模态共性与模态特性：跨模态行人再识别思考

对特定行人目标进行跨摄像头跨场景检索，在公共监控、智能安防、城市治理等领域具有重要应用价值，已成为计算机视觉的研究热点之一。以低质量行人特征表达、移动行人目标检测、跨视域、跨模态行人再识别等为代表的智能感知与推理在真实开放场景行人检索发展中起着非常重要的作用。近年来，团队结合平台积累和优势，在领域权威期刊和人工智能领域顶级会议等发表论文20余篇，本次报告将结合团队近年来在跨模态行人再识别领域的部分研究成果，探讨弱监督机器学习、跨模态特征一致性表达、模态融合与交互等在开放场景行人目标搜索的若干关键技术及其未来发展方向。

金一，北京交通大学计算机学院教授，博导，CCF杰出会员。担任CCF YOCSEF副主席(25-26)，智慧交通分会常委，多媒体专委会、大数据专委会执委等。任“十四五”国家重点研发计划重点专项专家组委员，中国图像图形学学会首批科普基地负责人等。主要研究领域包括：多模态数据融合感知、交通视频语义理解、可信行为分析及多媒体隐私保护等。主持国家级、省部级项目10余项，发表学术论文70余篇，ESI高被引论文5篇，其中包括IEEE/ACM汇刊等领域重要期刊和CCF A类会议CVPR，AAAI，ICCV，ICAI，ACM MM等。授权国家发明专利37项，参编国家、行业标准3项。获IEEE Computer Society年度最佳论文奖提名奖等国际论文奖励3项，第五十届日内瓦国际创新发明银奖、2023年中国产学研合作创新与促进奖创新成果奖、2024年通信学会科技进步奖、铁道学会科技进步奖等奖项，入选2023年度北京市轨道交通学会杰出青年人才。

高分辨极化SAR影像智能解译方法研究 / Deep learning for High-Resolution Polarimetric SAR Images: Challenges and Approaches

高分辨极化SAR影像精准分类是遥感影像智能解译的前提。报告针对高分辨极化SAR遥感影像的成像特点，结合物理散射机理，研究黎曼空间下的复杂遥感数据的特征学习方法，针对极化SAR影像分类中的存在的斑点噪声影响大、地物尺度差异大和复杂数据不可分等挑战问题，提出一系列遥感图像分类模型与方法。

Accurate classification of high-resolution polarimetric SAR images is a prerequisite for the intelligent interpretation of remote sensing data. This report aims to learn the unique imaging characteristics of high-resolution polarimetric SAR imagery by integrating physical scattering mechanisms with advanced feature learning methods on Riemannian manifolds. Specifically, it focuses on overcoming major challenges in polarimetric SAR image classification, including strong speckle noise, large variations in object scales, and the non-separability of complex data. To this end, this report gives a series of novel classification models and methodologies, providing a solid technical foundation for intelligent and precise interpretation of remote sensing imagery.

石俊飞，西安理工大学计算机学院副教授，硕士生导师，软件工程系副主任；IEEE高级会员，CCF高级会员，中国图象图形学学会(CSIG)会员，CCF女工委委员，CSIG女工委委员；获陕西省科学技术二等奖一项，中国发明协会创新创业奖二等奖一项，研究方向为：遥感图像处理、计算机视觉。近年主持国家基金面上项目1项，青年项目1项，陕西省基金1项，厅局级基金3项，参与国家级和省级基金十余项；以第一或通讯作者在IEEE TGRS, TCSVT, Pattern Recognition, Applied soft computing, 自动化学报等国际国内Top期刊发表论文30余篇，申请发明专利十余项，出版教材一部，撰写专著一部。现为TGRS, TNNLS, ISPRS, TMM, Knowledge-based system等国际知名期刊审稿人。

Junfei Shi, Associate Professor at the School of Computer Science, Xi’an University of Technology, and Deputy Director of the Department of Software Engineering. She is a IEEE Senior Member , CCF Senior Member , CSIG Member, a committee member of the Women’s Working Committee of CSIG and CCF. She has received the Second Prize of the Shaanxi Provincial Science and Technology Award and the Second Prize of the China Invention Association Innovation and Entrepreneurship Award. Her research interests include remote sensing image processing and computer vision.

In recent years, she has presided over one NSFC General Program, one NSFC Youth Program, one Shaanxi Provincial Science Foundation project, and three departmental-level projects, and has participated in more than ten national and provincial projects. As the first or corresponding author, she has published over 30 papers in top international and domestic journals such as IEEE TGRS, TCSVT, Pattern Recognition, Applied Soft Computing, and Acta Automatica Sinica. She has applied for more than ten invention patents, published one textbook, and authored one monograph. She currently serves as a reviewer for several prestigious international journals, including IEEE TGRS, TNNLS, ISPRS Journal, IEEE TMM, and Knowledge-Based Systems.

复杂地下空间火灾的精准识别与智能检测方法/ Accurate Identification and Intelligent Detection Methods for Fires in Complex Underground Spaces

复杂地下空间火灾隐蔽性强、烟气蔓延快，传统检测手段常因环境干扰误报漏报，精准识别与智能检测真的能破解这一难题吗？该技术如今发展到了什么阶段？

报告围绕复杂地下空间火灾的精准识别与智能检测方法这一核心任务，阐述课题组近年来在数据驱动的火灾检测技术上的研究成果与应用实践。同时，介绍技术在地铁隧道、地下商场等典型场景的部署案例，展现其在火灾早期预警、隐患排查中的实际成效，旨在为复杂地下空间的消防安全防控提供关键技术支撑，助力智慧消防的精准化、智能化发展。

Fires in complex underground spaces are highly concealed and spread rapidly with smoke. Traditional detection methods often suffer from false alarms or missed alarms due to environmental interference. Can accurate identification and intelligent detection really solve this problem? What stage has this technology reached now?

Centering on the core task of accurate identification and intelligent detection methods for fires in complex underground spaces, this report elaborates on the research achievements and application practices of the research team in data-driven fire detection technology in recent years. Meanwhile, it introduces the deployment cases of the technology in typical scenarios such as subway tunnels and underground shopping malls, demonstrating its practical effectiveness in early fire warning and hidden danger inspection. The report aims to provide key technical support for fire safety prevention and control in complex underground spaces, and contribute to the precise and intelligent development of smart fire protection.

崔金荣，副教授，博士，硕士研究生导师；CCF和CSIG会员，YOCSEF广州25-26学术AC，CSIG青工委委员，CSIG-生物特征识别专委会副秘书长，广东省图象图形学会计算机视觉专委会副秘书长，广东省2024年农村科技特派员。博士毕业于哈尔滨工业大学（深圳），主要研究方向是掌纹识别，缺失多视图聚类，计算机视觉，智慧消防等；2019年6月作为访问学者赴英国曼彻斯特大学交流访问一年。发表学术期刊和会议论文40余篇，并以第一作者/通讯作者在TIP、TNNLS、Neural Networks等国际权威期刊和ACM MM、ICME等重要国际会议上发表论文30余篇。先后主持和参与包括国家自然科学基金面上项目、青年项目等在内的国家级、省部级以及横向科技项目10余项，获广东省计算机学会优秀论文奖一等奖等。

Cui Jinrong: Associate Professor, Ph.D., Master Supervisor; Member of CCF (China Computer Federation) and CSIG (Chinese Society of Image and Graphics); Academic AC (Academic Committee) of CCF YOCSEF Guangzhou 25-26 Session ; Member of CSIG Youth Working Committee; Deputy Secretary-General of CSIG Biometric Recognition Special Committee; Deputy Secretary-General of Computer Vision Special Committee, Guangdong Society of Image and Graphics; Rural Science and Technology Commissioner of Guangdong Province (2024).

She obtained her Ph.D. from Harbin Institute of Technology, Shenzhen. Her main research interests include palmprint recognition, incomplete multi-view clustering, computer vision, and smart fire protection. In June 2019, she went to the University of Manchester, UK as a visiting scholar for one-year academic exchange .

She has published over 40 academic journal and conference papers, among which more than 30 were published as first author/corresponding author in international top journals such as IEEE Transactions on Image Processing (TIP), IEEE Transactions on Neural Networks and Learning Systems (TNNLS), Neural Networks, and important international conferences including ACM MM (ACM International Conference on Multimedia) and ICME . She has presided over and participated in more than 10 national, provincial, and horizontal scientific and technological projects, including General Program and Youth Program of the National Natural Science Foundation of China. She also won awards such as the First Prize of Excellent Paper Award from Guangdong Computer Society.

复杂场景下大模型驱动的多模态目标检测 / Large Models-Driven Multimodal Object Detection in Complex Environments

目标检测是计算机视觉领域的共性基础问题，致力于从图像中识别并定位出用户感兴趣的目标对象，在公共安全、智能制造、智能交通等诸多领域，具有重要的理论意义和应用价值。本报告主要针对复杂场景的对象形状多样、分布密集、样本稀缺等难题，深入探讨如何利用大模型的强大表征能力实现技术突破，并重点介绍形变自适应目标表征、大模型驱动的细粒度属性关联与零样本多模态检测等关键技术，总结大模型赋能复杂场景目标检测的技术优势与未来发展方向，为复杂场景下的智能感知提供新思路与新方法。

Object detection is a fundamental and common problem in computer vision, which aims to focus on identifying and localizing target objects of interest within images. It holds significant theoretical importance and practical value across numerous fields, including public security, intelligent manufacturing, and smart transportation. This presentation addresses the critical challenges in complex scenes, such as diverse object shapes, dense distributions, and a few of samples. It delves into leveraging the powerful representational capabilities of large models to achieve technological breakthroughs. Key technologies to be introduced include deformable adaptive object representation, large model-driven fine-grained attribute alignment, and zero-shot multimodal detection. The report will conclude by summarizing the technical advantages and future development directions of large model-empowered object detection in complex scenarios, aiming to provide new perspectives and methodologies for intelligent perception in challenging environments.

邱荷茜，电子科技大学信息与通信工程学院副教授，硕士生导师。入选CSIG博士学位论文激励计划提名、首批四川省博新计划，中国博士后特别资助（站前）。从事多媒体智能信息处理方向研究，在IEEE Transactions、CVPR、ICCV等国际重要期刊和会议发表论文50余篇，获CVPR2024W最佳论文，CVPR2023亮点论文，ICIP2024 最佳论文候选。申请与授权国家发明专利25项。承担科技部2030新一代人工智能重大项目、国家自然科学基金联合基金重点、重点基金、青年基金、中国博士后科学基金等10余项项目。先后荣获吴文俊人工智能科技进步一等奖、中国航天大会思源联盟优秀成果以及多项国内外学术竞赛冠军等奖励。担任CAAI智能成像专委会委员、CSIG女工委委员、CSIG多媒体专委会委员等。

Qiu Heqian, Associate Professor and Master's Supervisor at the School of Information and Communication Engineering, University of Electronic Science and Technology of China. She has been honored with nominations and selections for the CSIG Outstanding Doctoral Dissertation Award (Nominee), Sichuan Province Innovative Talent Funding Project for Postdoctoral Fellows, China Postdoctoral Science Foundation Special Grant. Her research focuses on multimedia intelligent information processing. She has published over 50 papers in internationally renowned journals and conferences, such as IEEE Transactions, CVPR, and ICCV. Her work has been recognized with multiple awards, including the CVPRW 2024 Best Paper Award, CVPR 2023 Highlight Paper, and ICIP 2024 Best Paper Candidate. She has applied for or been granted 25 national invention patents. Dr. Qiu has led or participated in more than 10 research projects, including the Scientific and Technological Innovation 2030, National Natural Science Foundation of China (NSFC) Joint Fund Key Project, Key Fund, Youth Fund, and China Postdoctoral Science Fund. Her achievements have earned her several accolades, such as the Wu Wenjun AI Science and Technology Progress Award (First Prize), the Outstanding Achievement Award of the Siyuan Alliance at the China Aerospace Conference, and championships in multiple international academic competitions. She also serves in various academic roles, including Member of the CAAI Special Committee on Intelligent Imaging, Member of the CSIG Women's Work Committee, and Member of the CSIG Special Committee on Multimedia.