Small-scale Data Underwater Acoustic Target Recognition with Deep Forest Model

Underwater acoustic target recognition is an issue of great interest, and its key lies in effective feature extraction. Nowadays, due to the rapid development of underwater acoustic signal processing technology and machine learning, some progress has been made in the field of underwater acoustic target recognition. However, traditional machine learning methods utilize shallow features, and the recognition ability needs to be further improved. Although neural network-based deep learning methods can extract deep features, they are prone to over-fitting and other undesirable phenomena in underwater small-scale data scenarios. This means that we need to find a method of underwater acoustic target recognition that can extract deep features, and it should be suitable for small-scale data scenarios. In this research, a method of underwater acoustic target recognition based on the deep forest model is come up with to meet the above requirements. This method adopts MFCC features and the deep forest model as the input feature vectors and classifier, respectively. Experimental results on the ShipsEar database show that the proposed method achieves satisfactory performance and has a promising application in the field of small-scale data underwater acoustic target recognition.