Target Recognition Method of Sonar Image Based on Deep Learning

The target detection and recognition technology of sonar images plays an important role in the field of marine environment monitoring. The traditional method mainly uses CNN to study sonar image recognition. However, the lack of data volume and the limitations of the CNN network itself reduce the accuracy of sonar image classification. To address the above issues, this paper first uses the DCGAN network for data augmentation. Data expansion generates fake images based on the collection of public data sets from the network and completes the expansion of the data set. Secondly, this paper uses ResNet network and DenseNet network to replace the traditional CNN network and uses focal loss to replace the traditional cross-entropy loss function. The model has a good classification effect for the data set in this paper. The classification accuracy of the improved ResNet network is 77%, and the classification accuracy of the improved DenseNet network is 84.1%.