The prevailing techniques for seismic phase analysis encompass waveform classification, seismic attribute feature mapping, and seismic geomorphological delineation. The waveform classification method is a well-established and extensively utilized technique for lithological, sand body, and oil and gas reservoir prediction. Nevertheless, the conventional approach relies on equal-length time window waveform similarity, which is only pertinent to the stable zone of formation thickness. As a consequence of changes in formation thickness, equal-length seismic waveforms cannot reflect the complete lithological information, or conversely, may lead to the phenomenon of ‘time-warp’, which in turn affects the accurate revelation of the relationship between reservoirs and waveforms. A seismic waveform classification method for unequally thick layers, intending proposes to reduce complexity and enhance classification efficacy. In comparison to the traditional classification method, this approach transfers unequal seismic signals from the time domain to the Hilbert domain with constant bandwidth, thereby ensuring the completeness of the waveform extraction and simplifying the traditional two-dimensional self-organized feature mapping network into a structure with fewer neurons and a one-dimensional output layer. This adaptation is better suited to the resolution of the seismic data and the need for classification efficiency. The enhanced network retains the capacity to modify the field and value of weight correction by with the responsiveness of the output neurons to the input neurons, thereby facilitating effective control of the network size, reducing the complexity of classification calculations, and enhancing classification efficacy. The practical results confirm the effectiveness of the method and significantly improve the accuracy of waveform classification.