Recognition of Side-scan Sonar Shipwreck Image Using Convolutional Neural Network

In order to enable the AUV to quickly and automatically identify the image of the side-scan sonar shipwreck, improve the speed of search and rescue of wrecked ships and the accuracy of the verification of shipwreck targets in submarine obstacles. Aiming at the limitation that AUV side-scan sonar image data cannot be transmitted in real time, a method for recognizing shipwreck image of side-scan sonar image based on convolutional neural network is proposed. Sample data is expanded through standardization and normalization processing, data augmentation, etc.; according to the characteristics of the side-scan sonar image sample data set, referring to the classic convolutional neural network VGG-16 model, a simplified convolutional neural network model is constructed; by adding Dropout, the joint adaptability between neuron nodes of the model is weakened, overfitting is reduced, and generalization ability is improved. The experimental results show that the accuracy of the model to the side-scan sonar shipwreck image recognition can reach 90.58%.