Overall vision
SFE lab’s primary research focus is developing and applying advanced artificial intelligence, robotics, and machine vision techniques to protect food safety, improve the food qualities and increase food consumer acceptance. SFE lab also has broad interests in applying these engineering solutions to areas related to human daily lives, including agriculture and healthcare.
Active project:
1. Autonomous mobile swabbing platform for visualizing the bacterial mapping in poultry processing facility (Funded by USDA NIFA)
Motivation: In the poultry processing facility, visualizing the bacterial mapping can better guide the sanitization protocols and understand pathogen transmission patterns. The autonomous mobile swabbing platform can quickly screen the facility in a high-throughput, and routine manner, which can effectively improve food safety and current protocols to meet the stakeholder needs.
Keywords: Robotic arm manipulation, UGV, SLAM, biomapping
2. Visual-tactile sensor guide dual-arm controlling for high throughput chicken picking and loading (Funded by NSF NRI 3.0/USDA NIFA)
Motivation: In the poultry supply chain, the poultry processing industry plays an important role in preparing disinfected and marketable chicken and value-added chicken products. During the COVID19 pandemic, the poultry industry is suffering from unprecedented challenges of the labor force, food safety, and supply chain robustness, which motivates this proposal to seek alternative smart and automated solutions to the existing workflow for meeting the industrial long-term needs
Keywords: Dual-arm manipulation, High throughput high resolution 3D sensing and understanding, multimodal sensor based robotic controlling, imitation learning.
3. Develop illumination robust computer vision model for agricultural applications (Funded by NSF DART)
Motivation: Most of the current imaging studies rely on human perceptions as the ground truth to train and evaluate model performances. However, in many real-world cases, human perception is not always reliable, and people may not be able to believe what they have seen. In these situations, mathematic models built upon simple
human perceptions cannot accurately reflect essential characteristics of items. Thus, there is an urgent need to under illumination effect on human and computer perceptions to build reliable deep learning models to prevent ulterior motives in manipulating model functionalities.
Keywords: Illumination estimation, Trustworthy AI, human perception
4. Quantitatively understand spectral signals for reliable bioproduct quality and nutrition estimation (Funded by USDA AFRI, UADA, SRSFC)
Motivation: Hyperspectral imaging techniques has been widely applied in remote sensing and precision agricultural applications. Transferring this non-invasive method from a qualitative analysis tool to a reliable quantitative tool is expected to expand its applications scopes.
Keywords: 1D-deep learning based spectral regression, active learning, transfer learning