Research and Comparison of Ship Classification Algorithms Based on Variant CNNs

At present, the image classification algorithm based on deep learning in the field of computer vision has come to maturity. However, the papers related to the main classification are aimed at, for example, life scenes and traffic scenes. This paper specially produced image datasets for ships, and carried out various image preprocessing experiments and classification algorithms for specific ship scenes. Various models based on convolutional neural networks were used for training, and a number of performances were performed for comparison, including the accuracy of the training from scratch, the accuracy after the transfer learning, the class accuracy of each model. The experimental results show that the model using transfer learning is not only faster in training, but also has a significant improvement in accuracy. And the recognition difficulty of the model is different for different classes, which provides useful data for subsequent research on this particular scenario.