The intelligent recognition and monitoring of sand particles in annular multiphase flow are of paramount importance for the safe production of high-yield gas wells. In this study, an experiment based on a uniaxial vibration method was initially designed to collect collision response signals between sand particles and the pipe wall. Utilizing wavelet packet analysis, the identification and classification of sand-carrying signals in the liquid film and gas core regions were first achieved. The results indicate that the excitation frequency range for sand-carrying signals impacting the pipe wall in the liquid film region was 19.2–38.4 kHz, while in the gas core region, it was 38.4–51.2 kHz. Finally, convolutional neural network (CNN) models, support vector machine (SVM) models, and CNN-SVM models were constructed to characterize and identify sand particles in annular multiphase flow. The results show that the CNN-SVM model improved the accuracy of sand-carrying data recognition by 2.0% compared to CNN and by 5.6% compared to SVM for gas core region data, and by 1.8% compared to CNN and by 8.6% compared to SVM for liquid film region data. Consequently, this research offers a high-accuracy recognition and classification method for sand particles in the gas core and liquid film regions of annular multiphase flow.