Active recognition of underwater targets using GAF-D3Net deep learning network

A method for active recognition of underwater targets based on Gramian angular field (GAF) and convolutional neural network (CNN) is proposed. GAF is used to encode the target echo signal into a two-dimensional image, and a lightweight convolutional neural network GAF-D3Net are bulit by dilated convolutions to achieve feature extraction and classification recognition of the target. Experiment results show that the classification accuracy of the proposed method is significantly improved to 99.65% compared to traditional image feature-based classification methods. In the generalization test, comparing the classical CNN migration learning method using sonar images, the area under curve (AUC) of the proposed method reaches 89%, verifying that the proposed method has better generalization performance as well as anti-interference capability. It provides a reliable method for achieving active recognition of underwater targets.