Search NASA⌕ Search

SEARCH · Search NASA

Results for “Data segmentation”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 325 records · Page 18

Optimizing inference of segmentation on high-resolution images in MLExchange

MLExchange is a machine learning (ML) operations platform providing web user-interfaces (UIs) for data visualization and analysis pipelines at synchrotron facilities. Among these UIs is the segmentation app which helps synchrotron users utilize ML algorithms to automatically segment high-resolution scientific images with minimal manual annotation effort. In this work, we share code optimizations that significantly speed up the segmentation inference workflow of large data in short time. By optimizing the sequence of CPU-GPU data transfers and introducing CPU parallelization to key operations, we improve the per-device, per-image frame computational efficiency and observe close to 3×$$\times$$ speedup over the original segmentation inference workflow run time when utilizing a single GPU. Further adaptations enabling multi-GPU inference yield more than 40×$$\times$$ speedup with 100 GPUs compared to the optimized single GPU inference workflow. This acceleration of the segmentation inference workflow will provide MLExchange users with easy access to segmentation results with little wait time.

Lu, Shizhao↗

MultiTaskDeltaNet: change detection-based image segmentation for operando ETEM with application to carbon gasification kinetics

Transforming in situ transmission electron microscopy (TEM) imaging into a tool for spatially-resolved operando characterization of solid-state reactions requires automated, high-precision semantic segmentation of dynamically evolving features. However, traditional deep learning methods for semantic segmentation often face limitations due to the scarcity of labeled data, visually ambiguous features of interest, and scenarios involving small objects. To tackle these challenges, we introduce MultiTaskDeltaNet (MTDN), a novel deep learning architecture that creatively reconceptualizes the segmentation task as a change detection problem. By implementing a unique Siamese network with a U-Net backbone and using paired images to capture feature changes, MTDN effectively leverages minimal data to produce high-quality segmentations. Furthermore, MTDN utilizes a multi-task learning strategy to exploit correlations between physical features of interest. In an evaluation using data from in situ environmental TEM (ETEM) videos of filamentous carbon gasification, MTDN demonstrated a significant advantage over conventional segmentation models, particularly in accurately delineating fine structural features. Notably, MTDN achieved a 10.22% performance improvement over conventional segmentation models in predicting small and visually ambiguous physical features. This work bridges key gaps between deep learning and practical TEM image analysis, advancing automated characterization of nanomaterials in complex experimental settings.

08 HYDROGEN↗

Automated Grain Boundary (GB) Segmentation and Microstructural Analysis in 347H Stainless Steel Using Deep Learning and Multimodal Microscopy

Austenitic 347H stainless steel offers superior mechanical properties and corrosion resistance required for extreme operating conditions such as high temperature. The change in microstructure due to composition and process variations is expected to impact material properties. Identifying microstructural features such as grain boundaries thus becomes an important task in the process-microstructure-properties loop. Applying convolutional neural network (CNN)-based deep learning models is a powerful technique to detect features from material micrographs in an automated manner. In contrast to microstructural classification, supervised CNN models for segmentation tasks require pixel-wise annotation labels. However, manual labeling of the images for the segmentation task poses a major bottleneck for generating training data and labels in a reliable and reproducible way within a reasonable timeframe. Microstructural characterization especially needs to be expedited for faster material discovery by changing alloy compositions. Here, in this study, we attempt to overcome such limitations by utilizing multimodal microscopy to generate labels directly instead of manual labeling. We combine scanning electron microscopy images of 347H stainless steel as training data and electron backscatter diffraction micrographs as pixel-wise labels for grain boundary detection as a semantic segmentation task. The viability of our method is evaluated by considering a set of deep CNN architectures. We demonstrate that despite producing instrumentation drift during data collection between two modes of microscopy, this method performs comparably to similar segmentation tasks that used manual labeling. Additionally, we find that naïve pixel-wise segmentation results in small gaps and missing boundaries in the predicted grain boundary map. By incorporating topological information during model training, the connectivity of the grain boundary network and segmentation performance is improved. Finally, our approach is validated by accurate computation on downstream tasks of predicting the underlying grain morphology distributions which are the ultimate quantities of interest for microstructural characterization.

36 MATERIALS SCIENCE↗

Patient-specific radiation dose and cancer risk estimation in CT: Part II. Application to patients

Purpose: Current methods for estimating and reporting radiation dose from CT examinations are largely patient-generic; the body size and hence dose variation from patient to patient is not reflected. Furthermore, the current protocol designs rely on dose as a surrogate for the risk of cancer incidence, neglecting the strong dependence of risk on age and gender. The purpose of this study was to develop a method for estimating patient-specific radiation dose and cancer risk from CT examinations. Methods: The study included two patients (a 5-week-old female patient and a 12-year-old male patient), who underwent 64-slice CT examinations (LightSpeed VCT, GE Healthcare) of the chest, abdomen, and pelvis at our institution in 2006. For each patient, a nonuniform rational B-spine (NURBS) based full-body computer model was created based on the patient's clinical CT data. Large organs and structures inside the image volume were individually segmented and modeled. Other organs were created by transforming an existing adult male or female full-body computer model (developed from visible human data) to match the framework defined by the segmented organs, referencing the organ volume and anthropometry data in ICRP Publication 89. A Monte Carlo program previously developed and validated for dose simulation on the LightSpeed VCT scanner was used to estimate patient-specific organ dose, from which effective dose and risks of cancer incidence were derived. Patient-specific organ dose and effective dose were compared with patient-generic CT dose quantities in current clinical use: the volume-weighted CT dose index (CTDI{sub vol}) and the effective dose derived from the dose-length product (DLP). Results: The effective dose for the CT examination of the newborn patient (5.7 mSv) was higher but comparable to that for the CT examination of the teenager patient (4.9 mSv) due to the size-based clinical CT protocols at our institution, which employ lower scan techniques for smaller patients. However, the overall risk of cancer incidence attributable to the CT examination was much higher for the newborn (2.4 in 1000) than for the teenager (0.7 in 1000). For the two pediatric-aged patients in our study, CTDI{sub vol} underestimated dose to large organs in the scan coverage by 30%-48%. The effective dose derived from DLP using published conversion coefficients differed from that calculated using patient-specific organ dose values by -57% to 13%, when the tissue weighting factors of ICRP 60 were used, and by -63% to 28%, when the tissue weighting factors of ICRP 103 were used. Conclusions: It is possible to estimate patient-specific radiation dose and cancer risk from CT examinations by combining a validated Monte Carlo program with patient-specific anatomical models that are derived from the patients' clinical CT data and supplemented by transformed models of reference adults. With the construction of a large library of patient-specific computer models encompassing patients of all ages and weight percentiles, dose and risk can be estimated for any patient prior to or after a CT examination. Such information may aid in decisions for image utilization and can further guide the design and optimization of CT technologies and scan protocols.

60 APPLIED LIFE SCIENCES↗

R and DE Robotic Sensor Intern Team: Advanced Sensor Suite Prototype

The Department of Energy (DOE) nuclear complex includes many hazardous and challenging environments that must be inspected remotely and routinely for structural integrity evaluation and long-term planning purposes. The H-Canyon Air Exhaust (HCAEX) Tunnel is an example of such an environment where a biannual inspection is currently performed. Presently inspections are performed by teleoperated robotic crawlers which provide low-res video feedback to the structural engineers. Advanced sensors: including Lidar and panospheric hi-res cameras have the potential to provide additional valuable data in future inspections. The primary goal of the sensor suite is to provide an internal 3D map of the HCAEX tunnel. Possible sensors were a stereo camera, a 3D Lidar, or a 2D Lidar that would need to be spun around. The issues with the stereo camera were that it does not have a large field-of-view and that its depth information was not nearly as accurate as even the baseline Lidars. The Hokuyo UTM-30LX-EW shown below provides a 270 deg. FOV with high accuracy and resolution while the largest FOV on a 3D Lidar is 40 deg. This was the reason a 2D Lidar was chosen over the 3D Lidar. Even though the 2D Lidar requires additional components and software to generate a 3D map, its FOV advantage is too great given that the HCAEX tunnel is a relatively confined space. Along with the Lidar, a camera sensor, IMU sensor, pan-tit unit, and on board computer needed to be identified. The two panospheric cameras allow for full 360 deg. FOV. The IMU will give data on the state of the robot inside the tunnel. The computer is used to gather and process data on board and to control the pan-tilt system. Mounts were designed and 3D printed to build the prototype of the sensor suite. It was designed to have the Lidar on top With its full FOV free while having the camera below and as close to together as possible while still leaving much of their FOV visible. The design also managed the cables such that they were unlikely to be damaged if in tension and so that they did not intrude much on the cameras vision. The sensor tree developed for the H-Canyon tunnel inspection project produces a large quantity of camera and Lidar data. a postprocessing pipeline was created to facilitate the analysis of these data. The pipeline follows the following routine: 1) Registration - Salient local features in subsequent scans are identified and used to align the scans With one another. This allows production of a single cohesive map of the tunnel. 2) Sensor Fusion - RGB camera data is 'painted' onto the depth data created by the Lidar. 3) Segmentation - Individual surfaces from within the scene are segmented out separately for individual analysis. 4) Degradation Analysis - Automatic mapping of local deviation from the plane in position or surface normal is performed, as is detection of exposed rebar using Lidar return intensity. 2) Obstacle Detection - Lidar data is also used in real time to detect obstacles in the environment. This could give feedback to the operator to successfully drive in the tunnel. 6) Graphic User Interface - a custom GUI was created to facilitate customer use of the system. The system provides real-time feedback from both the cameras and Lidar. The Lidar point-cloud data is also used for obstacle detection. As the Lidar spins, various algorithms are run to analyze the current 'slice' and obstacles in the environment are determined and registered on a cost-map. A Graphical User Interface (GUI) was developed to allow the user to easily command the crawler to set the rotational speed of the Lidar, perform a scan, and register and save the 3D generated cloud after a scan is done. For the good of the crawler, a critical aspect of the GUI design is to allow the user to cancel a scan any time after it is commanded and before it is saved. The sensor trees depth accuracy was evaluated in comparison to a high end commercial FARO Terrestrial Laser Scanner (TLS) with specifications for millimeter accuracy at distances above 10 m. For the purposes of these tests, the depth measurements produced by the TLS were taken as ground truth. Example test wall panels were scanned with the TLS and postprocessed with its proprietary software, and subsequently scanned with the sensor tree system. The two scans were manually aligned and for each point in the output cloud from the sensor tree, the distance to the nearest ground truth point in the FARO cloud was determined. 98.7% of all produced points were within the specified target range of 6 mm, with an average deviation of 1.75 mm. The test walls used have an overall in-wall depth variation of around 40 mm.

12 MANAGEMENT OF RADIOACTIVE AND NON-RADIOACTIVE W↗

Root identification in minirhizotron imagery with multiple instance learning

In this study, multiple instance learning (MIL) algorithms to automatically perform root detection and segmentation in minirhizotron imagery using only image-level labels are proposed. Root and soil characteristics vary from location to location, and thus, supervised machine learning approaches that are trained with local data provide the best ability to identify and segment roots in minirhizotron imagery. However, labeling roots for training data (or otherwise) is an extremely tedious and time-consuming task. This paper aims to address this problem by labeling data at the image level (rather than the individual root or root pixel level) and train algorithms to perform individual root pixel level segmentation using MIL strategies. Three MIL methods (multiple instance adaptive cosine coherence estimator, multiple instance support vector machine, multiple instance learning with randomized trees) were applied to root detection and compared to non-MIL approaches. The results show that MIL methods improve root segmentation in challenging minirhizotron imagery and reduce the labeling burden. In our results, multiple instance support vector machine outperformed other methods. The multiple instance adaptive cosine coherence estimator algorithm was a close second with an added advantage that it learned an interpretable root signature which identified the traits used to distinguish roots from soil and did not require parameter selection.

59 BASIC BIOLOGICAL SCIENCES↗

3D Imaging of Microbial Biofilms: Integration of Synchrotron Imaging and an Interactive Visualization Interface

Understanding the interactions of structured communities known as “biofilms” and other complex matrixes is possible through the X-ray micro tomography imaging of the biofilms. Feature detection and image processing for this type of data focuses on efficiently identifying and segmenting biofilms and bacteria in the datasets. The datasets are very large and often require manual interventions due to low contrast between objects and high noise levels. Thus new software is required for the effectual interpretation and analysis of the data. This work specifies the evolution and application of the ability to analyze and visualize high resolution X-ray micro tomography datasets.

Biofilm Viewer↗

The Automatic Reactor Control System (ARCS) Upgrade, version 2.0 for TREAT

The Transient REActor Test Facility (TREAT) recently underwent an upgraded to the Automatic Reactor Control System (ARCS). The purpose of the upgrade was to institute a new software architecture that is better suited for the programming environment, patched software bugs and applied two new segments for reactor control. This paper will focus on the new segments for reactor control. The two new control segments added to ARCS are called Generic Power and Generic Rods. The original version of ARCS provided only two power related functions, namely periods and ramps, that had to be spliced together to generate any power shape. The Generic Power segment allows the user to input data points to define the function and then the algorithm does its best to create the desired shape. The original ARCS program utilized two direct rod commands (i.e. open loop control), these were a rod stop and clip. A Generic Rod segment extends the ability of the computer to directly command any rod position. The functionality of these new segments was proven during the TREAT outage in 2024.

46 - INSTRUMENTATION RELATED TO NUCLEAR SCIENCE AN↗

Fuel Thermo-physical Characterization Project: Evaluation of Models to Calculate Thermal Diffusivity of Layered Composites

The Office of Material Management and Minimization Fuel Thermo-physical Characterization Project at Pacific Northwest National Laboratory (PNNL) is tasked with using PNNL facilities and processes to receive irradiated low enriched uranium-molybdenum fuel plate samples and perform analyses in support of the Office of Material Management and Minimization Reactor Conversion Program. This work is in support of the Fuel Development Pillar that is managed by Idaho National Laboratory. A key portion of the scope associated with this project was to measure the thermal properties of fuel segments harvested from plates that were irradiated in the Advanced Test Reactor. Thermal diffusivity of samples prepared from the fuel segments was measured using laser flash analysis. Two models, one developed by PNNL and the other developed by the Technische Universität München (TUM), were evaluated to extract the thermal diffusivity of the uranium-molybdenum alloy from measurements made on the irradiated, layered composites. The experimental data of the “TC” irradiated fuel segment was evaluated using both models considering a three-layer and five-layer system. Both models are in acceptable agreement with one another and indicate that the zirconium diffusion barrier has a minimal impact on the overall thermal diffusivity of the monolithic U-Mo fuel.

11 NUCLEAR FUEL CYCLE AND FUEL MATERIALS↗

DLSIA: Deep Learning for Scientific Image Analysis

DLSIA (Deep Learning for Scientific Image Analysis) is a Python-based machine learning library that empowers scientists and researchers across diverse scientific domains with a range of customizable convolutional neural network (CNN) architectures for a wide variety of tasks in image analysis to be used in downstream data processing. DLSIA features easy-to-use architectures, such as autoencoders, tunable U-Nets and parameter-lean mixed-scale dense networks (MSDNets). Additionally, this article introduces sparse mixed-scale networks (SMSNets), generated using random graphs, sparse connections and dilated convolutions connecting different length scales. For verification, several DLSIA-instantiated networks and training scripts are employed in multiple applications, including inpainting for X-ray scattering data using U-Nets and MSDNets, segmenting 3D fibers in X-ray tomographic reconstructions of concrete using an ensemble of SMSNets, and leveraging autoencoder latent spaces for data compression and clustering. As experimental data continue to grow in scale and complexity, DLSIA provides accessible CNN construction and abstracts CNN complexities, allowing scientists to tailor their machine learning approaches, accelerate discoveries, foster interdisciplinary collaboration and advance research in scientific image analysis.

97 MATHEMATICS AND COMPUTING↗

Fully convolutional neural network for removing background in noisy images of uranium bearing particles

A fully convolutional neural network (FCN) was developed to supersede automatic or manual thresholding algorithms used for tabulating SIMS particle search data. The FCN was designed to perform a binary classification of pixels in each image belonging to a particle or not, thereby effectively removing background signal without manually or automatically determining an intensity threshold. Using 8,000 images from 28 different particle screening analyses, the FCN was trained to accurately predict pixels belonging to a particle with near 99% accuracy. Background eliminated images were then segmented using a watershed technique in order to determine isotopic ratios of particles. A comparison of the isotopic distributions of an independent data set segmented using the neural network, compared to a commercially available automated particle measurement (APM) program developed by CAMECA, highlighted the necessity for effective background removal to ensure that resulting particle identification is not only accurate, but preserves valuable signal that could be lost due to improper segmentation. The FCN approach improves the robustness of current state-of-the-art particle searching algorithms by reducing user input biases, resulting in an improved absolute signal per particle and decreased uncertainty of the determined isotope ratios.

Tarolli, Jay G. (ORCID:0000000307369057)↗

Hestia SW-IFL Onroad Fossil Fuel Carbon Dioxide (FFCO2) product: Road segment-level annual FFCO2 emissions across Arizona (2017-2022), version 1.1

The SW-IFL onroad fossil fuel carbon dioxide (FFCO2) emissions data product represents CO2 emissions from the combustion of fossil fuels by motor vehicles (e.g., passenger cars, trucks, buses, motorcycles) traveling on designated roadways. The emissions are represented geographically on each road segment within the state of Arizona spanning the 2017 to 2022 time period. This data product was developed as part of the Southwest Urban Corridor Integrated Field Laboratory (SW-IFL) project, which aims to provide new knowledge and tools that address urban environmental issues by integrating high-resolution observations, modeling, and civic engagement. The emissions data are provided in CSV (input data, ONR_FFCO2_AZ_county.csv) and GeoPackage form (output polyline objects - about 786,000 road segments, XXXX_AZ_v1.1.gpkg) designated by road class (interstates, arterials, collectors, local). The metadata file (Metadata_SW-IFL_Onroad_annualFFCO2_v1.1.docx) provides details about attributes and data formats. The GeoPackage emissions data are provided separately for local roads and nonlocal roads (interstates, arterials, collectors). The method file (Methods_SW-IFL_Onroad_annualFFCO2_v1.1.docx) describes the data processing flow and data sources. Update on 2024-04-17: Updates were made to both the input emission data file (.csv) and output segment-level emission file (.gpkg). There was an update in county-level emission input data (ONR_FFCO2_AZ_county.csv) and the entire road segments were reprocessed to reflect this update.Update on 2024-04-29: Update was made to one output segment-level emission file (Nonlocal_AZ_v1.0.gpkg). There was an error in the AADT values and the data were reprocessed to reflect this update.Update on 2024-10-22: Temporal coverage was extended to include 2022. VMT values were recalculated using new AADT data and the entire road segments were reprocessed to reflect these updates.

54 ENVIRONMENTAL SCIENCES↗

Artificial neural network approach for multiphase segmentation of battery electrode nano-CT images

The segmentation of tomographic images of the battery electrode is a crucial processing step, which will have an additional impact on the results of material characterization and electrochemical simulation. However, manually labeling X-ray CT images (XCT) is time-consuming, and these XCT images are generally difficult to segment with histographical methods. We propose a deep learning approach with an asymmetrical depth encode-decoder convolutional neural network (CNN) for real-world battery material datasets. This network achieves high accuracy while requiring small amounts of labeled data and predicts a volume of billions voxel within few minutes. While applying supervised machine learning for segmenting real-world data, the ground truth is often absent. The results of segmentation are usually qualitatively justified by visual judgement. We try to unravel this fuzzy definition of segmentation quality by identifying the uncertainty due to the human bias diluted in the training data. Further CNN trainings using synthetic data show quantitative impact of such uncertainty on the determination of material’s properties. Nano-XCT datasets of various battery materials have been successfully segmented by training this neural network from scratch. We will also show that applying the transfer learning, which consists of reusing a well-trained network, can improve the accuracy of a similar dataset.

25 ENERGY STORAGE↗

Evaluation of a method for activity estimation in Sm-153 EDTMP imaging

Absolute activity evaluation is fundamental for internal radionuclide dosimetry when patient-specific therapy optimization is wanted. Often, quantification is attempted with 3D SPECT image based (IB) methods, but the true concentration values can be underestimated due to the partial volume effect (PVE). This is especially true when small diffuse lesions are present. In this paper, we describe a 3D region of interest (ROI) based quantification method (LS-ROI), which estimates the ROI concentration values directly from the projection data acquired in the tomographic scan once ROIs have been segmented on a CT and/or a SPECT image. The method, which has inherent PVE correction capabilities, was applied both on simulated and on real phantom data. Simulations reflected the case of a patient with bone metastases treated with {sup 153}Sm-EDTMP: Both the activity in the metastases and the total retention in the skeleton were evaluated. Thirty noisy data sets were produced in order to evaluate the accuracy and precision of the method. The effect of region segmentation errors on estimated concentrations was thoroughly investigated. Real data were acquired on a NEMA phantom, where a cylindrical central region (283 cm{sup 3}) simulated the bone and two spheres (10.3 and 25.5 cm{sup 3}) simulated the metastases. The results obtained with the LS-ROI method were compared with those of a conventional 3D IB method and those of a quantitative conjugate view approach derived from LS-ROI and applied to the anterior and posterior views acquired in the tomographic scan (LS-ROI anterior-posterior: LS-ROI-AP). Simulations showed that when the geometry of regions is known, the LS-ROI method recovered the simulated concentration values within 20%, while the IB method underestimated the concentration in high activity small lesions by as much as 49%. Segmentation errors, up to 44% of the true region volume, produced a higher variation in LS-ROI estimates than in IB ones; however, the overall bias of the LS-ROI estimates ({<=}25%) remained lower than that of IB estimates. In the case of the evaluation of the total retention in the skeleton, the LS-ROI method recovered the simulated value within 2%, while IB underestimated it up to 13%. In all the cases, the LS-ROI-AP method showed an accuracy comparable with that of the LS-ROI one, and a worse precision just because of the lower number of counts used in the analysis. However, a worsening of LS-ROI-AP performances was demonstrated in the case of strong overlap of regions: In this case, a bias of up to 40% was observed. The results obtained on real phantom data confirmed the simulation results: The IB method underestimated activity up to 47% in the smallest sphere, while the bias was reduced to 13% with LS-ROI and LS-ROI-AP estimates. The good quantification capabilities of the LS-ROI method can be useful for absolute activity quantification in the case of small active diffused lesions and constitute the basis for the development of an accurate patient-specific planning strategy in internal radionuclide treatments, provided there is a reliable segmentation of lesions.

62 RADIOLOGY AND NUCLEAR MEDICINE↗

Disclosure of the XRD pipeline software

The XRD pipeline software is a program to reduce 2D X-ray diffraction data from area detectors with advanced algorithms on automatic masking and image segmentation, which help characterization of multiple phases recorded in the data and facilitate data analysis.SF-25-114

XU, WENQIAN [Argonne National Laboratory (ANL), Ar↗

SU-E-T-100: Designing a QA Tool for Enhance Dynamic Wedges Based On Dynalog Files

Purpose: A robust quality assurance (QA) program for computer controlled enhanced dynamic wedge (EDW) has been designed and tested. Calculations to perform such QA test is based upon the EDW dynamic log files generated during dose delivery. Methods: Varian record and verify system generates dynamic log (dynalog) files during dynamic dose delivery. The system generated dynalog files contain information such as date and time of treatment, energy, monitor units, wedge orientation, and type of treatment. It also contains the expected calculated segmented treatment tables (STT) and the actual delivered STT for the treatment delivery as a verification record. These files can be used to assess the integrity and precision of the treatment plan delivery. The plans were delivered with a 6 MV beam from a Varian linear accelerator. For available EDW angles (10°, 15°, 20°, 25°, 30°, 45°, and 60°) Varian STT values were used to manually calculate monitor units for each segment. It can also be used to calculate the EDW factors. Independent verification of fractional MUs per segment was performed against those generated from dynalog files. The EDW factors used to calculate MUs in TPS were dosimetrically verified in solid water phantom with semiflex chamber on central axis. Results: EDW factors were generated from the STT provided by Varian and verified against practical measurements. The measurements were in agreement of the order of 1 % to the calculated EDW data. Variation between the MUs per segment obtained from dynalog files and those manually calculated was found to be less than 2%. Conclusion: An efficient and easy tool to perform routine QA procedure of EDW is suggested. The method can be easily implemented in any institution without a need for expensive QA equipment. An error of the order of ≥2% can be easily detected.

60 APPLIED LIFE SCIENCES↗

Supporting information for Few-Shot Learning Enables Population-Scale Analysis of Leaf Traits in Populus trichocarpa

In this work, we use few-shot learning to segment the body and vein architecture of P. trichocarpa leaves from high-resolution scans obtained in the UC Davis common garden. Leaf and vein segmentation are formulated as separate tasks, in which convolutional neural networks (CNNs) are used to iteratively expand partial segmentations until reaching stopping criteria. Our leaf and vein segmentation approaches use just 50 and 8 manually traced images for training, respectively, and are applied to a set of 2,634 top and bottom leaf scans. We show that both methods achieve high segmentation accuracy, in some cases exceeding even human-level segmentation. The leaf and vein segmentations are subsequently used to extract 68 morphological traits using traditional open-source image processing tools, which are validated using real-world physical measurements. For a biological perspective, we perform a genome-wide association study using the vein density trait to discover novel genetic architectures associated with multiple physiological processes relating to leaf development and function. In addition to sharing all of the few-shot learning code (see https://github.com/jlager/few-shot-leaf-segmentation), we are releasing all images, manual segmentations, model predictions, 68 extracted leaf phenotypes, and a new set of SNPs called against the v4 P. trichocarpa genome for 1,419 genotypes. The data folder includes all images, ground truth segmentations, predicted segmentations, and extracted leaf traits. All images encode the sample ID in the file name by indicating the treatment, block, row, position, and leaf side, respectively. For example, the file, C_1_1_2_bot.jpeg, indicates the control treatment, block 1, row 1, position 2, and the bottom side of the leaf. Tabulated results include position IDs as well as the corresponding genotype IDs. The images folder includes the 2,906 high-resolution leaf scans taken in the field. The leaf_masks folder includes 50 ground truth segmentations used for training the leaf tracing algorithm. The leaf_preds folder includes the 2,906 predicted segmentations from the leaf tracing algorithm. The vein_masks folder includes 8 ground truth segmentations used for training the vein growing algorithm. The vein_preds folder includes the 1,453 predicted segmentations from the vein growing algorithm. The vein_probs folder includes the 1,453 predicted probability maps from the vein growing algorithm before thresholding. The genomes folder includes the set of SNPs called against the v4 P. trichocarpa genome for 1,419 genotypes with a README file detailing the steps taken. The results folder includes: (i) raw values of the 68 predicted leaf traits in digital_traits.tsv, (ii) manually measured values of petiole length and width in manual_traits.tsv, (iii) thin plate spline (TPS) adjusted values of the vein density trait in vein_density_tps_adj.tsv, (iv) best linear unbiased prediction (BLUP) adjusted values of the vein density trait in vein_density_blups.tsv, and (v) GWAS results for the vein density trait, including chromosome positions and corresponding P values, in gwas_results.csv.

09 BIOMASS FUELS↗

CHESS 2025: Waveform LiDAR data from NEON AOP surveys

This dataset provides Level 1 (L1) full-waveform light detection and ranging (LiDAR) data collected for the 2025 Colorado Headwaters Ecological Spectroscopy Study (CHESS). These data were acquired to enable characterization of vegetation structure and other three-dimensional features of the land surface, and to evaluate structural changes that may have occurred between a prior LiDAR acquisition in 2018 and the 2025 overflight. Waveform LiDAR data can provide more detailed information about objects on the ground than discrete point clouds typically do, and they are often used for granular target segmentation and characterization of subcanopy vegetation. The data were acquired over three study domains in the Upper Gunnison river basin: the upper East River watershed (CRBU); Almont Triangle and Taylor Canyon (ALMO); and Upper Taylor River watershed (UPTA) between 2025-06-13 and 2025-07-15. LiDAR data were acquired using the Optech Galaxy Prime Airborne LiDAR Terrain Mapper onboard the National Ecological Observatory Network (NEON) Airborne Observation Platform (AOP). These are the primary waveform LiDAR data delivered by NEON and are provided per flightline in compressed Pulsewaves format, an open-source binary file standard. A Pulsewaves object comprises a two files: a pulse (.pls) file, which stores the geographic origin, outgoing vector, and metadata for every laser pulse emitted by the scanner, and a wave file (.wvs), which stores the sequential amplitude samples of the outgoing pulse and the returning signals. The files are published here in their compressed forms (.plz, .wvz). All waveform data were processed following the theoretical workflow described in the NEON L0-to-L1 Waveform LiDAR Algorithm Theoretical Basis Document (Krause and Goulden 2022a); however, the Pulsewaves output format differs from a legacy format described in that document. Waveform amplitude samples are recorded at 1 nanosecond intervals. All coordinates are provided in meters. Horizontal coordinates are referenced in Universal Transverse Mercator (UTM) zone 13N and the World Geodetic System (WGS) 1984 ensemble datum. Elevations are referenced to Geoid12A. Waveform data for the UPTA survey area were collected without incident and the published records are complete. However, both the ALMO and CRBU collections experienced issues that resulted in incomplete data for those areas. On collection day 2018-06-16 a hardware failure caused the waveform digitizer to lose data from the eastern edge of the ALMO site (Figure 22). The waveform data for flightlines 2–20 could not be extracted from the digitizer, and the data proved unrecoverable. As a result, a portion of the site does not have coverage with waveform data. Although no hardware failure was observed during collection over the CRBU area, final waveform files generated by vendor software contained only ~25% of the expected number of return pulses. After discovery, NEON initiated troubleshooting with the vendor. The root cause of the data ablation had not been identified at the time of publication. Additional data will be published in an update to this package if further recovery proves successful. CHESS Project Description: The Colorado Headwaters Ecological Spectroscopy Study (CHESS) comprised a multi-week airborne remote sensing and field observation campaign in the Upper Gunnison Basin, Colorado, conducted in June and July of 2025. Airborne remote sensing was conducted by the National Ecological Observatory Network Airborne Observation Platform (NEON AOP), concurrent with a field campaign run by the Rocky Mountain Biological Laboratory (RMBL), the Lawrence Berkeley National Laboratory (LBNL) and SLAC National Accelerator Laboratory Watershed Function Science Focus Area (SFA), and NASA-JPL (Jet Propulsion Laboratory) Earth Surface Mineral Dust Source Investigation (EMIT) program. Between June 10 and July 18, 2025, the NEON AOP flight team collected high-resolution aerial imaging spectroscopy and Light Detection and Ranging (LiDAR) data over three domains: the Upper East River (CRBU), Almont Triangle (ALMO), and the Upper Taylor Basin (UPTA). In coordination with the flights, a field campaign acquired ground-truth observations, including observations of vegetation composition, foliar traits, forest demography, and subsurface properties in 18 core sampling areas within the domains. Additional surface water observations were taken at over 380 point locations. All CHESS campaign datasets can be found within the CHESS ESS-DIVE data portal: https://data.ess-dive.lbl.gov/portals/chess. Funding Acknowledgement: Field and remote-sensing data acquisition was performed under a grant from the National Aeronautics and Space Administration (80NSSC24K1005). This work was also supported by the Watershed Function Science Focus Area at Lawrence Berkeley National Laboratory funded by the US Department of Energy, Office of Science, Biological and Environmental Research under Contract No. DE-AC02-05CH11231.

2018 NEON and 2025 CHESS Campaigns↗