Research on vibration quality recognition method based on multi-modal sensing in complex scenes

Key Points

• The aim is to develop a multi-modal recognition method to enhance vibration quality assessment in concrete casting.
• Integrated sensors including posture sensors, laser sensors, GNSS, and industrial cameras in the vibration platform.
• Implemented a MetaData-EfficientNet-B4 deep learning model for feature extraction and data fusion.
• Developed a user-friendly recognition system based on a B/S architecture for supporting on-site construction and acceptance.
• Recognition accuracy improved from 95.8% to 96.8% in daylight conditions; from 90.7% to 96.2% at night; and from 91.7% to 95.8% in foggy conditions.
• Multi-modal data fusion significantly enhanced model performance in complex scenarios.
• The MetaData fusion method outperformed other fusion methods in TOPSIS evaluation.