Filtern
Dokumenttyp
- Zeitschriftenartikel (4)
- Vortrag (1)
- Forschungsdatensatz (1)
Sprache
- Englisch (6)
Schlagworte
Organisationseinheit der BAM
- 1 Analytische Chemie; Referenzmaterialien (6)
- 6 Materialchemie (6)
- 6.1 Oberflächen- und Dünnschichtanalyse (6)
- 1.2 Biophotonik (5)
- 1.0 Abteilungsleitung und andere (2)
- 4 Material und Umwelt (2)
- 4.2 Material-Mikrobiom Wechselwirkungen (2)
- 5 Werkstofftechnik (2)
- 5.4 Multimateriale Fertigungsprozesse (2)
Paper des Monats
- ja (1)
Eingeladener Vortrag
- nein (1)
We present a workflow for obtaining fully trained artificial neural networks that can perform automatic particle segmentations of agglomerated, non-spherical nanoparticles from electron microscopy images “from scratch”, without the need for large training data sets of manually annotated images. This is achieved by using unsupervised learning for most of the training dataset generation, making heavy use of generative adversarial networks and especially unpaired image-to-image translation via cycle-consistent adversarial networks. The whole process only requires about 15 minutes of hands-on time by a user and can typically be finished within less than 12 hours when training on a single graphics card (GPU). After training, SEM image analysis can be carried out by the artificial neural network within seconds, and the segmented images can be used for automatically extracting and calculating various other particle size and shape descriptors.
This dataset accompanies the following publication, first published in Scientific Reports (www.nature.com/articles/s41598-021-84287-6):
B. Ruehle, J. Krumrey, V.-D. Hodoroaba, Scientific Reports, Workflow towards Automated Segmentation of Agglomerated, Non-Spherical Particles from Electron Microscopy Images using Artificial Neural Networks, DOI: 10.1038/s41598-021-84287-6
It contains electron microscopy micrographs of TiO2 particles, the corresponding segmentation masks, and their classifications into different categories depending on their visibility/occlusion. Please refer to the publication and its supporting information for more details on the acquisition and contents of the dataset, as well as the GitHub repository at https://github.com/BAMresearch/automatic-sem-image-segmentation
Reliable measurement of the size of polydisperse, complex-shaped commercial nanopowders is a difficult but necessary task, e.g., for regulatory requirements and toxicity risk assessment. Suitable methods exist for the accurate characterization of the size of non-aggregated, stabilized, spherical and monodisperse nanoparticles. In contrast, industrial nanoscale powders usually require dedicated sample preparation procedures developed for the analysis method of choice. These nano-powders tend to agglomerate and/or aggregate, a behavior which in combination with an innate broad particle size distribution and irregular shape often significantly alters the achievable accuracy of the measured size parameters. The present study systematically tests two commercially available nanoscale powders using different sample preparation methods for correlative analysis by scanning electron microscopy, dynamic light scattering, Brunauer–Emmet–Teller method and differential mobility analysis. One focus was set on the sample preparation by embedding nanoparticles in carbon-based hot-mounting resin. Literature on this topic is scarce and the accuracy of the data extracted from cross sections of these particles is unclearly stated. In this paper systematic simulations on the deviation of the size parameters of well-defined series of nanoparticles with different shapes from the nominal value were carried out and the contributing factors are discussed.
BAM is currently building up a platform of novel nanoRMs relying on iron oxide nanoparticles of different shape, size and surface chemistry. Iron oxide was chosen as a core material because of its relevance to the material and life sciences.
As a first candidate of this series, we present cubic iron oxide nanoparticles with a nominal edge length of 8 nm. These particles were synthesized by thermal decomposition of iron oleate in high boiling organic solvents adapting well-known literature procedures. After dilution to a concentration suitable for electron microscopy (TEM and SEM) as well as for small-angle X-ray scattering (SAXS) measurements, the candidate nanoRM was bottled and assessed for homogeneity and stability by both methods following the guidelines of ISO 17034 and ISO Guide 35.
The particle sizes obtained by both STEM-in-SEM and TEM are in excellent agreement with a minimum Feret of 8.3 nm ± 0.7 nm. The aspect ratio (AR) of the iron oxide cubes were extracted from the images as the ratio of minimum Feret to Feret resulting in an AR of 1.18 for TEM to 1.25 for SEM. Alternatively, a rectangular bounding box was fitted originating from the minimum Feret and the longest distance through the particle in perpendicular direction. This led to AR values of 1.05 for TEM and 1.12 for SEM, respectively. The results confirm the almost ideal cubic shape.
In this contribution different ways are explored with the aim to generate suitable training data for ‘non-ideal’ samples using various approaches, e.g., computer-generated images or unsupervised learning algorithms such as generative adversarial networks (GANs). We used these data to train simple CNNs to produce segmentation masks of SEM images and tested the trained networks on real SEM images of complex nanoparticle samples. The novel use of CNN for the automated analysis of the size of nanoparticles of complex shape and with a high degree of agglomeration has proved to be a promising tool for the evaluation of particle size distribution on a large number of constituent particles. Further development and validation of the preliminary model, respectively larger training and validation data sets are necessary.
We present a workflow for obtaining fully trained artificial neural networks that can perform automatic particle segmentations of agglomerated, non-spherical nanoparticles from scanning electron microscopy images “from scratch”, without the need for large training data sets of manually annotated images. The whole process only requires about 15 minutes of hands-on time by a user and can typically be finished within less than 12 hours when training on a single graphics card (GPU). After training, SEM image analysis can be carried out by the artificial neural network within seconds. This is achieved by using unsupervised learning for most of the training dataset generation, making heavy use of generative adversarial networks and especially unpaired image-to-image translation via cycle-consistent adversarial networks. We compare the segmentation masks obtained with our suggested workflow qualitatively and quantitatively to state-of-the-art methods using various metrics. Finally, we used the segmentation masks for automatically extracting particle size distributions from the SEM images of TiO2 particles, which were in excellent agreement with particle size distributions obtained manually but could be obtained in a fraction of the time.