About

Robert Hardin is an Associate Professor in the Department of Biological and Agricultural Engineering at Texas A&M University. He holds a B.S. degree in Biological and Agricultural Engineering and Animal Science from North Carolina State University, obtained in 2002. He further earned an M.S. and a Ph.D. in Biological and Agricultural Engineering from Texas A&M University, completed in 2004 and 2009 respectively. His areas of expertise include cotton engineering, postharvest processing, precision agriculture, automation, and control. Dr. Hardin's research focuses on advancing agricultural engineering practices, particularly in the context of crop management and processing technologies. He is actively involved in the Texas A&M AgriLife programs, contributing to research and extension efforts aimed at improving agricultural productivity and sustainability.

Research topics

• Computer Science
• Biology
• Agronomy
• Mathematics
• Artificial Intelligence
• Algorithm
• Agricultural engineering
• Remote sensing
• Economics
• Statistics
• Geography
• Engineering
• Ecology

Selected publications

• Soil background effects on UAS and proximal remote sensing‐derived vegetation indices

  Agronomy Journal · 2026-01-01

  articleOpen accessSenior author

  Abstract Exposed soil, due to low vegetation cover or in open canopy crops, influences scene reflectance derived from remotely sensed data. An experiment was conducted in College Station, TX, to investigate the potential of six unmanned aerial systems (UASs)‐derived and proximally sensed vegetation indices (VIs) in suppressing soil background brightness of four treatments in 2020 and 2021. The treatments were dry soil, dry soil with winter wheat ( Triticum aestivum L.) crop residue, wet soil (WS), and wet soil with winter wheat crop residue (CRWS) in 2020. In 2021, WS and CRWS were replaced with dry sand and dry compost (DC). The VIs were calculated from remotely sensed data of treatment plots. Cotton ( Gossypium hirsutum L.) canopy cover (%) on different dates of UAS flight was extracted using unsupervised classification. Factors such as shadows, crop residue, soil moisture, and uneven canopy growth influenced the scene reflectance. The shadow on the soil decreased the soil background reflectance to <10%. Soil background variations minimally impacted the UAS‐derived VIs. Soil wetness resulted in higher normalized difference vegetation index (NDVI) than dry treatment plots at an estimated mean canopy cover > 30% in 2020. Similarly, higher NDVI was observed for DC treatment plots at an estimated mean canopy cover of <35% in 2021. The perpendicular vegetation index was least influenced by canopy cover or soil background variations. The study suggests that UAS can be used for large‐scale research without being affected by soil variability when vegetation cover is above 30%.

  PublisherOA PDFDOI

• A Data-Driven Image Extraction and Analysis Pipeline for Plant Phenotyping in Controlled Environments

  bioRxiv (Cold Spring Harbor Laboratory) · 2026-02-27

  articleOpen access

  Abstract Advances in automation, imaging, and artificial intelligence have enabled researchers to capture large volumes of high-quality plant data for understanding crop growth, stress, and genotype-by-environment interactions. While genomics has achieved remarkable throughput, phenotypic data acquisition remains a critical bottleneck for accelerating crop improvement and biological discovery. To address this challenge, an integrated multispectral phenotyping framework was developed using imagery from the Texas A&M AgriLife Precision Automated Phenotyping Greenhouse, a fully controlled facility designed for reproducible plant monitoring throughout the entire growth cycle of most crops. The framework expands the Plant Growth and Phenotyping (PGP v2) dataset and establishes a standardized system for continuous image acquisition, segmentation, deep feature extraction, and temporal analysis across multiple crop species. The project was organized around five coordinated areas: Administration and Coordination, Imaging and Sensor Operations, Data Processing and Management, Artificial Intelligence and Analytics, and Plant Science and Discovery. This structure ensured consistent data quality, version-controlled workflows, and communication across disciplines. The analytical pipeline integrates pseudo-RGB generation, deep learning–based detection and segmentation, image stitching, and temporal (longitudinal) tracking to isolate individual plants and analyze changes in morphology, spectral reflectance, and texture over time. Beyond technical innovation, the framework provides a replicable model for interdisciplinary collaboration and administrative integration in plant phenomics. The combined dataset, workflow, and management framework enable scalable, reproducible, and data-driven plant science research that bridges engineering and biological discovery. Plain Language Summary Temporal imaging of plants in controlled environments helps scientists better understand growth and biological processes. However, analyzing large volumes of images has been limited by a lack of automated tools. Multispectral imagery captures additional information about plant pigments, structure, and stress beyond standard color images. We developed an automated analysis pipeline that identifies individual plants, tracks their growth over time, and measures traits such as height, area, shape, texture, and vegetation indices. Using artificial intelligence, the system efficiently processes thousands of images to provide consistent and repeatable measurements. By integrating engineering and plant biology, this work supports data-driven decisions for crop improvement and agricultural research.

  PublisherOA PDFDOI

• Integrating Machine Learning and Path Planning for UAS-based Weed Recognition and Site-specific Management in Turfgrass Systems

  Weed Science · 2026-03-26

  articleOpen access

  Abstract Annual bluegrass ( Poa annua L.) is an extremely problematic weed in turfgrass, posing a significant challenge for turfgrass management. Injudicious use of herbicides for controlling this weed has led to resistance issues and environmental concerns. Site-specific weed control offers an opportunity to achieve effective weed control with less herbicide use, but it requires the development of a pipeline for weed detection and localization, and a path planning algorithm. To achieve this, unmanned aerial system (UAS) based RGB imagery of P. annua plants in bermudagrass turf was collected at different weed growth stages at two locations in Texas: Deer Park and College Station. A CNN (YOLO11) and a transfer (RTDETRD) model were evaluated for weed detection. The results showed that the YOLO11n model achieved the highest F1-score (0.64) and mAP@0.50 (0.68), while the RTDETRD-x model achieved the lowest F1-score (0.52) and mAP@0.50 (0.51). The geo-transformation function transforms image coordinates into a world coordinate system with centimeter-level accuracy (mean error =1.5 cm). However, the precision of the transformation depends on the quality of the orthophoto georeferencing. Additionally, the path planning algorithm showed a significant reduction (37.7%) in travel distance compared to the original weed-model-derived distance. The research highlighted the potential of UAS-based imagery for weed detection and localization in turfgrass. Further improvements are needed to enhance model performance by modifying the model architecture (e.g., input image size, hyperparameters) and evaluating its robustness across different weed growth stages and turfgrass species.

  PublisherOA PDFDOI

• Low-Cost, Compact Mobile Robot for Autonomous Soil Monitoring in Crop Fields

  2025-06-30 · 2 citations

  article

  This paper presents the development and evaluation of a mobile robotic platform for autonomous crop field scouting and soil sensing. The system combines a durable commercial chassis kit with custom 3D-printed casings, enabling reliable operation across diverse outdoor field environments. The robot features encoder-controlled motors and a swivelmounted front frame, allowing versatile and agile navigation through narrow crop rows and uneven terrain, as demonstrated in field trials conducted in cotton and peanut fields. A soil sensing mechanism, driven by a 360° servo motor and employing a linear gear-and-rack mechanism, enables consistent soil penetration. Integrated with a low-cost 7 -in-1 soil sensor, the platform provides real-time mapping of key soil parame-ters-nitrogen, phosphorus, potassium, electrical conductivity, pH, temperature, and moisture-to support data-driven farm management decisions. Preliminary experiments evaluated the robot's field navigation and soil sensing performance. Results demonstrate the potential of the platform for low-cost, mobile soil sensing, while also highlighting limitations in the current sensor's accuracy.

  PublisherDOI

• Interventions for weed seed management in cotton around the harvest time

  Weed Technology · 2025-01-01

  articleOpen access

  Abstract Late-season escapes of Palmer amaranth and waterhemp (both are Amaranthus species) pose a significant challenge in cotton production due to their high fecundity, herbicide resistance, and ability to replenish the weed seedbank at harvest. While harvest weed seed control (HWSC) has been adopted in grain systems, its feasibility in cotton remains unknown due to differences in cotton harvesting equipment design. Therefore, this study aimed to determine the fate of Amaranthus spp. seeds during harvest with cotton pickers and stripper harvesters, and evaluated the efficacy of an impact mill to destroy a range of weed seeds present in different types of cotton debris. Along with the seed cotton, cotton strippers removed 52% of the Amaranthus seeds, compared with just 7% with pickers, which are then cleaned at the cotton gin. About 85% of the seeds were retained on the plant after harvest by the pickers, and about 15% by the strippers. Seeds shattered to the ground accounted for 8% with pickers and 18% with strippers. Additionally, the cotton stripper’s field cleaner mechanism removed 15% of the weed seeds. Together, seeds collected in seed cotton, retained on the plant, or separated by field cleaners (in strippers) represent points for HWSC implementation. Different types of cotton debris were then run through a stationary weed-seed impact mill with a known number of seeds for seven weed species to determine seed destruction efficacy. The stem debris had a 29% moisture content, which is too high for the impact mill and caused mill clogging; however, seed kill levels of 98% were achieved in bur debris and gin debris types, values similar to those reported in grain systems. Together, these findings provide a framework for incorporating HWSC practices into cotton, offering growers and processors a way to reduce weed seedbank inputs.

  PublisherOA PDFDOI

• Extracting Features from Oblique Ground-Based Multispectral Images for Monitoring Cotton Physiological Response to Nitrogen Treatments

  Remote Sensing · 2025-02-08 · 1 citations

  articleOpen accessSenior author

  Early detection of nitrogen deficiency in cotton requires timely identification of stress symptoms like leaf chlorosis (yellowing) and canopy stunting. Chlorosis initially appears in older, lower-canopy leaves, which are often not visible in conventional nadir-looking imaging. This study investigates oblique ground-based multispectral imaging to estimate plant height and capture spectral details from the upper (UC) and lower (LC) cotton canopy layers. Images were collected from four camera pitch and height configurations: set 1 (30°, 2 m), set 2 (55°, 2 m), set 3 (68°, 3 m), and set 4 (70°, 1.5 m). A pre-trained monocular depth estimation model (MiDaS) was used to estimate plant height from aligned RGB images and an empirically derived tangential model corrected for perspective distortion. Further, the lower and upper vertical halves of the plants were categorized as LC and UC, with vegetation indices (CIgreen, CIrededge) calculated for each. The aligned images in set 1 had the best sharpness and quality. The plant height estimates from set 1 had the highest correlation (r = 0.64) and lowest root mean squared error (RMSE = 0.13 m). As the images became more oblique, alignment and monocular depth/height accuracy decreased. Also, the effects of perspective and object-scale ambiguity in monocular depth estimation were prominent in the high oblique and relatively low altitude images. The spectral vegetation indices (VIs) were affected by band misalignment and shadows. VIs from the different canopy layers demonstrated moderate correlation with leaf nitrogen concentration, and sets 2 and 3 specifically showed high and low differences in VIs from the UC and LC layers for the no and high-nitrogen treatments, respectively. However, improvements in the multispectral alignment process, extensive data collection, and ground-truthing are needed to conclude whether the LC spectra are useful for early nitrogen stress detection in field cotton.

  PublisherOA PDFDOI

• UAS‐derived vegetation indices detect wheat leaf rust infection and its influence on grain yield and canopy temperature

  Crop Science · 2025-04-27 · 5 citations

  articleOpen accessSenior author

  Abstract Leaf rust is a major biotic factor affecting wheat yield globally. However, the visual scoring technique to assess fungal disease in breeding programs requires significant expert manual labor and time. Unmanned aerial systems have the potential to scan large acreage in a short time for disease screening. An experiment was conducted at College Station and Castroville, TX, in 2018–2019 and 2019–2020 to assess the performance of normalized difference vegetation index (NDVI), normalized difference red edge index (NDRE), and green chlorophyll index (GCI) in detecting leaf rust infection. Other measurements included proximal canopy temperature, grain yield, and visual screening for infection type and severity. A significant positive relationship ( p < 0.001; R 2 = 0.42–0.62) of grain yield with all three vegetation indices (VIs) was observed in mid‐April 2019 at College Station. At College Station, the highest leaf rust severity coincided with the senescence stage in mid‐April 2020. No relationship between the VIs and grain yield was observed. In mid‐April 2020, when the leaf rust infection was high, the VIs showed a significant negative relationship ( p < 0.05; R 2 = 0.27) with grain yield at Castroville. All three VIs showed a significant linear negative relationship with canopy temperature at College Station ( p < 0.05; R 2 = 0.3–0.34) and Castroville ( p < 0.001; R 2 = 0.52–0.54) in mid‐April 2020. At high leaf rust severity, the repeatability of GCI was less than NDVI and NDRE at both locations in 2019 and 2020. These results may differ if multiple factors affect winter wheat simultaneously.

  PublisherOA PDFDOI

• Integrating Machine Learning and Path Planning for UAS-based Weed Recognition and Site-specific Management in Turfgrass Systems

  SSRN Electronic Journal · 2025-01-01

  preprintOpen access
  PublisherDOI

• Radiometric calibration of UAV multispectral images under changing illumination conditions with a downwelling light sensor

  The Plant Phenome Journal · 2024-10-16 · 11 citations

  articleOpen access

  Abstract Unmanned aerial vehicle (UAV)‐acquired multispectral images commonly suffer from radiometric inaccuracies due to changing illumination produced by intermittent cloud cover. This study addressed cloud shadow effects by integrating direct irradiance measurements from a downwelling light sensor into reflectance calculations. Two reflectance calibration methods were proposed as follows: (1) use of a single calibration reference panel for all images (Method 1) and (2) employment of a minimum distance classifier to assign color‐graded in‐field reflectance calibration targets to individual images based on their proximity in irradiance levels (Method 2). Images were acquired from 30 and 75 m flight altitudes with fixed‐exposure and auto‐exposure settings. Conventional photogrammetric calibration of UAV multispectral images inadequately addressed cloud shadows, resulting in orthomosaics with poor to moderate spatial uniformity ( R 2 = 0.70–0.92) and high mean absolute percentage error (MAPE = 13.52%–49.18%). The proposed calibration methods improved radiometric accuracy, yielding average R 2 = 0.99 and 0.98 at 30 and 75 m images, respectively. Method 1 showed superior performance on the red‐edge and near infrared bands, and Method 2 worked best in the blue, green, and red bands. Post‐processing empirical line calibration of orthomosaics from Methods 1 and 2 reduced MAPE in reflectance estimation by at least 50% compared to conventional calibration. The clear sky reference histograms had lower Jensen–Shannon distance, higher Pearson's correlation, and improved intersection ratios with the histograms from the proposed methods, implying high similarity. Vegetation indices calculated with the proposed methods closely matched those from the reference orthomosaic, exhibiting significantly lower MAPE than conventionally calculated vegetation indices.

  PublisherOA PDFDOI

• Selection of appropriate multispectral camera exposure settings and radiometric calibration methods for applications in phenotyping and precision agriculture

  The Plant Phenome Journal · 2024-09-27 · 5 citations

  articleOpen access

  Abstract Exposure time and gain are camera‐related parameters that affect the radiometric accuracy of unmanned aerial vehicle (UAV)‐based multispectral images used in quantitative precision agriculture. This study quantified the potential radiometric errors from the conventionally used autoexposure settings and compared the agronomic implications of fixed exposure and autoexposure settings. The exposure and gain in the auto‐exposed UAV images varied scene‐to‐scene based on the reflectance range of objects on the ground. Hence, to capture multispectral images with fixed exposure, the ideal exposure ranges were determined to prevent loss of spectral detail from overexposure or underexposure of the canopy and soil. Reflectance measurements from the fixed exposure orthomosaic had a higher coefficient of determination ( R 2 = 0.97–0.99) and lower mean absolute percentage error (MAPE = 3.07%–5.97%) than those from autoexposure ( R 2 = 0.79–0.96 and MAPE = 7.42%–25.06%), indicating better radiometric uniformity and accuracy, respectively. Calibrating images with reflectance targets captured with different exposure settings resulted in MAPE < 5% for the blue, green, red, and near infrared bands and <7% for the red‐edge band when exposure settings were within the ideal ranges; outside of those ranges MAPE increased exponentially. These observations highlighted the challenges in appropriately calibrating canopy and soil reflectance values subjected to exposure setting variabilities. Finally, early‐season plant nitrogen uptake (g/m 2 ) from 2 years was estimated somewhat better with vegetation indices derived with fixed exposure ( R 2 = 0.32–0.40 and MAPE = 13%–14%) than with autoexposure ( R 2 = 0.00–0.19 MAPE = 15%–18%).

  PublisherOA PDFDOI

Frequent coauthors

• J. Alex Thomasson

  Mississippi State University

  26 shared

• Stephen W. Searcy

  Texas A&M University

  22 shared

• Pappu Kumar Yadav

  18 shared

• Ulisses Braga-Neto

  17 shared

• Sorin Popescu

  Texas A&M University

  17 shared

• Paul Funk

  16 shared

• Juan Enciso

  Texas A&M University

  15 shared

• Karem Meza

  Utah State University

  12 shared

Education

• B.S., Biological and Agricultural Engineering

  North Carolina State University

  2002

• B.S., Animal Science

  North Carolina State University

  2002

• M.S., Biological and Agricultural Engineering

  Texas A&M University

  2004

• Ph.D., Biological and Agricultural Engineering

  Texas A&M University

  2009