Researchers From UNSW Sydney Proposes A Deep Learning Algorithm That Produces High-Resolution Modeled Images From Lower-Resolution Micro X-ray Computerized Tomography (CT)

Researchers have turned away from fossil fuels in favor of clean and renewable energy sources in response to the significant change in climate over the past several years. There is substantial emphasis on hydrogen fuel cells and Proton Exchange Membrane Fuel Cells (PEMFCs) when it comes to different renewable energy sources. PEMFCs are a clean energy source capable of producing electricity using hydrogen fuel. They are mainly being developed for powering transport applications like cars, etc., but can also be used to power homes and businesses. These fuel cells work by electrochemically converting hydrogen into power, with pure water being the reaction’s only byproduct. PEMFCs are a great source of renewable energy, but they have a significant flaw that can render them fairly ineffective.

Due to the fuel cells’ extremely small size and complicated structural composition, engineers until now have had a very difficult time understanding the specific manner in which water drains, or even pools, inside them. If the water created as a byproduct of the reaction cannot adequately exit the cell and instead “floods” the system, PEMFCs may lose their efficiency. Researchers have previously attempted to use various water visualization techniques, including optical imaging, X-ray radiography, and X-ray micro-computed tomography, to improve PEMFC designs and increase their efficiency. However, these techniques’ current resolution and field of view are insufficient to resolve a PEMFC porous structure completely.

To better understand what is happening inside a PEMFC, researchers from the University of New South Wales, Australia (UNSW) have developed a super-resolution algorithm known as DualEDSR that produces high-resolution modeled images from lower-resolution micro X-ray computerized tomography (CT). The new technology has been tried on individual hydrogen fuel cells to precisely model the inside and increase their efficiency. DualEDSR can improve the field of view by about 100 times compared to the high-resolution image. The study has also been published in the prestigious Nature Communications journal.

No matter the hardware being utilized, it is normal for an image’s resolution to decline as it is zoomed out. Here, by overcoming this issue, UNSW researchers have made ground-breaking progress. The team applies deep learning techniques to a low-resolution X-ray image of the cell and data from a related high-resolution scan of a sub-section to produce a detailed 3D model of the cell. In simple terms, it’s like being able to precisely predict the layout of every road in a region from a fuzzy aerial shot of a town and a detailed photo of a few streets. The researchers’ methodology is relevant anywhere imaging is used, such as in medical applications, the fuel industry, or chemical engineering. Using the researchers’ deep learning technology on human X-rays would allow medical practitioners to diagnose various conditions more accurately and quickly by providing a deeper knowledge of the body’s underlying cellular structures.

The DualEDSR algorithm can create a complete 3D representation of the inside of a PEMFC for manufacturers to better regulate the water produced and increase the efficiency of the fuel cells. These issues can therefore be easily resolved in upcoming designs as the model can precisely identify where water tends to gather. Based on several experimental evaluations conducted by the researchers, it was revealed that the algorithm produced high-resolution modeling from low-resolution pictures with an accuracy of 97.3%. Compared to how long it would have taken to use a micro-CT scanner to get high-resolution images of the entire fuel cell, the algorithm created a high-resolution model significantly more quickly.

The researchers are extremely excited about uncovering the potential of PEMFCs in providing clean and environmentally-friendly electricity in the future. Due to the complexity of the materials, the flow of gases and liquids, and the electrochemical reactions occurring, it has become increasingly challenging to obtain an accurate model of PEMFCs during the past 20 years. But, the UNSW research team has succeeded in achieving just that. Another area that the researchers are keenly interested in is medical imaging. The team intends to expand their research in the future by using deep learning techniques for images that weren’t shot in the same place or, possibly, even with the same tool. 


Check out the Paper and Reference Article. All Credit For This Research Goes To the Researchers on This Project. Also, don’t forget to join our 14k+ ML SubRedditDiscord Channel, and Email Newsletter, where we share the latest AI research news, cool AI projects, and more.

Do You Know Marktechpost has 1.5 Million+ Pageviews per month and 500,000 AI Community members?
Want to support us? Become Our Sponsor

Khushboo Gupta is a consulting intern at MarktechPost. She is currently pursuing her B.Tech from the Indian Institute of Technology(IIT), Goa. She is passionate about the fields of Machine Learning, Natural Language Processing and Web Development. She enjoys learning more about the technical field by participating in several challenges.

🚀 LLMWare Launches SLIMs: Small Specialized Function-Calling Models for Multi-Step Automation [Check out all the models]