Abstract:The rational polyculture and close cultivation of multispecies freshwater fish have great practical significance in aquaculture. Aiming to identify the mixed proportions of freshwater fish, bream fish and crucian carp were taken as the research object. The passive acoustic signals of different proportions of freshwater fish were collected by hydrophone. The butter function was used for signal preprocessing. Then shorttime average energy, shorttime average zerocrossing rate, four layer wavelet packet decomposition frequency band energy, average Mel cepstrum coefficient, main peak frequency and principal peaks based on power spectrum were extracted to construct eigenvectors. The support vector machine model based on principal component analysis was used to realize the mixed proportion identification. The significant differences among the acoustic signals of freshwater fish with different mixed proportions were analyzed, and the influences of the number of principal component on the recognition rate of the model were studied. The results showed that the average Mel cepstrum coefficient had the most significant effect on the mixed proportions recognition of freshwater fish, and the effect of proportional recognition was the best by selecting the first 19 principal components. The average accuracy rate was 96.43% and Kappa coefficient was 0.96.