Texture and Shape Information Fusion of Convolutional Neural Network for Plankton Image Classification

Plankton is the foundation of marine ecosystem and of great value in environmental protection and fishing industries. As plankton plays an essential role in maintaining ecosystem balance and economic development, more and more researches are focused on plankton analysis. Plankton classification is very crucial for plankton analysis and related ecological study, but it is extremely difficult because of the huge amount and tiny volume of plankton. In this paper, we propose an effective plankton feature extraction method followed by an end-to-end hybrid convolutional neural network for plankton classification. The database we used is from a large scale database named WHOI-Plankton which contains about 3.6 million in situ images labeled into 103 classes. The experimental results show that our method can significantly improve plankton classification accuracy compared with other commonly used deep neural networks.