Abstract: Once the bolts are loosened, the structural stability of the tower will be affected. To address the problem that traditional methods struggle to identify weak sound signals, a voiceprint recognition method for bolt looseness based on the formant spectrogram of sound signals is proposed. The sound signal of bolt looseness in the excited state is collected using an excitation device, and noise reduction is then processed. The formant spectrogram is extracted from the sound signal and used as the input for an Dropout improved CNN model. Through model computation, the probability that the input data belongs to each category is obtained, and the category with the highest probability is selected as the recognition result. The results indicate that after model identification, towers numbered 5, 9, 13, 17, and 19 show no signs of bolt looseness. In contrast, other transmission towers exhibit a relatively higher probability of bolt looseness, suggesting the presence of such issues. This demonstrates that the improved CNN model with Dropout can accurately identify bolt looseness and shows significant performance advantages in the task of bolt looseness voiceprint recognition, indicating that this method has superior recognition performance.