Hyperosteogeny and Osteoporosis are two common bone diseases that have a high incidence in the middle-aged and elderly groups. Mild symptoms may only affect the daily life of patients, while severe ones are life-threatening. At present, detection methods based on X-ray film and ultrasound are generally applied. However, the former exist errors introduced by manual reading and a certain radiation hazard, the diagnostic results of the latter are not that satisfying as well. Photoacoustic effect combines the advantages of optics for sensitive light absorption contrast and acoustics for lower acoustic scattering in soft tissue. As a non-ionizing and non-invasive technique, its application in biomedicine is also emerging. In this paper, a classification model built on Convolutional Neural Network (CNN) was proposed to achieve automated diagnosis of hyperosteogeny, osteoporosis and normal bone. Time-domain photoacoustic signals generated by different bone types are set as the inputs of the CNN while the output results indicate the corresponding categories of the samples. The analysis results of ex vivo data demonstrated that the established model could accurately accomplish the research of classification. Thus, the proposed method has certain auxiliary value for improving the efficiency, accuracy and objectivity of clinical diagnosis of the three bone types.