Underwater image classification using deep convolutional neural networks and data augmentation
Yifeng Xu, Yang Zhang, Huigang Wang, Xing Liu · 2017
The classification accuracy of underwater image, which have the special image characteristic, is lower than the corresponding result of images in the air. A study was carried out to underwater image classification with deep convolutional neural networks and the classification ability was improved with two data augmentation methods. The experiments showed that the submarine image classification with the ILSVRC Championship GoogLenet model was still relatively low confidence. The classification probability can be improved by two augmentation methods. The first was optical transformation of raw data such as scale and aspect ratio augmentation and Color augmentation. The second was increasing the virtue data generated by Generative Adversarial Nets. The results of the study validated the effectiveness of two data augmentation methods. Especially, the generative adversarial nets approach gave a new path to increasing the train data.