Research on Side-scan Sonar Image Target Classification Method Based on Transfer Learning

摘要

In this paper, we propose a method which combine transfer learning method and deep learning method for side-scan sonar image classification task. The application of deep learning method can effectively improve the accuracy of classification and use transfer learning method to overcome the problem that the deep neural networks cannot be applied due to the small number of training samples. We fine-tune a pre-trained convolutional neural network (CNN) primarily trained for common image classification tasks where sufficient training data exists, make it specifically optimized for a side-scan sonar image classification task, we also present the pre-processing method for the source domain training samples which can influence the transfer efficiency. Experiments show that this method can effectively prevent the overfitting problem of training the deep neural network under small sample number conditions, and at the same time guarantees the high classification accuracy.