«上一篇/Previous Article|本期目录/Table of Contents|下一篇/Next Article»

《计算机技术与发展》[ISSN:1006-6977/CN:61-1281/TN]

卷:

33

期数:

2023年09期

页码:

155-160

栏目:

人工智能

出版日期:

2023-09-10

文章信息/Info

Title:

A Fish Classification System Based on Federal Convolution Neural Network

文章编号:

1673-629X(2023)09-0155-06

作者:

杨晓雨; 周彩凤*

华北电力大学 数理学院 河北省物理与能源技术重点实验室,河北 保定 071003

Author(s):

YANG Xiao-yu; ZHOU Cai-feng*

Hebei Key Laboratory of Physics and Energy Technology,School of Mathematics and Physics,North China Electric Power University,Baoding 071003,China

关键词:

鱼类分类; 联邦学习; 分布式计算; 卷积神经网络; 模糊化处理

Keywords:

fish classification; federated learning; distributed computing; convolutional neural network; fuzzy processing

分类号:

TP391

DOI:

10. 3969 / j. issn. 1673-629X. 2023. 09. 023

摘要:

由于鱼类数据的多样性以及应用的广泛性,为了进一步提高鱼类检测的效率,以及在处理鱼类图片时提取到更高维的特征来提高鱼类检测的准确率,将卷积神经网络与联邦学习相结合,将鱼类图片数据按照非独立同分布的形式分发给用户。 用户在本地训练模型,并将训练好的模型参数上传到云端,云端将完成模型参数的聚合与更新,并将更新好的参数返回到用户的终端,各个用户开始下一轮训练。 以此过程来训练网络,并模拟联邦学习的过程。 最后,用联邦卷积神经网络、联邦学习以及卷积神经网络分别对野生鱼类数据集上鱼类图片进行图像检测与识别,并将结果做对比。 结果表明,联邦卷积神经网络模型最终的分类准确率为 33. 3% ,传统的联邦学习的准确率为 26. 67%， Resnet50 的准确率为 87. 97% ,可以看出联邦卷积神经网络的分类准确率远高于传统的联邦学习。 并且联邦卷积神经网络模型在训练轮数较少的情况下就可以得到较好的实验结果。 联邦学习作为分布式计算的重要组成部分,它提供的快速模糊化处理以及数据独立的特性,为鱼类分类的效率和数据保护提供了有力保障。 卷积神经网络也提高了联邦学习的学习效率。 这使得提出的联邦卷积神经网络分类系统相比于传统的联邦学习在分类效率以及分类准确率方面有了较大程度的提高。

Abstract:

Due to the diversity of fish data and the wide range of applications, in order to further improve the efficiency of fishclassification and extract higher dimensional features when processing fish images to improve the accuracy of fish classification,the convolutional neural network is combined with federated learning to distribute fish image data to users in the form of nonindependentidentically distributed. The user trains the model locally and uploads the trained model parameters to the cloud. The cloud will completethe aggregation and update of model parameters,and return the updated parameters to the user’s terminal,and each user will start the nextround of training. This process is used to train the network and simulate the process of federated learning. Finally,we use federal convolutional neural network,federal learning,and convolutional neural network to classify fish images on the wild fish dataset,and comparethe results. It is showed that the final classification accuracy of the federal convolutional neural network model is 33. 3% ,the accuracy ofthe traditional federal learning is 26. 67% ,the accuracy of Resnet50 is 87. 97% . It can be seen that the classification accuracy of thefederal convolutional neural network is far higher than that of the traditional federal learning and convolutional neural network. And thefederal convolutional neural network model can get better experimental results when the number of training rounds is less. As animportant part of distributed computing,federal learning provides fast fuzzy processing and data independence,which provides a strongguarantee for the efficiency of fish classification and data protection. Convolutional neural network also improves the efficiency offederated learning. This makes the proposed federated convolutional neural network classification system has a greater degree ofimprovement in classification efficiency and classification accuracy than the traditional federated learning.

常用功能

更新日期/Last Update: 2023-09-10