报告时间:2026年3月16日10:00-12:00
报告地点:致原楼1214会议室
邀请人:邵永红
Bruce Z. Gao, PhD
The South Carolina SmartState Endowed Chair in Biofabrication Engineering, Clemson University, Clemson, SC, USA
Abstract:
The US NSF-funded EPSCoR research program, ADAPT in SC RII Track-1 project, aims to enhance research capacity in South Carolina by leveraging artificial intelligence, life and social sciences, and bioengineering. This interdisciplinary, statewide initiative focuses on fundamental research, education, workforce development, and industry partnerships, capitalizing on South Carolina's strengths in AI applications for biomedical devices while addressing the critical need to improve healthcare delivery in the state.
We present our research as part of the ADAPT program, focusing on the development of a deep learning model for hyperspectral image classification to analyze the growth of bacteria-based biofilms, which are designed to achieve fouling prevention for underwater vehicles. This project utilized several Python libraries for image analysis and processing with deep learning. A total of 767 input files were used, with image dimensions of 1024×4096 pixels, each row representing an individual image. The images were normalized using z-score normalization to set the dataset's mean to 0 and standard deviation to 1. They were then randomly divided into an 80/20 training/validation split, with 80% allocated for training and 20% for validation, using a batch size of 256. The autoencoder employed the Adam optimizer, with a learning rate of approximately 0.0013 and a weight decay of 8.75 × 10-6.
The Whiskbroom hyperspectral imaging (HSI) method was successfully applied alongside a supervised deep learning algorithm to collect, prepare, preprocess, and classify the images based on the recorded spectra. Our preliminary data indicates that the unsupervised approach effectively performs feature extraction, image reconstruction, and clustering, allowing differentiation among hyperspectral images obtained from 550 to 850 nm that contain marine microbes, empty space, or agar solutions.
Dr. Gao’s Bio:
Dr. Bruce Gao earned his BS in Physical Electronics and Optoelectronics in 1985 and his MS in Applied Laser Physics in 1988, both from Tianjin University, China. He received his PhD in Biomedical Engineering from the University of Miami in 1999 and completed a three-year postdoctoral training in cell and tissue engineering at the University of Minnesota. Currently, he is a professor in the Department of Bioengineering at Clemson University. His long-term research goal is to understand the mechanisms by which various cell types form functional tissues. To achieve this, he focuses on microfabrication, laser cell micromanipulation, coherent light, nonlinear optics-based 3D imaging, and multiscale modeling to explore cell-cell interactions within engineered microenvironments, such as biochips. His current research projects include: 1) a single-neuron-based cell biochip for investigating developmental neurotoxicity (NIH SC INBRE); 2) electrical and mechanical coupling between cardiogenic bone marrow stem cells and cardiac cells (NIH SC COBRE); and 3) the development of a hybrid laser microbeam and microfluidics system for high-throughput tissue scaffold creation (NSF EPSCoR).
