Research Progress and Prospect of Adaptive Optics Based on Deep Learning

Significance Adaive prs (AC) muligy enhances maging spalny by mening and gamsung fee wavelers. It whern widely used in groband telepus, hickgical imaging, seule alimsition morrerin, and lower miration, Cum Amb estored iw grimps depending the press or she of (WES). WWFS) AD terug spines the pupil phase in reveal algorithmn ham she light intensity distribution. This type if wchnology and divided in two kinds, singleimage and phase deity igy. Single image-hand wximadagy me the winners thigh single insity However, the pha listiti cainel from, a muditory intensity image follow on mapping relationship, rounding in limited cutiey. On the other hand, the phy AO weihnique ran irimsir the phase their field the pine by collecting image information of the heal plane and the helming plane, ruling in high desertion serursey However, lege der al Idera țimea and mexmurrent pred in altain sptimal resulte using traditial WFS AO trhligy, making it usualle highspeed and wali WFS AD play WFS had the interne principle in aditional grometrin opties principle in measure the waseim Examples WFS phase-shifting leurs WFS, Shuck Ha WFS (SHWFS), and pyramid WFS (PyWFS) A high mentre chieved using the militional phase* shing interfere WFS method, but its real-time performans is slegrimal and is reptile envial stres. The SHWFS widely used in AD systems has its simple struct and was of operation. However, result of its pupil wegmentation merhuman the spatial result of the image is and the dynamic range is small. While the PyWFS de wroker light than SHWES Army is expensive and has all linear range in the unmisluised mode Recently, the rigid develop of artifcial intellignes has scelerated develigmant var fra Deep haming waligy, a quilicam heck of artificial lligence, has exhibited remarkable papadalities in search, data mining, machine anal, and sitim. De hearing sigrithms an email us, which stige weight l bi in the given telles. The neural network, after being mined with a soft, urately establish the inpar output relationship. Despite the prolonged training time. results can be inferred quickly, making it useful in a multitude of technical domains. The combination of AO and deep leaming technology is expected to overcome the issues encountered in conventional AO techniques. It is hypothesized that deep learning can lead to faster and more precise wavefront correction, thereby enhancing the performance of AO technology Progress This review introduces several popular artificial leaming has been combined with AD technology are classifies into two categories, techniques with and techniques without WFS. The WFSless category is subdivided into single-image-hased ( WFS category includes examples of SIWFSs (Figs. 7-) review introduces a new diffraction neural network (Fig. how this diffraction neural network has been combined examples of deep learning combined with AO technology Traditional AO techniques. Finally, the review discusses the Conclusions and Prospects Utilizing deep learning with WFSless AO technology provides several favorable advantages, such as its simple structure and low cost. While the single-ma-hased method only uses one image to corret the aberration, the tamany mapping, ultimately resulting in inaccurate calculations. On the other hand, the phase-diversity hased method uses two images with known phase differences to determine the size of the aberrations yielding more accurate results than via the single-image-based method. Within the WFS AO technology field, numerous SHIWES based methods exist. With a focus on improving the accurates of centroid position and wavefrom reconstruction, the application of accuracy. Wavefront measurement methods based on sensors other than corresponding phase of the intensity image reveals a one deep learning networks has accelerated and further improved the SHWFS have gradually been integrated with deep learning technology. neural networks (Fig. 1) used in deep learning. The ways in which deep Fs 3-4) and phase diversity-based (Figs. 5-6) technologies, while the and other WFS technologies combined with deep learning. Moreover, the 10. building on the traditional neural network. and provides examples of with AO technology. The review notes that, over the past five years, have focused on improving the real-time performance and accuracy of nature development directions for deep learning-based AC technology. In the future, deep leaming algorithms will be combine with other technologies, including reinforcement learning, and applied to other types of sensors such as the PyWFS to further enhance AO performance. Furthermore, AO will likely be integrated with a novel optical neural network to optimize its performance. Ispite the growing body of literature on deep-learning hased AO, most studies have been limited to simulation data, thus, it is imperative to evaluate deep learning based AO asing real-world scenarios. Mareaves, while current AO technology focuses on the correction of pain-source wavefront ermis, the detection of extended-source wavefront errors should also be explored in future developments. . -

Automated summary

Adaptive optics (AC) multiplexing technology improves imaging quality by manipulating and sensing light waves. It is widely used in telecommunications, medical imaging, and astronomy. It can be divided into single-channel and phase-diversity techniques. Single-channel methods capture images using a single intensity, but accuracy is limited. Phase-diversity methods collect imaging information from different planes, yielding higher accuracy. The combination of deep learning and adaptive optics is expected to overcome limitations of traditional techniques and improve wavefront correction speed and accuracy.