(†: equal contribution, ~: corresponding author)

Early Project

• Intelligent Information Processing Laboratory
  3D Convolutional Neural Network-Based Spectral Unmixing for SERS Imaging.
  Tianyao Tan (Advisors: Dr.Ping Tang, Prof.Liyun Zhong)
  We have designed a 3D self-encoder algorithm based on discrete cosine transform with spatio-temporal global attention mechanism which is designed for realizing a component decomposition unmixing algorithm for scanning Raman imaging of 2D materials MoS2/WSe2.
  
  We have compared traditional machine learning NMF non-negative matrix decomposition and K-hype kernel abundance estimation algorithm with traditional and deep learning algorithms like A2SAN or VAE to realize the reconstruction of the Raman scanning imaging on SOTA level.
  Funding: This is supported by National Natural Science Foundation Project (500210181) with Dr. Ping Tang.

• [Reinforcement Learning-Driven Super-Resolution Imaging (5x Resolution)]
  ([thesis]).
  Tianyao Tan (Advisors: Dr.Ping Tang, Prof.Liyun Zhong)
  Proposed Markov Decision Process (MDP) framework to iteratively optimise system imaging to achieve super-resolution imaging from MoS2/WSe2 heterojunction at the electron microscope level, reducing imaging noise.
  Founding: This is supported by National Natural Science Foundation Project (500210181) with Dr. Ping Tang.

• Shanghai Zhangjiang Institute of Mathematics
  Diffusion Models for ISPRS Remote Sensing Data Augmentation.
  Tianyao Tan
  Training a state-of-the-art generative model based on stable diffusion on the ISPRS dataset generates synthetic images to improve semantic segmentation accuracy, comparing the generative stability of stable diffusion models with adversarial neural network GANs, integrating of diffusion-generated data into a U-Net pipeline (PyTorch), and explore the prospects of applying the stable diffusion model to different semantic segmentation networks(Swin-Unet, DeepLab v3, etc).
  

  
  

Degree Thesis

• Matlab-based GPU programming for optoelectronic image acceleration processing(graduation)
  Tianyao Tan (Advisor: Dr.Ping Tang).