Responsibilities include:

• Develop machine learning models to estimate phenotypes for crop breeding, leveraging both unstructured data (e.g., remote sensing imagery, climate grids) and structured data (e.g., sensor-based tabular datasets).
• Lead end-to-end ML pipelines from model development and validation to deployment and maintenance using automated solutions.
• Explore proof-of-concept solutions for phenotype prediction using cutting-edge sensors, newest satellites, and advanced model architectures to accelerate breeding cycles.
• Actively collaborate with cross-functional stakeholders to gather model requirements, share progress, and ensure alignment with research and business objectives.

Tech Stacks: PyTorch, TensorFlow, Rasterio, GDAL, Geopandas, Pandas, BigQuery, AWS, AirFlow, Docker.

Responsibilities include:

• Informed AI for Digital Agricultural Monitoring
  • Hybridize PROSAIL Radiative Transfer Model and Deep Neural Network through transfer learning to estimate corn biophysical traits
  • Developed an automatic solar-sensor geometry calculation tool using Python from UAV-based push-broom hyperspectral sensor.
• Multi-temporal UAV Images and 3D CNN for Yield Prediction
  • Devised an automated data processing pipeline to create 30,000 sample data from UAV orthomosaics.
  • Created 3D version of AlexNet, VGG, ResNet and DenseNet architectures to estimate soybean yield.
  • Scaled up the model training from local GPU cluster to AWS Sagemaker framework.
• Seed Composition Estimation using Multi-Sensor Fusion
  • Created Raw imagery-based decision-level fusion algorithms for early-stage soybean seed composition prediction.
  • Combined 2D and 3D CNN architectures from different sensor data (multispectral/hyperspectral, thermal, LiDAR).
  • Managed multi-sensor UAV data collection efforts during multiple growth stages of soybean.
• Field-level Yield Prediction using Satellite Images
  • Modeled Soybean and Corn Yield prediction from WorldView-3 and Planet satellites at plot scale.
  • Analyzed the performance of both classical machine learning algorithms (with hand-crafted feature engineering) and CNNs (without any feature engineering).
  • Reported the superiority of 3D ResNet-18 in model performance and highlighted the importance of temporal dataset in predictive modeling.
• TERRA-REF: Reference Phenotyping Platform for Crop Breeding
  • Developed a deep neural network-based SWIR Irradiance prediction model for a Gantry-based hyperspectral system.
  • Introduced a fast and unsupervised soil removal algorithm for hyperspectral images.
  • Utilized fractional calculus and statistical machine learning to predict leaf chlorophyll content from leaf spectroscopy.

Tech Stacks: Python, Rasterio, TensorFlow, Geopandas, AWS, Sensor Calibrations, Pix4D, ArcGIS, QGIS.

UAVs and Sensors: DJI Mavic, DJI Matrice, Headwall Hypersepc VNIR and SWIR, Velodyne LiDAR, Altum Multispectral, ICI and FLIR Thermal.

Responsibilities include:

• Implemented hydrologic connectivity models using TauDEM in a High-performance computing (HPC) cluster named BigDawg.
• Created a CNN model to identify bridges/culverts from LiDAR-derived high-resolution DEMs.
• Automated stream delineation pipeline from DEMs using ArcPy.

Tech Stacks: Python, ArcPy, ArcGIS, TensorFlow, HPC, SSH.

Responsibilities include:

• Devised geospatial data processing and modeling pipelines for embankment installation around major floodplains in Bangladesh; supported engineers and local governments by providing self-explanatory GIS maps.
• Transformed traditional map making procedures into automated systems using ArcPy-based scripting.
• Analyzed satellite images for detecting spatiotemporal changes of riverbanks using eCognition, ERDAS Imagine and Google Earth Engine (Python API).

Responsibilities include:

• Prepared community level risk and vulnerability map atlas using ArcGIS.
• Conducted earthquake risk and vulnerability assessment using RADIUS and HAZUS.