Abstract

In this study, we introduce an innovative methodology for robust twist angle identification in bilayer graphene using Raman spectroscopy, featuring the integration of generative adversarial network and convolutional neural network (GAN-CNN). Our proposed approach showcases remarkable resistance to noise interference, particularly in ultra-low Signal-to-Noise Ratio (SNR) conditions. We demonstrate the GAN-CNN model's robust learning capability, even when SNR reaches minimal levels. The model's exceptional noise resilience negates the necessity for preprocessing steps, facilitating accurate classification, and substantially reducing computational expenses. Empirical results reveal the model's prowess, achieving heightened accuracy in twist angle identification. Specifically, our GAN-CNN model achieves a test accuracy exceeding 99.9% and a recall accuracy of 99.9%, relying on an augmented dataset containing 4209 spectra. This work not only contributes to the evolution of noise-resistant spectral analysis methodologies but also provides crucial insights into the application of advanced deep learning techniques for bilayer graphene characterization through Raman spectroscopy. The findings presented herein have broader implications for enhancing the precision and efficiency of material characterization methodologies, laying the foundation for future advancements in the field.

Abstract

In this study, we introduce an innovative methodology for robust twist angle identification in bilayer graphene using Raman spectroscopy, featuring the integration of generative adversarial network and convolutional neural network (GAN-CNN). Our proposed approach showcases remarkable resistance to noise interference, particularly in ultra-low Signal-to-Noise Ratio (SNR) conditions. We demonstrate the GAN-CNN model's robust learning capability, even when SNR reaches minimal levels. The model's exceptional noise resilience negates the necessity for preprocessing steps, facilitating accurate classification, and substantially reducing computational expenses. Empirical results reveal the model's prowess, achieving heightened accuracy in twist angle identification. Specifically, our GAN-CNN model achieves a test accuracy exceeding 99.9% and a recall accuracy of 99.9%, relying on an augmented dataset containing 4209 spectra. This work not only contributes to the evolution of noise-resistant spectral analysis methodologies but also provides crucial insights into the application of advanced deep learning techniques for bilayer graphene characterization through Raman spectroscopy. The findings presented herein have broader implications for enhancing the precision and efficiency of material characterization methodologies, laying the foundation for future advancements in the field.

Abstract

In this study, we introduce an innovative methodology for robust twist angle identification in bilayer graphene using Raman spectroscopy, featuring the integration of generative adversarial network and convolutional neural network (GAN-CNN). Our proposed approach showcases remarkable resistance to noise interference, particularly in ultra-low Signal-to-Noise Ratio (SNR) conditions. We demonstrate the GAN-CNN model's robust learning capability, even when SNR reaches minimal levels. The model's exceptional noise resilience negates the necessity for preprocessing steps, facilitating accurate classification, and substantially reducing computational expenses. Empirical results reveal the model's prowess, achieving heightened accuracy in twist angle identification. Specifically, our GAN-CNN model achieves a test accuracy exceeding 99.9% and a recall accuracy of 99.9%, relying on an augmented dataset containing 4209 spectra. This work not only contributes to the evolution of noise-resistant spectral analysis methodologies but also provides crucial insights into the application of advanced deep learning techniques for bilayer graphene characterization through Raman spectroscopy. The findings presented herein have broader implications for enhancing the precision and efficiency of material characterization methodologies, laying the foundation for future advancements in the field.

Abstract

In this study, we introduce an innovative methodology for robust twist angle identification in bilayer graphene using Raman spectroscopy, featuring the integration of generative adversarial network and convolutional neural network (GAN-CNN). Our proposed approach showcases remarkable resistance to noise interference, particularly in ultra-low Signal-to-Noise Ratio (SNR) conditions. We demonstrate the GAN-CNN model's robust learning capability, even when SNR reaches minimal levels. The model's exceptional noise resilience negates the necessity for preprocessing steps, facilitating accurate classification, and substantially reducing computational expenses. Empirical results reveal the model's prowess, achieving heightened accuracy in twist angle identification. Specifically, our GAN-CNN model achieves a test accuracy exceeding 99.9% and a recall accuracy of 99.9%, relying on an augmented dataset containing 4209 spectra. This work not only contributes to the evolution of noise-resistant spectral analysis methodologies but also provides crucial insights into the application of advanced deep learning techniques for bilayer graphene characterization through Raman spectroscopy. The findings presented herein have broader implications for enhancing the precision and efficiency of material characterization methodologies, laying the foundation for future advancements in the field.