Fish recognition using convolutional neural network

Studying fish recognition has important realistic and theoretical significance to aquaculture and marine biology. Fish recognition is challenging problem because of distortion, overlap and occlusion of digital images. Previous researchers have done a lot of work on fish recognition, but the classification accuracy may be not high enough. Classification and recognition methods based on convolutional neural network (CNN) develop fast in recent years because of its higher accuracy and the support of GPU. In this paper, we design several architectures for convolutional neural network for the fish recognition. After performing a series of experiments, we can get the CNN architecture which has best performance and robustness.