Holographic imaging has always faced challenges when it comes to capturing clear and accurate images in dynamic environments. The traditional methods of using deep learning algorithms often struggle to adapt to diverse scenes due to their reliance on specific data conditions. However, researchers at Zhejiang University have made significant progress in this field by combining the principles of optics and deep learning. Their innovative method, TWC-Swin, leverages physical priors to ensure the alignment of data and pre-trained models, resulting in high-quality holographic images even in the presence of spatial coherence and turbulence disturbances.
Spatial coherence plays a critical role in holographic imaging as it determines how orderly light waves behave. In chaotic environments, such as those involving oceanic or atmospheric turbulence, the light waves become disordered, leading to blurry and noisy holographic images. This is because the disordered waves carry less information. To overcome this challenge, it is crucial to maintain spatial coherence throughout the imaging process.
The TWC-Swin method, which stands for “train-with-coherence swin transformer,” was developed by the researchers at Zhejiang University. This approach incorporates spatial coherence as a physical prior to guide the training of a deep neural network. The network is based on the Swin transformer architecture, which is known for its ability to capture both local and global image features.
To evaluate the effectiveness of the TWC-Swin method, the researchers designed a light processing system that produced holographic images with varying levels of spatial coherence and turbulence conditions. These holograms, based on natural objects, served as both training and testing data for the neural network. The results demonstrated that TWC-Swin successfully restored holographic images even under conditions of low spatial coherence and arbitrary turbulence. In fact, the method outperformed traditional convolutional network-based methods in terms of image quality. Additionally, the researchers noted that TWC-Swin exhibited strong generalization capabilities, allowing it to be applied to unseen scenes that were not included in the training data.
This research marks a significant breakthrough in the field of holographic imaging. By integrating physical principles into deep learning algorithms, the researchers have paved the way for enhanced holographic imaging techniques. This successful synergy between optics and computer science has the potential to revolutionize the field, enabling us to capture clear and accurate holographic images even in the presence of turbulence and other disturbances. As the future unfolds, we can expect further advancements and applications of this technology, empowering us to see through the turbulence and gain new insights through holographic imaging.