Research on a γ-Photon Flow-Field Image Denoising Algorithm Based on the SFGD Model

In the complex world of industrial engineering, monitoring the invisible currents of hot gas inside a jet engine is a formidable challenge. Traditional cameras cannot survive the heat, and magnetic fields can confuse electronic sensors. To see inside these turbulent environments without touching them, scientists turn to high-energy particles called gamma photons. These particles are born when radioactive atoms decay, and they possess a unique ability: they can pass through solid metal and extreme heat without being deflected or absorbed. By tracking how these particles travel, researchers can reconstruct a picture of the flow field, revealing where the air swirls, where it stalls, and how the engine breathes. However, this method is like trying to see a clear landscape through a heavy fog. The data collected is often sparse and noisy, resulting in images where the sharp edges of swirling air currents are blurred into a hazy mess, making it difficult for engineers to understand the physics at play.

To solve this problem, a team of researchers at Nanjing University of Aeronautics and Astronautics has developed a new way to clean up these blurry images. They created a sophisticated computer program designed to act like a highly skilled restorer, taking a grainy, distorted picture of a gas flow and reconstructing it into a sharp, clear image. The key to their success lies in understanding that the noise in these images behaves differently at different stages of the cleaning process. Instead of applying a single, static filter to the whole picture, their new system, called the Structure-Frequency Guided Diffusion model, adapts its strategy moment by moment. It learns to hold onto the large, important shapes of the flow while carefully sharpening the tiny, intricate details that usually get lost in the static.

The researchers began by simulating the entire process on a computer. They took detailed maps of gas flow from a jet engine and used a virtual detector to generate thousands of noisy, incomplete images, mimicking exactly what a real sensor would see. They then trained their new algorithm on these simulated pairs, teaching it to recognize the difference between random static and the actual shape of a swirling vortex. The system works by using two specialized tools working in tandem. The first tool focuses on the big picture. It looks at the blurry input and identifies the main outlines and boundaries of the flow, ensuring that the reconstructed image keeps the correct shape and position of the gas currents. This prevents the final image from becoming a smooth, featureless blob where the important structures have vanished.

The second tool is more subtle and dynamic. It understands that the type of noise changes as the image is being cleaned. At the very beginning of the process, the image is full of chaotic static, and the system knows to be gentle, avoiding any attempt to sharpen details that might just be noise. As the cleaning progresses and the main shapes become clearer, this tool gradually shifts its focus. It begins to selectively enhance the high-frequency details, such as the sharp edges of a vortex or the fine texture of a wake, which are crucial for understanding the engine's performance. This time-sensitive adjustment allows the system to suppress the noise without accidentally smoothing away the very details the engineers need to see.

When the team tested their new method against existing techniques, the results were striking. They compared their approach to several other advanced image-cleaning algorithms, including those based on standard deep learning and older diffusion models. In every test case, their new model produced images that were significantly clearer and more accurate. On a specific set of independent test images, the new method achieved a level of clarity that was nearly 5.5 units higher on a standard quality scale than the previous best method. Visually, this meant that the swirling patterns of the gas flow were preserved with remarkable precision. Where other methods left the edges of the flow blurred or distorted, the new model kept the boundaries sharp and the internal structures intact, closely matching the original, noise-free simulation.

The study confirms that combining a strong focus on the overall shape of the flow with a smart, changing strategy for handling fine details is the key to unlocking clear images from gamma photon data. By teaching the computer to adapt its cleaning process as the image evolves, the researchers have provided a powerful new tool for non-contact monitoring. This advancement suggests that engineers will soon be able to see the hidden dynamics of industrial flows with a clarity that was previously impossible, allowing for better design and safer operation of complex machinery. The work does not claim to have solved every imaging problem, but it demonstrates a clear and measurable improvement in how we can see the invisible world of high-speed gas flows.