Convolutional Neural Network-based Femtosecond Laser Processing with Adaptive Optics

George Miller^*
*Correspondence: George Miller, Department of Optical Research, University of Dallas, Dallas, USA, Email:

Introduction

In biomedical research, non-invasive, real-time imaging and deep focus into tissue are in high demand. However, the aberration generated by biological tissue's refractive index inhomogeneity obstructs progress. In this paper, machine learning is used to demonstrate quick focusing with sensorless aberration corrections. After training, the proposed method uses a Convolutional Neural Network to quickly calculate low-order aberrations from point spread function images with Zernike modes [1].

Description

In biomedical research, non-invasive, real-time imaging and deep focus into tissue are in high demand. However, the aberration generated by biological tissue's refractive index inhomogeneity obstructs progress, machine learning is used to demonstrate quick focusing with sensor-less aberration corrections. After training, the proposed method uses a Convolutional Neural Network to quickly calculate low-order aberrations from point spread function images with Zernike modes [2]. The progress of biological imaging in recent years has centred on real-time, high-resolution, and deep in vivo imaging. For high-resolution microscopy, adaptive optics becomes a valuable approach. It adjusts for specimen distortions and produces high-resolution pictures in deep was created for telescopes to compensate for atmospheric distortions that decrease the visual quality of alien objects.

Using an active device such as a deformable mirror or a spatial light modulator, it has recently been used in optical microscopy to recover diffractionlimited imaging deep in biological tissue [3]. The active element's refresh rate, on the other hand, is basically limiting the imaging speed. Furthermore, the total fluorescent photon budget is restricted, which indicates that fewer photons should be employed as the feedback signal to evaluate wave front aberrations in order to achieve a greater signal to background ratio. Traditional adaptive optics systems measure aberrations with a wave front sensor, such as a Shack- Hartman wave front sensor. As an example, Implementation is difficult and may result in measurement inaccuracies. Model-based wave front sensor-fewer techniques provide an alternate option. With parallel measurements, devised an approach to reconstruct the precise wave front for even discontinuous wave fronts and enhance the correction speed.

When the number of pupil segments is increased for finer wave front adjustments, however may consume significantly more time [4]. This approach may adjust for wave front aberrations quickly and efficiently, resulting in less photo bleaching and photo damage. Although there have been some previous methods that combine machine learning and live-cell super-resolution imaging to enable faster and gentler high-throughput and live-cell super-resolution imaging this is the first attempt to combine a machine learning algorithm with an correction method for aberrations compensation. Our technique is applicable for any system with good compatibility due to the general expression of the wave front with Zernike modes. We used our technology to correct. The difference in phase or optical path length between the ideal form and the wave front aberrations can be quantified [5]. It can be expressed mathematically as a summation of the Zernike polynomials, which are a set of orthogonal basic functions within a unit circle.

Conclusion

The net architecture we employed is based on a deep learning neural network called a convolutional neural network. Deep learning is a type of machine learning approach that analyses signals or data using multi-layered artificial neural networks. In comparison to fully linked networks have numerous convolutional layers. Convolutional layers employ a convolution process to the input, simulating an individual neuron's reaction to visual stimuli. These layers' convolution filters are randomly started and trained to learn how to extract certain visual task information. This means that the network learns the feature extraction that was previously done by hand using standard methodologies. The prior knowledge and human work in is a significant advantage.

Conflict of Interest

None.

References

1. Elsheikh, Ammar H., Taher A. Shehabeldeen, Jianxin Zhou and Ezzat Showaib. "Prediction of laser cutting parameters for polymethylmethacrylate sheets using random vector functional link network integrated with equilibrium optimizer." J  Intel Manufa 32 (2021): 1377-1388.

Indexed at, Google Scholar, Crossref

2. Scheinker, Alexander and Reeju Pokharel. "Adaptive 3D convolutional neural network-based reconstruction method for 3D coherent diffraction imaging." J Appl Phys 128 (2020): 184901.

Indexed at, Google Scholar, Crossref

3. Zhou, Hongping, Zhenzhen Pan, Maxime Irene Dedo and Zhongyi Guo. "High-efficiency and high-precision identification of transmitting orbital angular momentum modes in atmospheric turbulence based on an improved convolutional neural network." J Opt 23 (2021): 065701.

Indexed at, Google Scholar, Crossref

4. Li, Lu-Ning, Xiang Feng Liu, Fan Yang and Wei Ming Xu."A review of artificial neural network based chemometrics applied in laser-induced breakdown spectroscopy analysis." Spec Acta P B Atom Spect 180 (2021): 106183.

Indexed at, Google Scholar, Crossref

5. Tani, Shuntaro and Yohei Kobayashi. "Ultrafast laser ablation simulator using deep neural networks." Sci Rep 12 (2022): 1-8.

Google Scholar, Crossref