CNN模型转移到无人机图像灾害事件分类中的超参数优化系统配置

Q3 Computer Science
Supaporn Bunrit, Nittaya Kerdprasop, Kittisak Kerdprasop
{"title":"CNN模型转移到无人机图像灾害事件分类中的超参数优化系统配置","authors":"Supaporn Bunrit, Nittaya Kerdprasop, Kittisak Kerdprasop","doi":"10.18178/joig.11.3.263-270","DOIUrl":null,"url":null,"abstract":"Deep learning and computer vision-based approaches incorporated with the evolution of the relevant technologies of Unmanned Aerial Vehicles (UAVs) and drones have significantly motivated the advancements of disaster management applications. This research studied a classification method for disaster event identification from UAV images that is suitable for disaster monitoring. A Convolution Neural Network (CNN) of GoogleNet models that were pretrained from ImageNet and Place365 datasets was explored to find the appropriate one for fine-tuning to classify the disaster events. In order to get the optimal performance, a systematic configuration for searching the hyperparameters in fine-tuning the CNN model was proposed. The top three hyperparameters that affect the performance, which are the initial learning rate, the number of epochs, and the minibatch size, were systematically set and tuned for each configuration. The proposed approach consists of five stages, during which three types of trials were used to monitor different sets of the hyperparameters. The experimental result revealed that by applying the proposed approach the model performance can increase up to 5%. The optimal performance achieved was 98.77 percent accuracy. For UAV/drone applications, where a small onboard model is preferred, GoogleNet that is quite small in model size and has a good structure for further fine tuning is suitable to deploy.","PeriodicalId":36336,"journal":{"name":"中国图象图形学报","volume":null,"pages":null},"PeriodicalIF":0.0000,"publicationDate":"2023-09-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Systematic Configuration for Hyperparameters Optimization in Transferring of CNN Model to Disaster Events Classification from UAV Images\",\"authors\":\"Supaporn Bunrit, Nittaya Kerdprasop, Kittisak Kerdprasop\",\"doi\":\"10.18178/joig.11.3.263-270\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Deep learning and computer vision-based approaches incorporated with the evolution of the relevant technologies of Unmanned Aerial Vehicles (UAVs) and drones have significantly motivated the advancements of disaster management applications. This research studied a classification method for disaster event identification from UAV images that is suitable for disaster monitoring. A Convolution Neural Network (CNN) of GoogleNet models that were pretrained from ImageNet and Place365 datasets was explored to find the appropriate one for fine-tuning to classify the disaster events. In order to get the optimal performance, a systematic configuration for searching the hyperparameters in fine-tuning the CNN model was proposed. The top three hyperparameters that affect the performance, which are the initial learning rate, the number of epochs, and the minibatch size, were systematically set and tuned for each configuration. The proposed approach consists of five stages, during which three types of trials were used to monitor different sets of the hyperparameters. The experimental result revealed that by applying the proposed approach the model performance can increase up to 5%. The optimal performance achieved was 98.77 percent accuracy. For UAV/drone applications, where a small onboard model is preferred, GoogleNet that is quite small in model size and has a good structure for further fine tuning is suitable to deploy.\",\"PeriodicalId\":36336,\"journal\":{\"name\":\"中国图象图形学报\",\"volume\":null,\"pages\":null},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2023-09-01\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"中国图象图形学报\",\"FirstCategoryId\":\"1093\",\"ListUrlMain\":\"https://doi.org/10.18178/joig.11.3.263-270\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"Q3\",\"JCRName\":\"Computer Science\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"中国图象图形学报","FirstCategoryId":"1093","ListUrlMain":"https://doi.org/10.18178/joig.11.3.263-270","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q3","JCRName":"Computer Science","Score":null,"Total":0}
引用次数: 0

摘要

深度学习和基于计算机视觉的方法与无人机(uav)和无人机相关技术的发展相结合,极大地推动了灾害管理应用的进步。本文研究了一种适用于灾害监测的无人机图像灾害事件识别分类方法。通过对ImageNet和Place365数据集预训练的GoogleNet模型的卷积神经网络(CNN)进行探索,找到合适的卷积神经网络进行微调,以对灾难事件进行分类。为了获得最优的性能,提出了一种用于微调CNN模型超参数搜索的系统配置。影响性能的三个最重要的超参数是初始学习率、epoch数和minibatch大小,这些都是针对每种配置系统设置和调优的。提出的方法包括五个阶段,在此期间,三种类型的试验被用来监测不同的超参数集。实验结果表明,采用该方法可使模型性能提高5%。获得的最佳性能为98.77%的准确率。对于无人机/无人机应用,首选小型机载模型,模型尺寸相当小且具有良好的结构以进行进一步微调的GoogleNet适合部署。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
Systematic Configuration for Hyperparameters Optimization in Transferring of CNN Model to Disaster Events Classification from UAV Images
Deep learning and computer vision-based approaches incorporated with the evolution of the relevant technologies of Unmanned Aerial Vehicles (UAVs) and drones have significantly motivated the advancements of disaster management applications. This research studied a classification method for disaster event identification from UAV images that is suitable for disaster monitoring. A Convolution Neural Network (CNN) of GoogleNet models that were pretrained from ImageNet and Place365 datasets was explored to find the appropriate one for fine-tuning to classify the disaster events. In order to get the optimal performance, a systematic configuration for searching the hyperparameters in fine-tuning the CNN model was proposed. The top three hyperparameters that affect the performance, which are the initial learning rate, the number of epochs, and the minibatch size, were systematically set and tuned for each configuration. The proposed approach consists of five stages, during which three types of trials were used to monitor different sets of the hyperparameters. The experimental result revealed that by applying the proposed approach the model performance can increase up to 5%. The optimal performance achieved was 98.77 percent accuracy. For UAV/drone applications, where a small onboard model is preferred, GoogleNet that is quite small in model size and has a good structure for further fine tuning is suitable to deploy.
求助全文
通过发布文献求助,成功后即可免费获取论文全文。 去求助
来源期刊
中国图象图形学报
中国图象图形学报 Computer Science-Computer Graphics and Computer-Aided Design
CiteScore
1.20
自引率
0.00%
发文量
6776
期刊介绍: Journal of Image and Graphics (ISSN 1006-8961, CN 11-3758/TB, CODEN ZTTXFZ) is an authoritative academic journal supervised by the Chinese Academy of Sciences and co-sponsored by the Institute of Space and Astronautical Information Innovation of the Chinese Academy of Sciences (ISIAS), the Chinese Society of Image and Graphics (CSIG), and the Beijing Institute of Applied Physics and Computational Mathematics (BIAPM). The journal integrates high-tech theories, technical methods and industrialisation of applied research results in computer image graphics, and mainly publishes innovative and high-level scientific research papers on basic and applied research in image graphics science and its closely related fields. The form of papers includes reviews, technical reports, project progress, academic news, new technology reviews, new product introduction and industrialisation research. The content covers a wide range of fields such as image analysis and recognition, image understanding and computer vision, computer graphics, virtual reality and augmented reality, system simulation, animation, etc., and theme columns are opened according to the research hotspots and cutting-edge topics. Journal of Image and Graphics reaches a wide range of readers, including scientific and technical personnel, enterprise supervisors, and postgraduates and college students of colleges and universities engaged in the fields of national defence, military, aviation, aerospace, communications, electronics, automotive, agriculture, meteorology, environmental protection, remote sensing, mapping, oil field, construction, transportation, finance, telecommunications, education, medical care, film and television, and art. Journal of Image and Graphics is included in many important domestic and international scientific literature database systems, including EBSCO database in the United States, JST database in Japan, Scopus database in the Netherlands, China Science and Technology Thesis Statistics and Analysis (Annual Research Report), China Science Citation Database (CSCD), China Academic Journal Network Publishing Database (CAJD), and China Academic Journal Network Publishing Database (CAJD). China Science Citation Database (CSCD), China Academic Journals Network Publishing Database (CAJD), China Academic Journal Abstracts, Chinese Science Abstracts (Series A), China Electronic Science Abstracts, Chinese Core Journals Abstracts, Chinese Academic Journals on CD-ROM, and China Academic Journals Comprehensive Evaluation Database.
×
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