Fish Image Classification Using Adaptive Learning Rate In Transfer Learning Method

R. Suhana, W. Mahmudy, Agung Setia Budi
{"title":"Fish Image Classification Using Adaptive Learning Rate In Transfer Learning Method","authors":"R. Suhana, W. Mahmudy, Agung Setia Budi","doi":"10.17977/um018v5i12022p67-77","DOIUrl":null,"url":null,"abstract":"The existence of fish species diversity in coastal ecosystems which include mangrove forests, seagrass beds and coral reefs is one of the benchmarks in determining health in coastal ecosystems. It is certain that we must maintain, preserve and care for so that conservation efforts need to be carried out in water areas. Many experts at the Indonesian Fisheries and Marine Research and Development Agency often classify fish images manually, of course it will take a long time, therefore with today's developments they can use the latest technology.  One of the reliable techniques in terms of image classification is Convolutional Neural Network (CNN). As time goes by, of course, many people want fast learning and solving new problems faster and better, so transfer learning appears, which adopts part of CNN, the name is modified convolution layer. Observing the needs of experts in the field of marine conservation, the researchers decided to solve this problem by using transfer learning modifications. The transfer learning used is an architectural model from the pre-trained Mobilenet V2, which is known for its light computing process and can be applied to our gadgets and other embedded tools. The research image data used is 49.281 data of various sizes and there are 18 types of fish, in the pre-processing data there is a resize of the image to a size of 224x224 pixels. testing with the modified transfer learning architectural model obtained an accuracy score of 99.54%, this model is quite reliable in classifying fish images.","PeriodicalId":52868,"journal":{"name":"Knowledge Engineering and Data Science","volume":"1 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2022-06-07","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"Knowledge Engineering and Data Science","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.17977/um018v5i12022p67-77","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0

Abstract

The existence of fish species diversity in coastal ecosystems which include mangrove forests, seagrass beds and coral reefs is one of the benchmarks in determining health in coastal ecosystems. It is certain that we must maintain, preserve and care for so that conservation efforts need to be carried out in water areas. Many experts at the Indonesian Fisheries and Marine Research and Development Agency often classify fish images manually, of course it will take a long time, therefore with today's developments they can use the latest technology.  One of the reliable techniques in terms of image classification is Convolutional Neural Network (CNN). As time goes by, of course, many people want fast learning and solving new problems faster and better, so transfer learning appears, which adopts part of CNN, the name is modified convolution layer. Observing the needs of experts in the field of marine conservation, the researchers decided to solve this problem by using transfer learning modifications. The transfer learning used is an architectural model from the pre-trained Mobilenet V2, which is known for its light computing process and can be applied to our gadgets and other embedded tools. The research image data used is 49.281 data of various sizes and there are 18 types of fish, in the pre-processing data there is a resize of the image to a size of 224x224 pixels. testing with the modified transfer learning architectural model obtained an accuracy score of 99.54%, this model is quite reliable in classifying fish images.
基于自适应学习率的鱼类图像分类迁移学习方法
包括红树林、海草床和珊瑚礁在内的沿海生态系统中鱼类物种多样性的存在是决定沿海生态系统健康状况的基准之一。可以肯定的是,我们必须维护、保护和照顾,以便在水域开展保护工作。印尼渔业和海洋研究与发展局的许多专家经常手动对鱼类图像进行分类,当然这需要很长时间,因此随着今天的发展,他们可以使用最新的技术。卷积神经网络是图像分类的可靠技术之一。当然,随着时间的推移,许多人希望快速学习,更快更好地解决新问题,因此出现了迁移学习,它采用了CNN的一部分,名称为修改卷积层。考虑到海洋保护领域专家的需求,研究人员决定通过迁移学习修改来解决这个问题。所使用的迁移学习是预训练的Mobilenet V2的架构模型,该模型以其轻计算过程而闻名,可以应用于我们的小工具和其他嵌入式工具。所使用的研究图像数据是各种大小的49.281个数据,有18种类型的鱼,在预处理数据中,将图像调整为224x224像素的大小。用改进的迁移学习结构模型进行测试,获得了99.54%的准确率,该模型在鱼类图像分类中是相当可靠的。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
求助全文
约1分钟内获得全文 求助全文
来源期刊
自引率
0.00%
发文量
4
审稿时长
8 weeks
×
引用
GB/T 7714-2015
复制
MLA
复制
APA
复制
导出至
BibTeX EndNote RefMan NoteFirst NoteExpress
×
提示
您的信息不完整,为了账户安全,请先补充。
现在去补充
×
提示
您因"违规操作"
具体请查看互助需知
我知道了
×
提示
确定
请完成安全验证×
copy
已复制链接
快去分享给好友吧!
我知道了
右上角分享
点击右上角分享
0
联系我们:info@booksci.cn Book学术提供免费学术资源搜索服务,方便国内外学者检索中英文文献。致力于提供最便捷和优质的服务体验。 Copyright © 2023 布克学术 All rights reserved.
京ICP备2023020795号-1
ghs 京公网安备 11010802042870号
Book学术文献互助
Book学术文献互助群
群 号:481959085
Book学术官方微信