In order to study stick-slip vibration of water-lubricated rubber stern tube bearing， firstly， stick-slip vibration images are collected by machine vision technology. Secondly， images are analyzed by the methods of persistent homology based machine learning and simplicial complex， the corresponding barcodes are obtained by calculating the homology of the vibration images' simple complex. Then， the topological characteristics of the vibration images are obtained based on the barcode images. Finally， the improved support vector machine learning is used to study the topological features， the classification and identification of the stick-slip vibration of water-lubricated rubber stern tube bearing are completed. The results have shown that the length of the longest Betti barcode is closely related to vibration， which can effectively warn the beep， establish an intelligent description of the beep process， and provide a new idea for stick-slip vibration of the stern bearing.