日本語 English | 中文 |
Website Home Page Self-Introduction Research Introduction Participated Projects Teaching Introduction Other Content Contact Map
2024.11 Teacher Han Yuexing is Certified as CCF Distinguished Membership
Congratulations !

2024.10.22 Welcome to the research group, Cao Zhen, Chen Yiming, Li Yang, Sang Chen, Zhang Haoxuan, Zhang Wenjun, and Zhao Yinkang! We are excited to have you join our team!
Cao Zhen, male, studied both undergraduate and master's degrees at the School of Computer Engineering and Science, Shanghai University, currently majoring in Electronic Information. Funny words, positive and optimistic attitude, like nonsense and abstract things, daily hobbies are play badminton, running and exploring shops. I hope to exercise my abilities in all aspects during my graduate studies, strengthen my fellow students, and cultivate professional academic skills while keeping fit.
Chen Yiming, male, graduated with a bachelor's degree in Internet of Things Engineering from the School of Computer Science and Communication Engineering at Jiangsu University. He is currently pursuing a master's degree in Computer Science at Shanghai University. He is optimistic, outgoing, and friendly, and enjoys sports, gaming, and other activities. I hope to improve my research skills and teamwork abilities during my graduate studies, and join forces with everyone.
Li Yang, male, studied undergraduate at Wuhan University of Technology and is currently a master's student majoring in Computer Science and Technology at Shanghai University. Usually I like play badminton, and I can run when I am alone. Enjoy socializing with excellent classmates. I have a cleanliness obsession and a bit of perfectionism, wanting to do things to the best of my ability. I hope to have a good body while doing well in scientific research at a new starting point, and enjoy my graduate life with everyone.
Sang Chen, male, graduated with a bachelor's degree in Artificial Intelligence from the School of Computer Science at Shanghai University, and is currently pursuing a master's degree in Computer Science and Technology. Outgoing personality, positive and optimistic, sincere in dealing with people, and responsible. I hope to enhance my professional and practical abilities during my graduate studies, and continuously grow through learning and exploration.
Zhang Haoxuan, graduated from the Department of Computer Science at Jiangsu University with a bachelor's degree, and is currently pursuing a master's degree in Computer Science and Technology at Shanghai University. In terms of personality, I am a person who combines internal and external traits. I enjoy both communicating with others and immersing myself in my own world. I have a wide range of interests and hobbies, mainly enjoying sports such as ball and chess. I hope to explore the forefront of academia and make progress together with everyone in the coming days. I will also work hard to learn and actively contribute my own strength.
Zhang Wenjun, male, studied software engineering at the School of Computer Science and Technology, Wuhan University of Technology for his undergraduate degree, and is currently pursuing a master's degree in computer science and technology at Shanghai University. Sincere and responsible, with hobbies related to firearms, knives, bows and arrows, slingshots, FPVs, unrestricted combat, and a strong interest in software development. I hope to experience a different kind of life and explore the meaning of life during my graduate studies.
Zhao Yinkang, male, graduated with a bachelor's degree in Data Science and Big Data Technology from Changshu Institute of Technology, and is currently pursuing a master's degree in Computer Science and Technology at Shanghai University. I am positive and optimistic, and enjoy thinking. I like but am not good at ball sports such as badminton and basketball. I enjoy taking walks at night. I hope to improve my professional skills during my graduate studies and make more friends to progress together.
2024.07 Congratulations to Chi Ruting on her successful graduation!
Chi Ruting graduated with a bachelor's degree from Shanghai University and began pursuing an academic master's degree in the School of Computer Engineering and Science at Shanghai University in 2021. Since her senior year of undergraduate studies, student Chi Yuting has been studying and researching image processing related technologies and applications in the Han Yuexing research group. Under the careful guidance of Professor Han, she has continued and advanced the following research:
1. To address the issues of small sample size and complex microstructure features in material image semantic segmentation, a dual branch semantic segmentation network based on feature pyramid and cross attention is proposed. This network is divided into main branch and auxiliary branch. The main branch uses a feature pyramid model to aggregate multi-level image features to enhance detail information; The auxiliary branch uses low-level features of the backbone network to segment images, and the auxiliary network learns texture and boundary information. Under the collaboration of multi task supervision and multi-scale features, this method achieves the best performance on multiple small sample material image datasets compared to the comparative model.
2. To address the significant issue of small sample size in material image instance segmentation, a instance segmentation method based on multimodal fusion and pseudo labeling technology is proposed to improve the utilization of existing data. This method improves the accuracy of object classification, localization, and segmentation by fusing multimodal data of images and text. This paper divides training into two stages: fully supervised training and semi supervised training. During the semi supervised training phase, pseudo labeling techniques are used to involve unlabeled data in the optimization of the supervised model. Under the collaboration of multimodal information and two-stage training, compared to the comparison model, this method achieves superior performance in various small sample instance segmentation scenarios.
3. Design statistical analysis methods for the microscopic images of 2205 duplex stainless steel to explore the intrinsic relationship between material image segmentation results and material properties. This paper is based on semantic segmentation results and instance segmentation results, using material empirical formulas to efficiently and accurately obtain the growth rate and material performance changes of microstructures that are similar to manual analysis results.
After graduation, Chi Yuting joined Xiaomi to work on camera algorithms. During her three-year graduate studies at Shanghai University, student Chi Yuting worked hard and conducted research diligently, constantly enhancing her professional knowledge. She was fortunate enough to have met many good teachers and friends. I hope that student Chi Yuting will not forget her original intention, keep in mind her mission, overcome obstacles, and forge ahead on the road ahead.
Essay: Research on Small Sample Image Segmentation Method and Its Application in Material Performance Mining

2024.07 Congratulations to Wan Guanxin on his successful graduation!
Wan Guanxin graduated from Guilin University of Technology with a bachelor's degree and began pursuing a master's degree in Computer Engineering and Science at Shanghai University in September 2021. After joining the research group, he studied shape space theory and image feature enhancement related technologies and applications under Professor Han Yuexing. Under the careful guidance of Professor Han, the following research was completed:
1. Aiming at the characteristics of limited available data and insufficient diversity in small sample image scenes, a feature enhancement method based on pre shape space geodesic curves, abbreviated as FAGC-PSS (Feature Augmentation on Geodetic Curves in Pre Shape Space), is proposed. Firstly, use deep learning models to extract features from small sample images; Enhance the dimensionality of image features based on shape space theory and project them onto a pre shape space; Construct corresponding geodesic curves for each category of feature data separately; Finally, generate feature data along the optimal geodesic curve for training the image processing model. The innovation of this method includes the following three points: firstly, achieving small sample image feature enhancement, which helps the model comprehensively understand the distribution and rules of training samples, and improves the robustness and reliability of the model; Secondly, the proposed FAGC-PSS feature enhancement method can be applied to multiple downstream tasks, such as achieving good results when combined with machine learning classification models; Thirdly, designing a random probability function and influencing factors in the cross entropy loss function of small sample image classification tasks can balance the impact of generated features and image features on the model.
2. In response to the characteristics of small sample size and poor prediction accuracy of material properties in material images, this paper proposes a material property prediction method based on FAGC-PSS. This method achieves material performance prediction tasks on small sample material images by designing the downstream task framework structure of FAGC-PSS and combining it with a pseudo labeling mechanism. The specific process includes four steps: extracting material image features, generating features through FAGC-PSS, using pseudo labeling mechanism to label performance value labels for the generated features, and finally using enhanced features to train the material performance prediction model. The innovation of this method includes the following: introducing the FAGC-PSS module into the predictive material performance model to enhance the diversity and complexity of feature data; Design a pseudo labeling mechanism to annotate the generated feature data. The experimental results show that this method can demonstrate good effectiveness and universality for predicting the properties of different types of materials.
After graduation, Wan Guanxin entered Huawei company. During his three-year graduate studies at Shanghai University, Wan Guanxin worked hard to learn, participated in research projects, and demonstrated excellent programming skills and algorithm development abilities. For complex technical problems, the ability to quickly analyze and propose effective solutions demonstrates strong independent research capabilities and innovative thinking. I hope that Wan Guanxin will not forget his original intention, keep in mind his mission, move forward bravely, and create a more brilliant future on the road ahead.
Essay: Image feature enhancement based on shape space theory and its application in material performance prediction

2024.07 Congratulations to Han Sifan on his successful graduation!
Han Sifan entered the School of Computer Engineering and Science at Shanghai University in September 2021, starting his master's degree program. After joining the research group, he followed two teachers, Chen Qiaochuan and Han Yuexing, and focused on the research of deep learning based processing of material images to predict material properties. Under the guidance of two teachers, the following research content was completed:
1. Constructed a deep learning prediction material performance network based on global local feature extraction and multi feature fusion. This network adopts a dual branch multi-scale structural design, utilizing both global and local branch networks to extract global and local features from material microstructure images, without disrupting the modeling process of each feature. Integrating multi head self attention mechanism into the global branch network, dividing the feature map into multiple different subspaces, and mining the inherent correlation between features. Compared to existing methods, this network has successfully established a more complete and accurate "structure performance" mapping relationship.
2. In response to the problem of insufficient understanding of material microstructure images in complex scenes by current algorithms, which affects prediction accuracy, an efficient multimodal feature fusion network is proposed. This model includes a spectral feature extraction module, a local element feature extraction module, and a GLFS Net module for extracting material microstructure features. Through multi information fusion and the strategy of using material elements to assist in enhancing network details and material image microstructure, the network can ultimately achieve accurate prediction of material properties in complex scenes.
3. Applied for a patent for a method for predicting material properties. This patent is based on a lightweight network architecture, which comprehensively utilizes multimodal information combining images and text of materials, further improving the accuracy of material microstructure analysis and performance prediction.
After graduation, Han Sifan worked in the Zhuzhou Locomotive Research Institute of CRRC Zhuzhou Electric Locomotive Co., Ltd. to conduct research on autonomous driving. During my three years of graduate studies at Shanghai University, I worked hard to learn and also made many good teachers and friends. He has walked through many corners of Shanghai and will never forget it. He hopes to have the opportunity to meet everyone again in the future.
Essay: Research on Material Performance Prediction Based on Multi Feature Fusion

2024.04 Recent achievements of the team - using deep learning to predict the thermal conductivity of tetragonal YSZ coatings doped with Al2O3
Our team has published a paper titled 'Thermal Conductivity Prediction of Al2O3 Doped Tetragonal YSZ Coatings Using Deep Learning' in the international journal Journal of the European Ceramic Society (IF: 5.7, Chinese Academy of Sciences). The paper is authored by the School of Computer Engineering and Science at Shanghai University, with Chen Qiaochuan as the first author, Han Sifan as the second author, Song Xuemei as the third author, and Han Yuexing and Zeng Yi as co corresponding authors.

Predicting material image performance based on deep learning faces challenges such as data scarcity, inability to simultaneously extract local and global features of material images, and discovering correlations between features. In the field of materials, due to factors such as manufacturing costs and commercial protection, it is not possible to obtain data in bulk like obtaining natural scene images, which makes it difficult to directly apply deep learning models to the materials field due to insufficient data volume. On the other hand, unlike natural scene images, material images often have very fine and complex texture structures due to their own characteristics, and macroscopic performance is not only affected by local microscopic structures. The correlation between features and the interaction between structures, that is, global features, are also very important. Most existing deep learning methods often directly apply CNN models that perform well on natural scene images to the field of materials without targeted optimization. Due to the fixed size of the convolution kernel, CNN's receptive field is limited and often only extracts local features of the image, ignoring global features. Therefore, the algorithm suffers from insufficient training due to the scarcity of data and the inability to simultaneously extract local and global features, resulting in insufficient prediction accuracy and poor robustness. To address these issues, this paper proposes a dual structure feature extraction and multi-scale attention fusion network (RCFNet). This model adopts a dual branch structure of global feature extraction module and local feature extraction module, independently extracting global and local features of material images without damaging the original modeling of their respective features. By proposing a multi-scale attention fusion module (Merge), the global and local features extracted at each scale are fused, and the fusion module accumulates information from the previous fusion result. The final fused features are fed into FCNN for processing and prediction results are obtained. The following is a schematic diagram of the structure of the multi-scale attention fusion module.

For both global and local features at each scale, the SCSE attention mechanism is first employed to simultaneously stimulate important information in both channel and spatial dimensions, highlighting the most critical features and assigning significant weights. Next, global features with global semantics are further processed through channel attention mechanisms, while local features with local semantic information are further processed through spatial attention mechanisms. Subsequently, high-dimensional spatial mapping and nonlinear transformation are performed to obtain the fusion result of the current stage. The proposed method takes into account both local and global feature extraction of material images, with the Merge module performing multi-level feature fusion, multi attention continuously focusing on key features, suppressing noise information, reducing information loss, and reducing the model's dependence on a large amount of training data. In the field of material images with scarce samples, very impressive prediction results have also been obtained.
Essay: Thermal Conductivity Prediction of Al2O3-Doped Tetragonal YSZ Coatings Using Deep Learning
Our code and paper are both publicly available at: https://github.com/han-yuexing/RCFNet_Conv_Resnet50_MHSA/tree/main

2023年 Congratulations to Li Ruiqi for receiving the 2023 Shanghai Computer Society Outstanding Master's Thesis Nomination Award
描述性文本

Name: Li Ruiqi

Unit: Shanghai University

Topic: Research on Complex Texture Image Segmentation Method
Based on Incomplete Annotation

Tutor: Prof. Han Yuexing
2024 Recent achievements of the team - prediction of tensile strength of Al Si alloy based on multimodal fusion learning
Our team has published a paper titled 'Prediction of ultimate tensile strength of Al Si alloys based on multimodal fusion learning' online in the new international journal 'Materials Genome Engineering Advances' in the field of materials genetic engineering. The School of Materials Science and Engineering at Shanghai University is the first unit, the School of Computer Engineering and Science is the second unit, Zhu Longfei is the first author, Chen Qiaochuan is the third author, and Associate Professor Han Yuexing and Professor Li Qian from the School of Materials Science and Engineering are co corresponding authors of this paper.

At present, the tensile strength of Al Si alloys is mainly obtained through tensile testing, which involves sample preparation, processing, and testing, requiring professional skills and testing equipment. There are problems such as long testing cycles, high costs, and material waste. In addition, preparing standardized tensile specimens for performance testing is a challenge for components with complex shapes. Therefore, how to efficiently and accurately obtain the tensile strength of materials is currently a challenging problem. To address this issue, an innovative multimodal fusion learning framework that comprehensively considers composition and microstructure is proposed to predict the tensile strength of Al Si alloys. This work focuses on hypoeutectic Al Si alloys with wide applications. Firstly, we collected data from literature and experimental data on different modes, including alloy composition, added alloy elements, α - Al images, eutectic Si images, tensile sample size, and tensile rate. Secondly, using image processing techniques to extract microstructural feature parameters, the image is segmented and quantitatively analyzed. Then, a three-step feature screening was performed on 33 features from different modalities to obtain 12 key features. Finally, using 12 key features as inputs, four machine learning models (Decision Tree (DT), Random Forest (RF), Adaptive Boosting (AdaBoost), and Extreme Gradient Boosting (XGBoost)) were employed to construct a regression prediction model for tensile strength. The results showed that the XGBoost model performed the best among all models, achieving high accuracy in predicting tensile strength (R2=0.94, relative error less than 8.1%, absolute error less than 14.2 MPa) with limited data mainly from different literature. In addition, five mixed features (Grain size, Ti, Si, ECD, Number density) that significantly affect UTS and their critical values were identified through feature importance analysis and SHAP analysis. This work is expected to provide insights into the mapping relationship between the composition, microstructure, and properties of hypoeutectic Al Si alloys, and can be applied to other alloys.
Essay: Prediction of ultimate tensile strength of Al-Si alloys based on multimodal fusion learning

2023
2025