A platform for research: civil engineering, architecture and urbanism
A platform for research: civil engineering, architecture and urbanism
Dual attention-based deep learning for construction equipment activity recognition considering transition activities and imbalanced dataset
https://orcid.org/0000-0003-4085-6041
https://orcid.org/0000-0002-7838-2120
Abstract With the advancement of sensor and data acquisition technology, the development of multi-sensor integrated construction equipment has become increasingly prominent. Activity recognition for construction equipment is confronted with challenges such as complex input from multiple sensors, difficulties in accurately identifying transition activities, and imbalanced datasets across various activities. To address these challenges, a dual attention-based deep learning approach is proposed for more accurate activity recognition. Sensor data features are extracted using temporal convolutional networks and dual attention mechanisms. Employing the constrained dynamic time warping minority oversampling method, a larger number of smaller samples are synthesized, thereby balancing the training dataset. Subsequently, gated recurrent units are utilized to recognize the basic and transition activities of construction equipment. The proposed method has been implemented on excavators and one wheel loader, yielding promising results. In the future, a more lightweight network will be designed to accommodate the deployment on real construction equipment.
Highlights Multiple ontological signals from construction equipment are used to recognize its activity. A dual attention-based TCN network is used to extract features from Multi-sensor signals. Solving the problem that the switching phase cannot be recognized accurately. CDTW-based minority class oversampling addresses imbalance activity samples. The proposed method is verified on both excavator and wheel loader.
Dual attention-based deep learning for construction equipment activity recognition considering transition activities and imbalanced dataset
https://orcid.org/0000-0003-4085-6041
https://orcid.org/0000-0002-7838-2120
Abstract With the advancement of sensor and data acquisition technology, the development of multi-sensor integrated construction equipment has become increasingly prominent. Activity recognition for construction equipment is confronted with challenges such as complex input from multiple sensors, difficulties in accurately identifying transition activities, and imbalanced datasets across various activities. To address these challenges, a dual attention-based deep learning approach is proposed for more accurate activity recognition. Sensor data features are extracted using temporal convolutional networks and dual attention mechanisms. Employing the constrained dynamic time warping minority oversampling method, a larger number of smaller samples are synthesized, thereby balancing the training dataset. Subsequently, gated recurrent units are utilized to recognize the basic and transition activities of construction equipment. The proposed method has been implemented on excavators and one wheel loader, yielding promising results. In the future, a more lightweight network will be designed to accommodate the deployment on real construction equipment.
Highlights Multiple ontological signals from construction equipment are used to recognize its activity. A dual attention-based TCN network is used to extract features from Multi-sensor signals. Solving the problem that the switching phase cannot be recognized accurately. CDTW-based minority class oversampling addresses imbalance activity samples. The proposed method is verified on both excavator and wheel loader.
Dual attention-based deep learning for construction equipment activity recognition considering transition activities and imbalanced dataset
https://orcid.org/0000-0003-4085-6041
https://orcid.org/0000-0002-7838-2120
Abstract With the advancement of sensor and data acquisition technology, the development of multi-sensor integrated construction equipment has become increasingly prominent. Activity recognition for construction equipment is confronted with challenges such as complex input from multiple sensors, difficulties in accurately identifying transition activities, and imbalanced datasets across various activities. To address these challenges, a dual attention-based deep learning approach is proposed for more accurate activity recognition. Sensor data features are extracted using temporal convolutional networks and dual attention mechanisms. Employing the constrained dynamic time warping minority oversampling method, a larger number of smaller samples are synthesized, thereby balancing the training dataset. Subsequently, gated recurrent units are utilized to recognize the basic and transition activities of construction equipment. The proposed method has been implemented on excavators and one wheel loader, yielding promising results. In the future, a more lightweight network will be designed to accommodate the deployment on real construction equipment.
Highlights Multiple ontological signals from construction equipment are used to recognize its activity. A dual attention-based TCN network is used to extract features from Multi-sensor signals. Solving the problem that the switching phase cannot be recognized accurately. CDTW-based minority class oversampling addresses imbalance activity samples. The proposed method is verified on both excavator and wheel loader.
Dual attention-based deep learning for construction equipment activity recognition considering transition activities and imbalanced dataset
Shen, Yuying (author) / Wang, Jixin (author) / Feng, Chenlong (author) / Wang, Qi (author)
2024-01-26
Article (Journal)
Electronic Resource
English
Plant Disease Identification under Imbalanced Dataset using Hybrid Deep Learning Method
| Springer Verlag | 2025
Plant Disease Identification under Imbalanced Dataset using Hybrid Deep Learning Method
| Springer Verlag | 2025
Mapping Forests Using an Imbalanced Dataset
| Springer Verlag | 2022