Temporal-Wavelet CNN for RFID-Based Multi-Target Vibration Sensing

Traditional single-channel vibration sensing struggles with multi-target signal aliasing and inadequate feature extraction. This paper proposes a dual-channel fusion method integrating RFID-based contactless sensing and deep learning. The architecture includes a temporal CNN capturing temporal dynamics, and a Morlet wavelet-based branch generating time-frequency maps for 2D CNN frequency analysis. Features are fused via concatenation and fully connected layers. Experiments show 95% accuracy on fine-grained 600–840 RPM vibration classification and 100% accuracy in multi-band tasks (low, medium, high, ultra-high frequency). The method replaces costly sensors with RFID tags, enabling scalable applications in machinery fault warning and structural health monitoring.