We propose an improved wavelet clustering algorithm to address the issue of cluster overlap in the In-phase and Quadrature (IQ) domain caused by multi-tag collisions in Radio Frequency Identification (RFID) backscatter systems. This algorithm offers significant advantages in dealing with cluster overlap problems that are not effectively resolved by conventional clustering algorithms. In the improved wavelet clustering algorithm, we leverage the characteristics of multi-tag collision signals based on the original wavelet clustering algorithm. In general the transition point will be removed as noise points, however, it has a certain segmentation effect on the signal multi-tag collision signal. By utilizing the discrete transition points between combined state clusters, we successfully separate overlapped combined state clusters and accurately identify the number of correctly classified combined state clusters. Experimental results demonstrate that the improved wavelet clustering algorithm can effectively distinguish overlapped combined state clusters affected by noise and achieve accurate symbol clustering.