Projects
Research projects I have completed and are currently working on
Embodied Intelligence Visual Perception and Motion Coordination
Multispectral Visual Information Processing Laboratory Project
Leveraging embodied intelligence visual perception capabilities, integrating large models and deep learning technologies to build an intelligent system with complex scene perception, dynamic target tracking, and motion coordination control.
Project Overview
This project belongs to the cutting-edge research direction of the Multispectral Visual Information Processing Laboratory, focusing on core technological breakthroughs in the field of embodied intelligence. It aims to solve key problems such as poor perception-motion coordination and high latency of traditional intelligent systems in complex dynamic scenarios. The project intends to take visual perception as the core, deeply integrate large models and deep learning technologies to build a complete embodied intelligence system framework with real-time perception of complex scenes, accurate tracking of dynamic targets, and multi-dimensional motion coordination control capabilities.
During the project development, we aim to focus on breaking through the technical bottlenecks of modal semantic understanding and front-end perception latency. By constructing feature distillation and feedback networks, we enhance the system's ability to understand multi-modal data while reducing perception latency. Targeting the differences in different application scenarios, we are developing scene style transfer technology to enhance the system's adaptability to complex dynamic scenes, enabling stable applications in service robots, industrial collaborative robots, autonomous driving, and other fields.
My Responsibilities
As a core R&D member, I am primarily responsible for building the feature distillation and feedback network, which is the core module for improving modal semantic understanding capabilities. Meanwhile, I will deeply participate in the research and development of scene style transfer technology, designing scene adaptation algorithms based on generative adversarial networks to enhance the robustness of the system in different environments.
In addition, I will participate in building a multi-scene dataset containing over 100,000 samples, covering various complex dynamic indoor and outdoor scenes, providing solid data support for algorithm training and verification.
Project Achievements
It is expected that the embodied intelligence system built in this project can operate with low latency and high robustness in complex dynamic scenes, improving motion response speed and environmental adaptation accuracy compared with traditional systems. In terms of data accumulation, the multi-scene dataset we aim to build will become the core data resource of the laboratory, forming key technical documents and laying a solid foundation for subsequent research.
Currently, relevant research results are being organized for submission to top conferences and journals in the field of artificial intelligence. After completion, the system will undergo preliminary landing tests in service robot and industrial collaborative robot scenarios, with potential for industrial application in intelligent manufacturing and intelligent service fields in the future.
Infrared Light Field Modal Perception Imaging
Multispectral Visual Information Processing Laboratory Project
Research on core technologies of infrared light field perception, focusing on infrared and visible light modal processing, breaking through bottlenecks such as noise interference and detail loss, providing algorithm support and technical verification for high-precision perception systems.
Project Overview
As a key research project of the Multispectral Visual Information Processing Laboratory, this project is committed to overcoming core technical challenges in the field of infrared light field perception. Infrared imaging technology has irreplaceable application value in intelligent security, industrial inspection, autonomous driving, and other fields. However, traditional infrared imaging systems generally suffer from severe noise interference, loss of detailed information, sparse data, and other issues, which greatly limit their application in high-precision perception scenarios.
The project team has conducted in-depth research on the fusion processing of infrared and visible light multi-modal data, innovatively proposing a dual-branch fusion algorithm architecture that can effectively integrate the complementary advantages of different modal data and significantly enhance feature expression capabilities. Meanwhile, targeting the problem of detail loss in infrared images, we designed a Transformer-based detail extraction module that accurately captures subtle features in images through attention mechanisms, achieving effective preservation and enhancement of infrared image details.
My Responsibilities
As a core R&D member of the project, I am primarily responsible for the research, development, and optimization of the dual-branch fusion algorithm. During the algorithm design process, I drew on cross-modal attention mechanisms to design an adaptive feature fusion module, enabling the system to automatically adjust modal weights according to different scenes, which greatly improved the feature expression ability of multi-modal data. Meanwhile, I led the design of the Transformer-based detail extraction module, effectively enhancing the detail preservation effect of infrared images by optimizing the network structure and attention calculation method.
In addition, I deeply participated in the construction of a million-level light field dataset covering paired infrared and visible light data under different scenes and illumination conditions, providing sufficient data support for algorithm training and verification. In the algorithm landing phase, I was responsible for deploying the optimized algorithm to embedded platforms, ensuring real-time processing performance of the algorithm on embedded devices through model compression and operator optimization to meet the requirements of practical application scenarios.
Project Achievements
This project has achieved significant academic and technical results. Relevant research findings have been published in the IEEE TCCF journal, and another paper is being submitted to top conferences in the field of computer vision. Technically, the perception accuracy of the prototype system we developed in complex scenes has increased by 35% compared with traditional methods, and noise suppression capability has improved by 40%, with all indicators reaching advanced industry levels.
Based on the research results of this project, my graduation design was awarded the Excellent Graduation Design of Soochow University 2025 (top 2%). Currently, the system has completed landing tests in intelligent inspection, industrial defect detection, and other scenarios, verifying its effectiveness and stability in practical applications, and laying a solid foundation for the industrial application of high-precision infrared perception systems.
XinmaiTong - Continuous Blood Pressure Monitoring System Based on Dual-Modality
National College Student Innovation and Entrepreneurship Training Program Project
Developed a high-precision, non-invasive continuous dynamic blood pressure monitoring system integrating ECG and PPG dual-modal monitoring technologies, building an embedded blood pressure monitoring model and supporting "XinmaiTong" software.
Project Overview
The XinmaiTong project is a funded project of the National College Student Innovation and Entrepreneurship Training Program, aiming to solve the pain points of traditional blood pressure monitoring equipment such as requiring manual operation, inability to achieve continuous monitoring, and poor invasive detection experience. The project team developed a high-precision, non-invasive continuous dynamic blood pressure monitoring system that innovatively integrates electrocardiogram (ECG) and photoplethysmography (PPG) dual-modal monitoring technologies, enabling 24-hour uninterrupted blood pressure data collection and analysis.
The entire system consists of hardware and software components. The hardware end adopts a low-power embedded design, integrating high-precision biological signal sensors to accurately collect human ECG and PPG signals; the software end developed a supporting "XinmaiTong" APP that can not only display blood pressure data in real-time but also has functions such as data storage, trend analysis, and abnormal warning. By building a machine learning-based blood pressure monitoring model, the system can accurately extract blood pressure features from collected biological signals to achieve non-invasive continuous blood pressure monitoring, providing important data support for the early prevention and intervention of cardiovascular diseases.
My Responsibilities
As a core member of the project, I was primarily responsible for the research and optimization of dual-modal monitoring algorithms. During the algorithm R&D phase, I collected and organized a large amount of clinical data, extracted key features from ECG and PPG signals through feature engineering methods, and participated in designing a fusion model based on random forests and neural networks, effectively improving the accuracy and stability of blood pressure monitoring.
During the hardware development phase, I assisted the team in completing the debugging and verification of embedded hardware, focusing on testing the sensor data collection module. By optimizing the sampling frequency and filtering algorithms, I ensured the accuracy and stability of biological signal collection. Meanwhile, I participated in the functional design and testing of the "XinmaiTong" APP, responsible for the development of the data visualization module to ensure users can intuitively view and understand blood pressure data. In addition, I was responsible for organizing and analyzing part of the experimental data, verifying the effectiveness of the algorithm through statistical analysis methods, providing important basis for algorithm iteration.
Project Achievements
Research results of this project have been published as an EI conference paper, which elaborates on the design ideas and experimental verification results of the dual-modal blood pressure monitoring algorithm in detail. Technically, the dual-modal blood pressure monitoring model we built performed excellently in clinical tests, with monitoring errors of systolic and diastolic blood pressure controlled within ±3mmHg and ±2mmHg respectively, and relevant indicators reaching advanced industry levels.
The project team has completed the development of a prototype machine and preliminary clinical testing. Test results show that the system can stably achieve 24-hour continuous blood pressure monitoring with a data accuracy rate of over 95%. Currently, we are advancing the industrialization of the product, conducting cooperation negotiations with multiple medical institutions and technology enterprises, with potential to transform this technology into medical health products for ordinary consumers, providing new solutions for the prevention and management of cardiovascular diseases.
Intelligent Environmental Interaction System Based on Speech and Image
Provincial College Student Innovation and Entrepreneurship Training Program Project
This project evolved through multiple stages, starting with an intelligent waste sorting cart, developing into an intelligent shooting mobile device and motion target tracking system, and later expanding to the direction of intelligent logistics robots. Integrating image and speech recognition, IoT technologies to achieve full-scenario intelligent interaction.
Project Overview
As a Provincial College Student Innovation and Entrepreneurship Training Program Project, this project lasted nearly two years and experienced technological evolution from single function to multi-scenario application. It is a comprehensive R&D project integrating mechanical design, embedded development, artificial intelligence algorithms, and IoT technologies. The core goal of the project is to build an intelligent environmental interaction system based on speech and image recognition to achieve intelligent and autonomous interaction between equipment and the environment.
This phase focused on the intelligent needs of waste sorting, developing a fully intelligent waste cart system. The system integrates image recognition and speech recognition modules, capable of automatically identifying different types of waste and completing classified disposal, with functions such as full-load packaging and fixed-point delivery. To achieve remote management, we built an IoT monitoring platform, allowing managers to view equipment status in real-time and control equipment operation through a mobile APP, solving the problems of traditional waste sorting relying on manual labor and low efficiency.
Based on the technical foundation of waste sorting equipment, we expanded the research direction to intelligent shooting, developing an intelligent cart photography assistant. This device has functions such as gesture-guided movement, fixed-track panoramic shooting, and automatic positioning group photo shooting, capable of adjusting shooting position and angle according to user gesture commands to meet shooting needs in different scenarios. Technically, we focused on optimizing the three modules of image processing, servo control, and object tracking of the motion system, achieving accurate recognition and tracking of targets in real-time image streams.
Based on technical accumulation from the previous two phases, we began exploring the R&D direction of intelligent logistics robots, applying object tracking and autonomous navigation technologies to logistics scenarios. The aim is to develop an intelligent logistics robot capable of independently completing cargo handling, path planning, and obstacle avoidance navigation, laying a technical foundation for subsequent industrial applications.
My Responsibilities
As the project captain, I was fully responsible for the overall management and core technology R&D of the project, leading a 5-person team to complete the full-cycle R&D of the project. In the early stage of the project, I led the literature review and market research, conducted in-depth analysis of the market demand and technical status of intelligent interaction equipment, formulated a detailed project R&D plan and task division plan to ensure the project progressed as scheduled.
In the mid-stage of the project, I focused on the design of the embedded system and algorithm development of the speech recognition module. According to the product requirements of different stages, I designed adaptive hardware circuits and control programs, optimized the accuracy and response speed of the speech recognition algorithm, enabling the system to achieve stable speech interaction in complex environments. Meanwhile, I coordinated team members to complete mechanical structure design and image recognition algorithm development, ensuring technical connection and system integration of each module.
In the late stage of the project, I was responsible for building the IoT module and UI design of the APP, developing a cloud platform-based equipment monitoring system to achieve real-time monitoring and remote control of equipment status. In addition, I was responsible for debugging the mechanical structure and system joint debugging, solving various technical problems during system integration to ensure the project successfully achieved the expected goals. Throughout the project cycle, I regularly organized team meetings to coordinate and solve technical problems in the R&D process, ensuring the smooth delivery of multi-stage R&D tasks.
Project Achievements
This project has achieved remarkable results. Both the scientific research plan project and extracurricular scientific research fund project were successfully concluded with an "Excellent" rating, fully verifying the technological innovation and application value of the project. In terms of intellectual property rights, the project applied for and accepted 1 utility model patent covering the core structural design of intelligent interaction equipment; meanwhile, 2 software copyrights were obtained, corresponding to the waste sorting monitoring system and intelligent shooting control system respectively.
In terms of academic achievements, based on technical accumulation during R&D, the project team published 2 EI conference papers elaborating on the design ideas and experimental results of the intelligent interaction system in detail. Technically, the project achieved a technological leap from static scene recognition to dynamic object tracking, with image recognition accuracy increased to 98% and speech recognition response time reduced to less than 0.5 seconds, laying a solid technical foundation for the subsequent R&D of intelligent logistics robots.
Through the R&D of this project, the technical capabilities and project management capabilities of team members have been significantly improved. The developed prototype equipment has received positive feedback in pilot applications on campus and in communities, showing high application value and market potential.
Information Processing Laboratory Project
Phase Research Assistant
As a research assistant in the laboratory, I was exposed to and participated in scientific research projects to accumulate experience. Mainly involved in the R&D of dormitory access control system based on voiceprint recognition and speech recognition, as well as diode simulation research project.
Project Overview
From May 2022 to August 2023, I joined the Information Processing Laboratory as a research assistant, systematically exposed to and participated in two core research projects of the laboratory. This was my first formal participation in scientific research work, accumulating valuable experience for subsequent project R&D.
This project is a Provincial College Student Innovation and Entrepreneurship Training Program Project, aiming to solve problems such as low security and inconvenience of traditional dormitory access control systems. The project innovatively integrates voiceprint recognition and speech recognition technologies to build an intelligent access control system based on biometric features. Compared with traditional password and card-swiping access control, this system has higher security and convenience, effectively preventing problems such as proxy check-in and card loss, while supporting extended functions such as remote authorization and visitor management.
This project is a research topic hosted by the supervisor, focusing on the performance research and application exploration of new diode structures. The project uses simulation methods to study the influence of different structural parameters on the electrical performance of diodes, providing theoretical basis for the design and optimization of new semiconductor devices. The research content covers multiple aspects such as volt-ampere characteristics, switching characteristics, and temperature characteristics of diodes, involving multiple technical fields such as semiconductor physics, device simulation, and data processing.
My Responsibilities
In the access control system project, I was mainly responsible for programming debugging and structural design. For the algorithm optimization of the voiceprint recognition module, I wrote a large number of test codes, improving the accuracy of voiceprint recognition by comparing different feature extraction methods and classification algorithms. Meanwhile, I participated in the mechanical structure design of the access control equipment, optimizing the size and installation method of the equipment according to practical application scenarios to ensure the practicality and stability of the system. In addition, I was responsible for the development of the supporting APP, implementing functions such as access control permission management and record query to ensure the complete operation of the system.
In the diode simulation research project, my main work focused on background sorting and data processing. I systematically reviewed relevant domestic and foreign research literature, sorted out the research status and development trends of new diode structures, and formed a detailed literature review report. During the simulation experiment phase, I was responsible for collecting and organizing experimental data, conducting descriptive statistical analysis on simulation results using statistical methods, extracting key characteristic parameters, and providing important data support for the research topic.
By participating in these two projects, I systematically improved my capabilities in literature retrieval, information inquiry, and data processing, mastered basic methods of embedded system development and semiconductor device simulation, cultivated scientific research thinking and working methods, laying a solid foundation for subsequent participation in more complex scientific research projects.
Project Achievements
Regarding the access control system project, this Provincial College Student Innovation and Entrepreneurship Training Program Project has been successfully concluded, completing the expected R&D goals. The developed prototype system performed stably in tests in laboratories and pilot dormitories, with a voiceprint recognition accuracy rate of over 95%, verifying the feasibility of the technical solution.
Regarding the diode simulation research project, the research work I participated in provided important support for the smooth progress of the research topic. Relevant research results have been published as an EI conference paper, which analyzes in detail the influence rules of different structural parameters on diode performance, providing theoretical reference for the design of new diodes.
More importantly, through this phase of research assistant work, I accumulated practical experience in embedded system development and semiconductor device simulation, cultivated the ability to independently solve technical problems and scientific research literacy. These experiences and capabilities laid a solid foundation for my subsequent hosting and participation in higher-level scientific research projects.