Empirical mode decomposition has been used to decompose the pressure fluctuation signals in the spouted bed into several intrinsic mode functions, and these modes were transformed from the time domain into the frequency domain by Hilbert transformation. According to the characteristic parameters extracted from these modes, flow regimes were recognized by RBF neural network, and parameters in RBF neural network were optimized by adaptive genetic algorithm. The recognition accuracy of packed bed, spouted bed, bubbly fluidized bed and slugging bed can reach 90%, 85%, 85%, 95%, respectively.