Cavitation in centrifugal pumps causes damage, leading to poor performance, noise, vibration, and material erosion. Early prediction is important in avoiding pump breakdowns. In this study, we have used the Continuous Wavelet Transform (CWT) technique for cavitation detection at various levels based on casing vibration signals of the pump. Experiments have been done on a centrifugal pump having 10 blades (2850 rpm) in four states: normal, incipient, developed, and severe cavitation. The vibration data has been collected using an accelerometer and analyzed using the CWT method with the db9 mother wavelet function. It was observed that cavitation continuously generates scales 2 (725 Hz) and 5-10 (145-290 Hz). The CWT coefficient value for these scales rises gradually with increasing cavitation severity. Even in the presence of heavy white Gaussian noise (Signal-to-Noise Ratio (SNR) 0 dB), these scale bands can be easily identified. The proposed technique is efficient and noise resilient.