1. Monitor crop disease severity with satellite image:Collaborated with disease teams across multiple regions to develop a machine learning model for estimating crop disease severity using satellite imagery2. Improved and deployed UAV hyperspectral image processing pipelineDesigned and implemented unsupervised vegetation mask and quality control metrics for processed UAV hyperspectral imagery, establishing a solid foundation for subsequent analysis3. Identified novel traits for… Show more 1. Monitor crop disease severity with satellite image:Collaborated with disease teams across multiple regions to develop a machine learning model for estimating crop disease severity using satellite imagery2. Improved and deployed UAV hyperspectral image processing pipelineDesigned and implemented unsupervised vegetation mask and quality control metrics for processed UAV hyperspectral imagery, establishing a solid foundation for subsequent analysis3. Identified novel traits for corn yield prediction from UAV image: Utilized AWS (Amazon SageMaker) to deploy machine learning models, achieving predictions of end-of-season corn yields across multiple locations with an value of approximately 0.75 with UAV hyperspectral imagery Show less

▪ Built convolutional neural network (AutoEncoder) to extract features from high-dimensional data (hyperspectral images) for biomass prediction and tassel detection with Pytorch▪ Built regression model (LSTM RNN) for biomass prediction using UAV based remote sensing data (hyperspectral images, LiDAR) with Tensorflow 2▪ Utilized machine learning methods (Support Vector Regression and so on) to classify different genotypes of maize/sorghum with limited training… Show more ▪ Built convolutional neural network (AutoEncoder) to extract features from high-dimensional data (hyperspectral images) for biomass prediction and tassel detection with Pytorch▪ Built regression model (LSTM RNN) for biomass prediction using UAV based remote sensing data (hyperspectral images, LiDAR) with Tensorflow 2▪ Utilized machine learning methods (Support Vector Regression and so on) to classify different genotypes of maize/sorghum with limited training data. Utilized PCA and manifold learning for linear/non-linear dimension reduction ▪ Developed visual analytics system for selecting the best subsets of features from high dimensional data (hyperspectral data with more than 200 bands) with flask and React.js Framework▪ Collected high-resolution, phenotypic data on energy crops with unmanned aerial vehicle. Developed a python-based hyperspectral image processing software with others. Show less

▪ Applied ArcGIS to display car crash thematic map and analyze data with linear regression model▪ Developed heatmap web page with Html+JavaScript+CSS and ECharts, use Python to clear data▪ Utilized CLIP and Earth 3D Map to correct car crash coordinates in Indiana state