Nonlinear characteristics based on ultrasonic guided waves are more sensitive to bolt loosening and have become a research hotspot in recent years. However, due to the dispersion and multi-mode characteristics of ultrasonic guided waves, their propagation mechanism in bolt-connected structures is complicated, making it difficult to establish a direct correlation between nonlinear characteristics and bolt loosening. In recent years, machine learning methods have been widely applied to structural health monitoring. This paper builds and validates an intelligent identification method for bolt joint loosening based on nonlinear guided waves for accurate identification of bolt loosening. First, vibration-acoustic modulation (VAM) is used in the bolt looseness monitoring experiment to extract vibration-acoustic modulation signals, enabling accurate extraction of nonlinear features such as modulation sidebands and second harmonics. Subsequently, using the acoustic modulation signals extracted from the vibration-acoustic modulation experiment, a deep convolutional network is established to achieve bolt looseness identification based on deep neural networks and vibration-acoustic modulation. Using this method, end-to-end detection based on vibration-acoustic modulation can be achieved, enabling automatic extraction of nonlinear features. The effectiveness of this method has been verified through experiments, and the results show that this method achieves high accuracy in bolt looseness identification.