Objective To explore the detection efficacy of models constructed by combining time-frequency image features of the fragrant pears in acoustic and vibration with machine learning algorithms.Methods Acoustic and vibration signals are collected from fragrant pears exhibiting varying degrees of internal browning using a custom-built non-destructive detection device. Time-frequency images are transformed using the short-time Fourier transform (STFT), continuous wavelet transform (CWT), and the S-transform. Then, texture feature parameters in 15 time-frequency domains are extracted, respectively, from 3 types of time-frequency images using the Gray-Gradient Co-occurrence Matrix (GGCM). Additionally, correlation analysis is performed between these feature parameters and the fragrant pear browning levels. Subsequently, 5 types of machine learning models [random forest, na?ve Bayes, k-nearest neighbors (KNN), extreme learning machine (ELM), and support vector machine (SVM)] are built using the feature datasets of 3 distinct time-frequency domains for internal browning detection.Results The SVM model achieves the best classification performance for feature parameters of the S-transform time-frequency images, exceeding 90% in accuracy, precision, recall, and weighted score. Based on the model confusion matrix, the constructed model achieves identification accuracies of 93.1% for healthy pears, 94.9% for mildly browned pears, and 92.1% for severely browned pears.Conclusion Time-frequency image information of acoustic vibration signals can be applied to the detection of early browning inside fragrant pears.