使用基于 CSA 模型的改进型斯温变换器进行物联网大数据视频人脸跟踪

Anbumani K, Cuddapah Anitha, Achuta Rao S V, Praveen Kumar K, Meganathan Ramasamy, M. R.
{"title":"使用基于 CSA 模型的改进型斯温变换器进行物联网大数据视频人脸跟踪","authors":"Anbumani K, Cuddapah Anitha, Achuta Rao S V, Praveen Kumar K, Meganathan Ramasamy, M. R.","doi":"10.53759/7669/jmc202404029","DOIUrl":null,"url":null,"abstract":"Even though Convolutional Neural Networks (CNNs) have greatly improved face-related algorithms, it is still difficult to keep both accuracy and efficiency in real-world applications. The most cutting-edge approaches use deeper networks to improve performance, but the increased computing complexity and number of parameters make them impractical for usage in mobile applications. To tackle these issues, this article presents a model for object detection that combines Deeplabv3+ with Swin transformer, which incorporates GLTB and Swin-Conv-Dspp (SCD). To start with, in order to lessen the impact of the hole phenomena and the loss of fine-grained data, we employ the SCD component, which is capable of efficiently extracting feature information from objects at various sizes. Secondly, in order to properly address the issue of challenging object recognition due to occlusion, the study builds a GLTB with a spatial pyramid pooling shuffle module. This module allows for the extraction of important detail information from the few noticeable pixels of the blocked objects. Crocodile search algorithm (CSA) enhances classification accuracy by properly selecting the model's fine-tuning. On a benchmark dataset known as WFLW, the study experimentally validates the suggested model. Compared to other light models, the experimental findings show that it delivers higher performance with significantly fewer parameters and reduced computing complexity.","PeriodicalId":516221,"journal":{"name":"Journal of Machine and Computing","volume":"69 1","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2024-04-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Video Face Tracking for IoT Big Data using Improved Swin Transformer based CSA Model\",\"authors\":\"Anbumani K, Cuddapah Anitha, Achuta Rao S V, Praveen Kumar K, Meganathan Ramasamy, M. R.\",\"doi\":\"10.53759/7669/jmc202404029\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"Even though Convolutional Neural Networks (CNNs) have greatly improved face-related algorithms, it is still difficult to keep both accuracy and efficiency in real-world applications. The most cutting-edge approaches use deeper networks to improve performance, but the increased computing complexity and number of parameters make them impractical for usage in mobile applications. To tackle these issues, this article presents a model for object detection that combines Deeplabv3+ with Swin transformer, which incorporates GLTB and Swin-Conv-Dspp (SCD). To start with, in order to lessen the impact of the hole phenomena and the loss of fine-grained data, we employ the SCD component, which is capable of efficiently extracting feature information from objects at various sizes. Secondly, in order to properly address the issue of challenging object recognition due to occlusion, the study builds a GLTB with a spatial pyramid pooling shuffle module. This module allows for the extraction of important detail information from the few noticeable pixels of the blocked objects. Crocodile search algorithm (CSA) enhances classification accuracy by properly selecting the model's fine-tuning. On a benchmark dataset known as WFLW, the study experimentally validates the suggested model. Compared to other light models, the experimental findings show that it delivers higher performance with significantly fewer parameters and reduced computing complexity.\",\"PeriodicalId\":516221,\"journal\":{\"name\":\"Journal of Machine and Computing\",\"volume\":\"69 1\",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2024-04-05\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Journal of Machine and Computing\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.53759/7669/jmc202404029\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Journal of Machine and Computing","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.53759/7669/jmc202404029","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
引用次数: 0

摘要

尽管卷积神经网络(CNN)极大地改进了与人脸有关的算法,但在实际应用中仍难以保证准确性和效率。最前沿的方法使用深度网络来提高性能,但计算复杂度和参数数量的增加使其在移动应用中的使用变得不切实际。为了解决这些问题,本文提出了一种结合 Deeplabv3+ 和 Swin 变换器的物体检测模型,其中包含 GLTB 和 Swin-Conv-Dspp (SCD)。首先,为了减少洞穴现象的影响和细粒度数据的丢失,我们采用了 SCD 组件,它能够有效地提取各种大小物体的特征信息。其次,为了妥善解决因遮挡而对物体识别造成挑战的问题,本研究建立了一个带有空间金字塔池化洗牌模块的 GLTB。该模块可以从被遮挡物体的少数明显像素中提取重要的细节信息。鳄鱼搜索算法(CSA)通过适当选择模型的微调来提高分类精度。本研究在一个名为 WFLW 的基准数据集上对所建议的模型进行了实验验证。实验结果表明,与其他轻型模型相比,该模型以更少的参数和更低的计算复杂度实现了更高的性能。
本文章由计算机程序翻译,如有差异,请以英文原文为准。
Video Face Tracking for IoT Big Data using Improved Swin Transformer based CSA Model
Even though Convolutional Neural Networks (CNNs) have greatly improved face-related algorithms, it is still difficult to keep both accuracy and efficiency in real-world applications. The most cutting-edge approaches use deeper networks to improve performance, but the increased computing complexity and number of parameters make them impractical for usage in mobile applications. To tackle these issues, this article presents a model for object detection that combines Deeplabv3+ with Swin transformer, which incorporates GLTB and Swin-Conv-Dspp (SCD). To start with, in order to lessen the impact of the hole phenomena and the loss of fine-grained data, we employ the SCD component, which is capable of efficiently extracting feature information from objects at various sizes. Secondly, in order to properly address the issue of challenging object recognition due to occlusion, the study builds a GLTB with a spatial pyramid pooling shuffle module. This module allows for the extraction of important detail information from the few noticeable pixels of the blocked objects. Crocodile search algorithm (CSA) enhances classification accuracy by properly selecting the model's fine-tuning. On a benchmark dataset known as WFLW, the study experimentally validates the suggested model. Compared to other light models, the experimental findings show that it delivers higher performance with significantly fewer parameters and reduced computing complexity.
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
自引率
0.00%
发文量
0
×
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:604180095
Book学术官方微信