vasey, Margarita Chli, Fabio Gramazio, Matthias Kohler, Marco Hutter
Summary: This article presents a robotic construction pipeline that uses local materials to build stone walls and landscapes, reducing carbon emissions. The system learns from real and simulated data to detect and segment stone instances, allowing for precise positioning using a geometric planning algorithm.
Article
Environmental Sciences
Kunsen Lin, Tao Zhou, Xiaofeng Gao, Zongshen Li, Huabo Duan, Huanyu Wu, Guanyou Lu, Youcai Zhao
Summary: This study develops an efficient method to sort Construction and Demolition (C&D) waste using deep learning and knowledge transfer approach. The CVGGNet-16 model shows excellent performance in various types of C&D waste, paving the way for better waste management.
JOURNAL OF ENVIRONMENTAL MANAGEMENT
(2022)
Article
Environmental Sciences
Kunsen Lin, Tao Zhou, Xiaofeng Gao, Zongshen Li, Huabo Duan, Huanyu Wu, Guanyou Lu, Youcai Zhao
Summary: The study introduces the use of CVGGNet models for efficient sorting of Construction and Demolition (C&D) waste, incorporating knowledge transfer and data augmentation techniques for optimal performance. Among the CVGGNet models, CVGGNet-16 exhibits outstanding performance in classifying various types of C&D waste, achieving higher accuracy and overall performance compared to other models.
JOURNAL OF ENVIRONMENTAL MANAGEMENT
(2022)
Article
Chemistry, Analytical
Sulabh Kumra, Shirin Joshi, Ferat Sahin
Summary: This study proposes a dual-module robotic system capable of generating and performing antipodal robotic grasps at real-time speeds. The model achieved state-of-the-art accuracy of 98.8%, 95.1%, and 97.4% on three standard datasets. Experimental results show significant improvement in grasp detection compared to prior work.
Article
Automation & Control Systems
Qian-Qian Hong, Liang Yang, Bi Zeng
Summary: This paper presents a novel grasp generative residual attention network (RANET) for generating antipodal robotic grasp from multi-modal images with the pixel-wise method. The proposed network integrates a coordinate attention mechanism and a symmetrical skip connection, emphasizing meaningful information and preserving fine-grained details of the feature map. Evaluation on public datasets demonstrates high accuracy and real-time performance.
INTERNATIONAL JOURNAL OF CONTROL AUTOMATION AND SYSTEMS
(2022)
Article
Robotics
Boyan Wei, Xianfeng Ye, Chengjiang Long, Zhenjun Du, Bangyu Li, Baocai Yin, Xin Yang
Summary: This article proposes a Discriminative Active Learning (DAL) framework for robotic grasping algorithms. DAL utilizes a shared encoder to extract latent features from labeled and unlabeled data, and a discriminator to estimate the informativeness of unlabeled data samples for annotation. Experimental results demonstrate that DAL performs well with limited labeled data and is stable to noisy annotation.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2023)
Article
Engineering, Environmental
Jiantao Li, Huaiying Fang, Lulu Fan, Jianhong Yang, Tianchen Ji, Qiang Chen
Summary: This study establishes an RGB-D detection platform, introduces three fusion models, and uses an instance segmentation network for waste classification. The experimental results show that these fusion models outperform the traditional RGB model in terms of classification accuracy and meet the requirements of real-time detection.
Article
Remote Sensing
Xue Zhao, Yang Yang, Fuzhou Duan, Miao Zhang, Guofu Jiang, Xing Yan, Shisong Cao, Wenji Zhao
Summary: In this study, a method for construction and demolition waste (CDW) identification based on change detection and deep learning was proposed. The method involved initial sample preparation using multi-spectral images and difference images, and spatial analysis methods were used to extract specific forms of CDW. The results demonstrated that change detection improves the accuracy of deep learning models, and this method achieves high recognition accuracy.
INTERNATIONAL JOURNAL OF REMOTE SENSING
(2022)
Article
Robotics
Gang Yan, Alexander Schmitz, Satoshi Funabashi, Sophon Somlor, Tito Pradhono Tomo, Shigeki Sugano
Summary: This research explores the possibility of constructing a multi-phase, multi-output framework for robotic tactile manipulation and proposes using different neural network architectures, including attention mechanisms, to improve prediction and detection accuracy. Experimental results demonstrate that the proposed model exhibits more reliable and flexible performance in robot experiments.
IEEE ROBOTICS AND AUTOMATION LETTERS
(2022)
Article
Construction & Building Technology
Peter Davis, Fayeem Aziz, Mohammad Tanvi Newaz, Willy Sher, Laura Simon
Summary: The management of Construction and Demolition Waste (C&DW) is complicated and costly. The research developed a deep convolutional neural network to accurately classify C&DW in construction sites, achieving 94% accuracy which is crucial for cost reduction and waste diversion from landfill.
AUTOMATION IN CONSTRUCTION
(2021)
Article
Thermodynamics
Jingkuang Liu, Yuxuan Li, Zhenshuang Wang
Summary: Energy savings and emission reductions in the construction industry are vital for China's carbon peaking and carbon neutrality goals. This study uses a system dynamics model to explore the relationship among factors affecting carbon emissions in construction and demolition waste (C&DW) management and investigates the effects of different policies. The findings suggest that subsidies and carbon taxes can significantly influence contractors' choices of sorting modes, leading to reduced carbon emissions in C&DW treatment.
Article
Green & Sustainable Science & Technology
Kunsen Lin, Youcai Zhao, Tingting Zhou, Xiaofeng Gao, Chunbo Zhang, Beijia Huang, Qinyan Shi
Summary: This study aims to develop an efficient method for finely classifying construction and demolition waste using convolutional neural networks and transfer learning. The results show that knowledge transfer can significantly reduce training time and improve model performance. C&DWNet-18 is more suitable for the classification of construction and demolition waste in terms of training time, accuracy, precision, and F1 score.
ENVIRONMENT DEVELOPMENT AND SUSTAINABILITY
(2023)
Article
Automation & Control Systems
Eduardo Godinho Ribeiro, Raul de Queiroz Mendes, Valdir Grassi Jr
Summary: This study explores the potential of using supervised deep learning for robotic grasping in unstructured and dynamic environments by training a Convolutional Neural Network (CNN) with the Cornell Grasping Dataset (CGD). The developed controller achieves millimeter accuracy in the final position considering a target object seen for the first time, showing promising results for autonomous robotic manipulation.
ROBOTICS AND AUTONOMOUS SYSTEMS
(2021)
Article
Multidisciplinary Sciences
Wen Xiao, Jianhong Yang, Huaiying Fang, Jiangteng Zhuang, Yuedong Ku
Article
Materials Science, Characterization & Testing
Jiangteng Zhuang, Jianhong Yang, Huaiying Fang, Wen Xiao, Yuedong Ku
JOURNAL OF TESTING AND EVALUATION
(2019)
Article
Engineering, Environmental
Wen Xiao, Jianhong Yang, Huaiying Fang, Jiangteng Zhuang, Yuedong Ku
Article
Green & Sustainable Science & Technology
Wen Xiao, Jianhong Yang, Huaiying Fang, Jiangteng Zhuang, Yuedong Ku, Xiaojun Zhang
CLEAN TECHNOLOGIES AND ENVIRONMENTAL POLICY
(2020)
Article
Automation & Control Systems
Yue-Dong Ku, Jian-Hong Yang, Huai-Ying Fang, Wen Xiao, Jiang-Teng Zhuang
INTERNATIONAL JOURNAL OF AUTOMATION AND COMPUTING
(2020)
