The slope stability of muddy submarine channel is very important for the safe operation of the channel.Due to the uncertainty of channel slope instability and collapse，it is impossible to predict and evaluate the degree of instability，which seriously affects the safety of passing ships and management and maintenance of the channel.Based on the engineering experiment simulation，this paper obtains the full cycle voiceprint image of channel slope instability and collapse.By analyzing the energy change of voiceprint image，the instability process is divided into five stages：very stable，stable，unstable，extremely unstable and collapse.The voiceprint image features at each stage are extracted based on the Gabor transform，and the discrete cosine transform is used to reduce the dimensionality of the features.The voiceprint image features after the dimensionality reduction are input into Fisher function to build the instability degree evaluation and prediction model.The results show that the recognition accuracy reaches 90% when the features of voiceprint image samples to be evaluated are input into the prediction model.Therefore，it is feasible to build Fisher model based on Gabor characteristics to evaluate and predict the instability degree of muddy submarine channel slope，which is of great significance to the study of channel slope stability.