This paper proposes a novel underwater acoustic target recognition method based on MFCC and RACNN. The method aims to accurately recognize the type of ship-radiated noise in underwater environments. The Residual Attentional Convolutional Neural Network (RACNN) is designed to extract internal features of Mel Frequency Cepstral Coefficients (MFCC) of underwater ship-radiated noise. The RACNN integrates residual and attention mechanisms to enhance learning capability, fault tolerance, and emphasis on vital information. The network is capable of suppressing environmental noises, extracting deep abstract features, and improving sensitivity to critical information. The RACNN model is trained and evaluated using the ShipsEar dataset, which contains 91 samples of underwater target noise. The model achieves an overall accuracy of 99.34%, surpassing conventional recognition methods and other deep learning models. The RACNN outperforms other networks such as Vgg16 and ResNet34 in terms of accuracy and parameter scale. The model demonstrates superior performance in underwater target recognition, with high accuracy and efficiency. The RACNN is a robust and efficient solution for underwater target recognition, showing potential for real-world applications where efficiency and accuracy are crucial. The study highlights the effectiveness of the RACNN in learning distinctive features from underwater acoustic signals, and its potential for further development and application in various maritime environments.This paper proposes a novel underwater acoustic target recognition method based on MFCC and RACNN. The method aims to accurately recognize the type of ship-radiated noise in underwater environments. The Residual Attentional Convolutional Neural Network (RACNN) is designed to extract internal features of Mel Frequency Cepstral Coefficients (MFCC) of underwater ship-radiated noise. The RACNN integrates residual and attention mechanisms to enhance learning capability, fault tolerance, and emphasis on vital information. The network is capable of suppressing environmental noises, extracting deep abstract features, and improving sensitivity to critical information. The RACNN model is trained and evaluated using the ShipsEar dataset, which contains 91 samples of underwater target noise. The model achieves an overall accuracy of 99.34%, surpassing conventional recognition methods and other deep learning models. The RACNN outperforms other networks such as Vgg16 and ResNet34 in terms of accuracy and parameter scale. The model demonstrates superior performance in underwater target recognition, with high accuracy and efficiency. The RACNN is a robust and efficient solution for underwater target recognition, showing potential for real-world applications where efficiency and accuracy are crucial. The study highlights the effectiveness of the RACNN in learning distinctive features from underwater acoustic signals, and its potential for further development and application in various maritime environments.