Recognition of common fish images by deep convolutional neural networks

Abstract

A fast and accurate fish identification system is urgently needed for the development of fishery industry. In this paper, ten kinds of common fish are taken as the research object, and the existing achievements of deep Convolutional Neural Networks in the field of image classification are used to provide technical support for the establishment of intelligent and efficient fish recognition system in fishery production. The research mainly includes three parts : data collection, model building and optimization comparison. Firstly, fish images are collected through web crawler and field collection, and then the collected image format is normalized to establish a fish data set. data augmentation is used to expand the training dataset and reduce the dependence of the model on the original dataset. Secondly, AlexNet, VGGNet and ResNet convolutional neural networks are used for image classification tools, and the generalization ability of the model is improved by combining the transfer learning method. Finally, through the comparison of the training results of each model, the ResNet-50 network model with better performance based on transfer learning is found as the main model of fish recognition, and its recognition rate is 98.06 %.