Hodeghatta, Ph.D - Research

My research focuses on the following areas:

Ethical and Responsible AI – Exploring frameworks and methodologies to ensure fairness, transparency, and accountability in AI systems.
AI in Precision Agriculture – Leveraging satellite imagery and GIS data to enhance crop monitoring, resource optimization, and sustainable farming practices.
Natural Language Processing (NLP) – Developing models for understanding and generating human language, with applications across various domains.
Cybersecurity and AI – Applying AI techniques to strengthen cybersecurity measures, detect threats, and enhance resilience in digital systems.

Below is an overview of my current research work

Removal of cloud noise in satellite imagery data using Artificial Intelligence(AI)

This research project addresses the complex challenge of cloud noise in satellite imagery employing advanced artificial intelligence models for detection and removal. Our investigation spans multiple phases, beginning with a baseline examination of image classification models like LightCSPNet and progressing to transformer-based models such as Vision Transformer (ViT), CLIP, and Align. The study focuses on their advantages and limitations in cloud and cloud-shadow detection, leveraging datasets like SPARCS, Landsat 8, and Band 9 satellite data for comprehensive evaluation.

Building upon successful model development, the project extends toward practical deployment on small satellites. Two main objectives guide this phase: optimizing models for image preprocessing tasks to enhance efficiency in small satellite systems, and fine-tuning models using data from the Maine Space Complex small satellite for adaptation to regional characteristics. Beyond technical advancements, our research holds broader impacts, including improved environmental monitoring, conservation efforts, and community engagement. The models, equipped with effective cloud noise removal, enable answering broader research questions. Specifically, the project explores the effects of global warming in coastal areas through image processing and object detection tasks.

The deployment of our models will contribute to real-time environmental monitoring, supporting conservation initiatives, and aiding decision-makers in coastal management. Additionally, educational outreach initiatives aim to inspire future generations to space technology, artificial intelligence, and environmental science. In summary, our project is a multidimensional endeavor that combines cutting-edge technology with practical applications, environmental stewardship, and societal engagement, aiming to make a lasting positive impact on both local communities and the broader scientific community.

Research Student: Emmanuel Asante, CPS, Roux Institute (click more)

HyperSpectral Image (HSI) Segmentation:

The objective of this project is to analyze hyperspectral data using advanced models like the Vision Transformer (ViT). ViT utilizes the transformer architecture for image classification purposes, offering simplicity, excellent performance, and scalability. By segmenting hyperspectral images using such models, we can accurately assess crop health, identifying areas that are healthy, water-deficient, or lacking essential nutrients.

Hyperspectral imaging (HSI) is a powerful technique that captures detailed spectral information across numerous wavelengths, enabling the identification and differentiation of materials based on their chemical compositions. In agriculture, HSI serves as a unique signature to assess crop health by detecting specific spectral reflectance patterns associated with various plant conditions. For instance, changes in a plant's foliar chemistry due to stress or disease produce distinct spectral signatures, which can be detected through hyperspectral imaging.

In summary, integrating hyperspectral imaging with state-of-the-art models like ViT enhances our ability to monitor and manage crop health effectively, leading to more informed agricultural practices and improved crop yields.

Research Student: Seaqueue Cheng (click more)

Autonomous UAVs for crop monitoring in the state of Maine

Climate change is affecting crops and forest ecosystems in unprecedented ways. To adapt to changing ecosystems, we need to continuously monitor crops and forests with high resolution images and ground data. Traditionally, satellites have been used to collect remote-sensing data evaluating crop health. However, satellite based remote-sensing data has been found to be sparse and often obstructed by cloud cover. More recently, manually-controlled UAVs have been used to collect high-resolution data about crop health to complement satellite data. However, manually-controlled UAVs limit the amount of data that can be captured with the available manpower. Over the long term, we intend to automate the data-collection, data-analysis and recommendation generation pipeline to save manual resources.

Capturing the spatial heterogeneity of crop health and resource needs (e.g. water, pesticides, and fertilizers) is the prerequisite for precise management and for improving the on-farm resource use efficiency. Wild blueberry fields show high spatial heterogeneity in environmental conditions and crop performances (Barai et al. 2022), thus it provides a model system for mapping spatial variations and precise management. Our previous work has shown that thermal and near-infrared imagery-based indices could be used to predict crop water stress (Chan et al. 2021) and yield (Barai et al. 2022) in wild blueberries. Now we will expand the monitoring to include canopy nutrient concentrations and health status (disease infection). Also, using satellite data and regression analysis, we found some important factors influencing wild blueberry yield (Barai et al. 2021). More insights could be revealed by collecting higher-resolution data and using modern machine learning algorithms. (Click here)

Research Students: K. Barai, T. Trang (click here)

Publications: 16th International Conference on Precision Agriculture 21-24 July 2024 Manhattan, Kansas, United States

1. Y. Zhang, U.R. Hodeghatta, V. Dhiman, K. Barai, T. Trang, (2024, July). “Airborne Spectral Detection of Leaf Chlorophyll Concentration in Wild Blueberries”. Conference proceedings of 16th International Conference on Precision Agriculture, Manhattan, Kansas. July 21-24, 2024. (Proceedings | International Society of Precision Agriculture (ispag.org))

2. K. Barai, C. Ewanik, V. Dhiman, Y. Zhang, U.R. Hodeghatta (2024, July), “Predicting Water Potentials of Wild Blueberries During Drought Treatment Using Hyperspectral Sensor and Machine Learning”. Conference proceedings of 16th International Conference on Precision Agriculture, Manhattan, Kansas. July 21-24, 2024.( Proceedings | International Society of Precision Agriculture (ispag.org))

Download: Proceedings | International Society of Precision Agriculture (https://www.ispag.org/proceedings/?action=author_abstracts)

Data Privacy and Challenges in Building AI Models

The objective of this project is to investigate the critical issues of data privacy and bias in Artificial Intelligence (AI) models, with a specific focus on applications deployed for public use, including GIS & Satellite applications. AI systems are increasingly used across various sectors such as healthcare, finance, education, and law enforcement, where sensitive data is processed and decisions directly impact individuals. However, the rapid adoption of AI has raised significant concerns about how these models handle user data and whether they introduce or amplify biases that can lead to unfair outcomes.

Research Student: Samidha Rathore (Click more)

Presentations

Project Title: Skill-in-a-Flash Phase-IIB

Proposal for FOA NUMBER: DE-FOA-0003279 (FY 2024 PHASE II RELEASE 2)

Topic: 10-b

Company Name: iZen ai, Inc.

The lack of skilled workers at all levels is challenging the growth of the U.S. bioeconomy, and

disproportionately affecting small and rural businesses that cannot afford higher-priced labor and don’t have the means to train the new labor force on their own. The solution is a combination of a new device and framework for workforce development and promoting DEIA.

The solution, named iZen-LEAPTM, comprises two elements: (1) an eLearning platform featuring (a) “iZen LEAP-Cloud”, a cloud-based platform, and (b) “iZen-LEAP-Box”, a standalone device capable of delivering education and training programs to people without requiring the Internet, computers or infrastructure-power, on any device, anywhere, anytime. And (2) eLearning content, encompassing foundational and practitioner-level skills, tailored to specific job roles.

Learners with a computer or mobile device but no internet can connect wirelessly to iZen-LEAP-Box. Those without a computer or internet, can connect iZen-LEAP-Box to a TV or monitor and begin learning.

In Phase-I, research and development of prototypes and beta-testing were completed successfully, proving the technical feasibility and the commercialization potential.

During Phase-II, we delivered a fully operational solution equipped with essential features and the necessary level of maturity to undergo comprehensive field trials as part of the customer discovery process.

In the proposed Phase-IIB project we will build upon the achievements of the Phase-II project to further develop usability, security, field serviceability, and ecosystem integration-related features to optimize opportunities for commercialization and support the development of a diverse bioenergy workforce.

BB_cropmonitoring_UH_Oct_2024V0.3.pdf