Since the early 1930s, electron microscopes have provided unprecedented access to the incredibly tiny alien world, revealing the intricate details that conventional light microscopes have. cannot distinguish. But to achieve high resolution over a large sample area, the energy of the electron beams needs to be increased, which is costly and detrimental to the specimen being observed.
Texas A&M University researchers may have come up with a new way to improve the quality of low-resolution electron microscopy images without compromising sample integrity. By training the deep neural network, a type of artificial intelligence algorithm, on pairs of images from the same pattern but at different physical resolutions, they find that the details in the image have Lower resolutions can be further enhanced.
“Typically, a high-energy electron beam is passed through the sample at locations where the image has a higher resolution than expected. But with our image-processing technique, we can hyper-resolution. the entire image using only a few smaller, high-resolution images, ”said Dr. Yu Ding, Mike and Professor Sugar Barnes at the Ws Michael Barnes’ Department of Systems and Industry Engineering. . “This method is less destructive because most parts of the specimen do not need to be scanned with high energy electron beams.”
The researchers have published their image processing techniques in Transaction of electrical and electronic engineers on image processing in June.
Unlike in a light microscope where photons, or small packages of light, are used to illuminate an object, in an electron microscope, an electron beam is used. Electrons that are reflected from or passing through an object are then collected to form an image, called an electron microscope image.
Therefore, the energy of the electron beam plays an important role in determining the image resolution. That is, the higher the energy electrons, the better the resolution. However, the risk of damaging the specimen is also increased, as with ultraviolet rays, a relative of more energy visible light, can damage sensitive materials such as skin.
“There’s always a dilemma for scientists,” said Ding. “In order to maintain specimen integrity, high energy electron beams are used economically. But if one does not use energy beams, high resolution or the ability to see at the beams.” The smaller scale will be limited. “
But there are ways to get high resolution or super resolution by using low resolution images. One method involves using multiple low resolution images of essentially the same area. Another method learns common patterns between small image arrays and uses unrelated high-resolution images to enhance existing low-resolution images.
These methods use almost exclusively natural light images instead of electron microscopy images. Therefore, they had problems with the super-resolution electron microscope image because the basic physics for light and electron microscopy are different, Ding explained.
Researchers have turned to pairs of low-resolution and high-resolution electron microscopy images for a given sample. Although these types of pairs are not very common in public image databases, they are relatively common in materials science research and medical imaging.
For their experiments, Ding and his team first took a low-resolution image of a specimen and then injected about 25% of the observed area into the high-energy electron beam to get the result. High resolution images. The information in the pair of high-resolution and low-resolution images is strongly correlated, the researchers note. They say that this property can be leveraged even though the available data set may be small.
For their analysis, Ding and his team used 22 image pairs of materials infused with nanoparticles. They then divide the high-resolution image and its equivalent area in the low-resolution image into three for three subpages. Next, each pair of sub-images is used to “self-train” the deep neural network. After training, their algorithm became familiar with the recognition of visual features, such as edges.
When they examined the trained deep neural network on a new position on the low-resolution image without high-resolution collation, they found that their algorithm could enhance the recognizable features up to 50%.
While their image processing techniques show a lot of promise, Ding notes that it still requires a lot of computational power. In the near future, his team will direct their efforts towards developing algorithms that are much faster and can be supported by lower computer hardware.
“Our stitched image processing technique shows details in low-resolution images that were previously unnoticeable,” says Ding. “We are all familiar with the magic wand feature on our smartphones. It makes the image clearer. What we aim for in the long run is to provide the research community with a tool. Similar convenience for enhancing electron microscopy images. ”
Electronic refrigeration microscope: Uses inexpensive technology to produce high-resolution images
Yanjun Qian et al., Efficient Super Resolution Method for Coupled Electron Microscopy Imaging, Transactions IEEE on image processing (Year 2020). DOI: 10.1109 / TIP.2020.3000964
Provided by Texas A&M University
Quote: New super resolution method that shows fine details without constant zooming (2020, August 13) retrieved August 13, 2020 from https://techxplore.com/news/ 2020-08-super-resolution-method-reveals-fine- continuity.html
This material is the subject to have fake rights. Other than any fair dealings for academic or personal research purposes, no part may be reproduced without written permission. The content provided is for informational purposes only.