Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
Eine Plattform für die Wissenschaft: Bauingenieurwesen, Architektur und Urbanistik
A Deep Learning Framework for Construction Equipment Activity Analysis
Systematic evaluation of construction equipment activities is essential to efficient management of the fleet. Recent research has made significant progress in developing machine learning activity recognition frameworks. Deep learning in particular can circumvent the need for complex manually-designed feature extraction/selection procedures that contribute to lower accuracies in traditional and shallow models. The research presented in this paper develops and compares deep learning algorithms for construction equipment activity recognition in different levels of detail. Data are collected in a non-controlled environment from real-world activities. A convolutional neural network (CNN) called BaselineCNN and a hybrid network that contains both convolutional and recurrent long short-term memory (LSTM) layers called DeepConvLSTM are studied. In summary, DeepConvLSTM proved superior to BaselineCNN. In a six-class identification task, DeepConvLSTM achieved a validation accuracy of 77.1%. In the vibration-setting-only subproblem, DeepConvLSTM achieved a validation accuracy of 75.2%. In the direction-only subproblem, DeepConvLSTM achieved a validation accuracy of 96.2%.
A Deep Learning Framework for Construction Equipment Activity Analysis
Systematic evaluation of construction equipment activities is essential to efficient management of the fleet. Recent research has made significant progress in developing machine learning activity recognition frameworks. Deep learning in particular can circumvent the need for complex manually-designed feature extraction/selection procedures that contribute to lower accuracies in traditional and shallow models. The research presented in this paper develops and compares deep learning algorithms for construction equipment activity recognition in different levels of detail. Data are collected in a non-controlled environment from real-world activities. A convolutional neural network (CNN) called BaselineCNN and a hybrid network that contains both convolutional and recurrent long short-term memory (LSTM) layers called DeepConvLSTM are studied. In summary, DeepConvLSTM proved superior to BaselineCNN. In a six-class identification task, DeepConvLSTM achieved a validation accuracy of 77.1%. In the vibration-setting-only subproblem, DeepConvLSTM achieved a validation accuracy of 75.2%. In the direction-only subproblem, DeepConvLSTM achieved a validation accuracy of 96.2%.
A Deep Learning Framework for Construction Equipment Activity Analysis
Hernandez, Carlos (Autor:in) / Slaton, Trevor (Autor:in) / Balali, Vahid (Autor:in) / Akhavian, Reza (Autor:in)
ASCE International Conference on Computing in Civil Engineering 2019 ; 2019 ; Atlanta, Georgia
Computing in Civil Engineering 2019 ; 479-486
13.06.2019
Aufsatz (Konferenz)
Elektronische Ressource
Englisch
A Deep Learning Framework for Construction Equipment Activity Analysis
| British Library Conference Proceedings | 2019
Vision-Based Activity Analysis Framework Considering Interactive Operation of Construction Equipment
| British Library Conference Proceedings | 2017