Prestressed concrete beams are widely used in bridge engineering due to their long span, lightweight, and good integrity. However, the grouting quality in the bellows of the beams will directly affect the safety and durability of the entire structure. Therefore, it is fairly important to detect and evaluate the quality of the grouting in bellows. The current mainstream detection methods are mainly horizontal detection methods that can only identify local defects and that have low detection efficiency. This paper proposes a longitudinal transmission detection method based on the BP neural network algorithm. This method could quickly identify the grouting compactness of an entire pipeline. The main process was as follows: First, a signal at one end of the beam was transmitted and the signals at both ends of the beam were collected. Second, the features of the processed signal were extracted. Through the qualitative analysis of wave propagation theory, the extracted characteristic parameters including the wave speed, energy attenuation, wavelet high-frequency energy attenuation, and frequency shift value were all closely related to the compactness of the grouting, which confirmed their rationality as the network input vector. Third, advantage was taken of the close relationship between the characteristic parameters and the grouting saturation to establish a BP neural network regression model. Moreover, the network parameters were adjusted to make the model prediction effect the best. After adjusting the network parameters, the network performance was significantly changed and improved, in that the training set error dropped by 0.0081 and the test set error dropped by 0.0011. The end of the paper describes how a full-scale experimental model was designed to verify the reliability of the method. The result showed that the error between the predicted grouting compactness based on the proposed method and the actual compactness in the test was within 3%, which met the actual engineering requirements and verified the reliability of the proposed method. The significance of this method could significantly improve detection efficiency and prepare for defect location and quantification.