Abstract: To solve the problem of underwater multi-target azimuth estimation, a method to estimate the target sound source azimuth using the convolutional neural network (CNN) model is proposed. In this method, unequal intensity sound source data are used for training and the focal loss function is taken as the training loss function. Through the feature extraction of the signals received by the array, the focal loss function is used to guide the convolutional neural network training, and finally the trained convolutional neural network model is used to estimate the target azimuth. By comparison with the training results of different model parameters, it is shown that the trained convolutional neural network model can correctly estimate the azimuth of weak targets under the condition of low SNR. And, the experimental results show that in contrast with the convolutional neural network model using the binary cross-entropy loss function, the method in this paper has a stronger ability to estimate weak target azimuth and improves the estimation accuracy.