[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"doc-seo-663526-105":3,"detail-sidebar-cat-0-en-105":80,"doc-detail-663526-en":130},{"code":4,"msg":5,"data":6},0,"ok",{"site_id":7,"language":8,"slug":9,"title":10,"keywords":11,"description":12,"schema_data":13,"social_meta":73,"head_meta":75,"extra_data":77,"updated_unix":79},105,"en","mu-net-optlstm-two-stream-spatialtemporal-feature-extraction-and-classification-architecture","MU-Net-optLSTM - Two-Stream Spatial–Temporal Feature Extraction and Classification Architecture","","Networked cameras continuously capturing video create strong demand for hybrid edge-to-cloud systems that can process live streams in real time. Although art museum environments are seldom studied, visual analysis is crucial for distinguishing individuals and crowds in smart surveillance. This research presents a deep learning framework for detecting abnormal behavior from museum video, combining U-Net-based spatial feature extraction, MobileNetV2 encoding, improved optimized LSTM for temporal features, and optical flow for motion tracking. Results reach 97.67±1.23% average accuracy on video datasets, offering efficiency and stronger generalization than prior work.",{"@graph":14,"@context":72},[15,34,55],{"@type":16,"itemListElement":17},"BreadcrumbList",[18,23,27,31],{"item":19,"name":20,"@type":21,"position":22},"https://docshare.wps.com","Home","ListItem",1,{"item":24,"name":25,"@type":21,"position":26},"https://docshare.wps.com/document/","Document",2,{"item":28,"name":29,"@type":21,"position":30},"https://docshare.wps.com/document/research-report/","Research & Report",3,{"item":32,"name":10,"@type":21,"position":33},"https://docshare.wps.com/document/mu-net-optlstm-two-stream-spatialtemporal-feature-extraction-and-classification-architecture/663526/",4,{"url":32,"name":10,"@type":35,"image":36,"author":41,"headline":10,"publisher":44,"fileFormat":47,"inLanguage":8,"description":12,"dateModified":48,"datePublished":49,"encodingFormat":47,"isAccessibleForFree":50,"interactionStatistic":51},"DigitalDocument",{"url":37,"@type":38,"width":39,"height":40},"https://docshare.wps.com/thumbnails/mu-net-optlstm-two-stream-spatialtemporal-feature-extraction-and-classification-architecture/663526.png","ImageObject",300,407,{"name":42,"@type":43},"Margaret","Person",{"url":19,"name":45,"@type":46},"DocShare","Organization","application/pdf","2026-10-30","2026-10-28",true,{"@type":52,"interactionType":53,"userInteractionCount":30},"InteractionCounter",{"@type":54},"ViewAction",{"@type":56,"mainEntity":57},"FAQPage",[58,64,68],{"name":59,"@type":60,"acceptedAnswer":61},"What problem does the paper address in crowded art museums?","Question",{"text":62,"@type":63},"It targets abnormal-behavior identification from continuously captured video in crowded art museum settings, where real-time analysis is required for reliable monitoring.","Answer",{"name":65,"@type":60,"acceptedAnswer":66},"Which deep learning components are combined in the proposed MU-Net-optLSTM framework?",{"text":67,"@type":63},"The method uses U-Net for spatial feature extraction, MobileNetV2 as the encoder, and an improved optimized LSTM for temporal feature extraction, enhanced by optical flow.",{"name":69,"@type":60,"acceptedAnswer":70},"How is performance evaluated and what accuracy is reported?",{"text":71,"@type":63},"The approach is tested on a collection of video datasets, achieving an average accuracy of 97.67±1.23%, with claims of computational efficiency and broader generalizability versus state-of-the-art methods.","https://schema.org",{"og:url":32,"og:type":74,"og:title":10,"og:site_name":45,"og:description":12},"article",{"robots":76,"canonical":32},"index,follow",{"doc_id":78,"site_id":7},663526,1793228934,{"code":4,"msg":81,"data":82},"success",[83,87,91,95,100,105,110,114,119,122,126],{"id":22,"doc_module":4,"doc_module_name":25,"category_name":84,"show_sort_weight":85,"slug":86},"Story & Novel",90,"story-novel",{"id":26,"doc_module":4,"doc_module_name":25,"category_name":88,"show_sort_weight":89,"slug":90},"Literature",80,"literature",{"id":33,"doc_module":4,"doc_module_name":25,"category_name":92,"show_sort_weight":93,"slug":94},"Exam",70,"exam",{"id":96,"doc_module":4,"doc_module_name":25,"category_name":97,"show_sort_weight":98,"slug":99},5,"Comic",60,"comic",{"id":101,"doc_module":4,"doc_module_name":25,"category_name":102,"show_sort_weight":103,"slug":104},6,"Technology",50,"technology",{"id":106,"doc_module":4,"doc_module_name":25,"category_name":107,"show_sort_weight":108,"slug":109},7,"Healthcare",40,"healthcare",{"id":111,"doc_module":4,"doc_module_name":25,"category_name":29,"show_sort_weight":112,"slug":113},8,30,"research-report",{"id":115,"doc_module":4,"doc_module_name":25,"category_name":116,"show_sort_weight":117,"slug":118},9,"Religion & Spirituality",20,"religion-spirituality",{"id":117,"doc_module":4,"doc_module_name":25,"category_name":120,"show_sort_weight":117,"slug":121},"World Cup","world-cup",{"id":123,"doc_module":4,"doc_module_name":25,"category_name":124,"show_sort_weight":123,"slug":125},10,"Lifestyle","lifestyle",{"id":127,"doc_module":4,"doc_module_name":25,"category_name":128,"show_sort_weight":96,"slug":129},19,"General","general",{"code":4,"msg":81,"data":131},{"doc_id":78,"user_id":132,"nickname":42,"user_avatar":133,"doc_module":4,"category_id":111,"category_name":29,"doc_title":10,"doc_description":12,"doc_content":134,"file_id":135,"file_url":136,"file_type":137,"file_size":138,"view_count":30,"is_deleted":4,"is_public":22,"is_downloadable":22,"audit_status":22,"page_count":139,"language":140,"language_code":8,"site_id":7,"html_lang":8,"table_of_contents":141,"faqs":142,"seo_title":143,"seo_description":12,"update_tm":79,"read_time":144},137451207643,"https://ap-avatar.wpscdn.com/davatar_3d24733baf745e90a7e4bdd5f77d97b2","TSINGHUA SCIENCE AND TECHNOLOGY ISSN 1007-0214 23/23 pp2304−2321 DOI: 10 . 26599/TST . 2024 . 9010003 Volume 31 , Number 4 , August 2026  \nResearch Article  \nMU-Net-optLSTM: Two-Stream Spatial–Temporal Feature Extraction and Classification Architecture for Automatic Monitoring of Crowded Art Museums  \nMukun Wang, Rongju Yao, and Khosro Rezaee*  \nAbstract: Networked cameras that continuously capture video data have generated a high demand for hybrid edge-to-cloud servers that can process live videos in real time. The environment of art museums is rarely studied, but visual analysis is an important factor in categorizing and distinguishing individuals and crowds through smart surveillance systems. This paper demonstrates how video surveillance data from art museums can be analyzed to identify abnormal behavior using an innovative deep learning framework. To enhance the extracted features, a spatial feature extraction method based on the U-Net architecture is applied, along with the encoder component of the proposed approach, MobileNetV2 . Additionally, we propose an improved LongShort-Term Memory (LSTM) algorithm for extracting temporal features. Optical flow enhances surveillance in art museums by tracking individuals and crowds. Our approach yields an average accuracy of 97.67±1 .23% when applied to a collection of video datasets. Using U-Net, MobileNetV2, and optimized LSTM algorithms, the model recognizes patterns in video data, such as crowd motion in museums. Consequently, this methodology generates reliable results as well as being computationally efficient. Compared to the state-of-the-art, the proposed method is more comprehensive and generalizable for analyzing atypical museum visitor behavior.  \nKey words: art museums; automatic monitoring ; deep learning ; long short-term memory; U-Net architecture ; MobileNetV2 ; optical flow  \n1 Introduction  \nEfficient Video surveillance technology is employed by security professionals to evaluate collective behavior  \n Mukun Wang is with Department of Design, Graduate School, Dongseo University, Busan 47011, Republic of Korea, and also with Tianshui Normal University, Tianshui 741000, China. [E-mail: miachiyu2022@gmail.com](E-mail: miachiyu2022@gmail.com).  \n Rongju Yao is with Shandong Provincial University Laboratory for Protected Horticulture, Weifang University of Science and Technology, Weifang 262700, China. E-mail: yaorongju8@ [wfust.edu.cn](wfust.edu.cn).  \n Khosro Rezaee is with Department of Biomedical Engineering, Meybod University, Yazd 89195, Iran. E-mail: [kh.rezaee@meybod.ac.ir](kh.rezaee@meybod.ac.ir).  \n* To whom correspondence should be addressed.  \nManuscript received: 2023-11-24; revised: 2023-12-18;  \naccepted: 2023-12-29  \nin public spaces[1, 2] . In recent years, the significance of employing many smart cameras in an edge-fog-cloud infrastructure for continuous surveillance of individuals and crowds has increased[3, 4] . Data-driven crowd simulations[4, 5] and advanced driving assistance systems[6, 7] heavily depend on multi-person and multicamera tracking techniques. Among the significant places that require surveillance and video analysis are art museums[8] . Insufficient research has been conducted to assess the effectiveness of automated video surveillance and crowd analysis in art museum environment However, art museums and other organizations responsible for safeguarding precious artifacts are legally obligated to implement all essential measures to protect data confidentiality[9] . Maintaining records of surveillance actions inside monitored zones  \n© The author(s) 2026. The articles published in this open access journal are distributed under the terms of the  \nCreative Commons Attribution 4.0 International License ([http://creativecommons.org/licenses/by/4.0/](http://creativecommons.org/licenses/by/4.0/)).  \nMukun Wang et al.: MU-Net-optLSTM: Two-Stream Spatial–Temporal Feature Extraction and Classification... 2305  \nis difficult, and recording and ","cbCailBBprw3I0lN","https://ap.wps.com/l/cbCailBBprw3I0lN","pdf",3958592,18,"English","# Introduction\n## Related work and motivation","[{\"question\":\"What problem does the paper address in crowded art museums?\",\"answer\":\"It targets abnormal-behavior identification from continuously captured video in crowded art museum settings, where real-time analysis is required for reliable monitoring.\"},{\"question\":\"Which deep learning components are combined in the proposed MU-Net-optLSTM framework?\",\"answer\":\"The method uses U-Net for spatial feature extraction, MobileNetV2 as the encoder, and an improved optimized LSTM for temporal feature extraction, enhanced by optical flow.\"},{\"question\":\"How is performance evaluated and what accuracy is reported?\",\"answer\":\"The approach is tested on a collection of video datasets, achieving an average accuracy of 97.67±1.23%, with claims of computational efficiency and broader generalizability versus state-of-the-art methods.\"}]","MU-Net-optLSTM - Two-Stream Spatial–Temporal Feature Extraction and Classification Architecture | PDF",45]