Search

Research on metamaterials shows excellent potential in the field of solar energy harvesting. In recent years, the design of broadband solar metamaterial absorbers (SMAs) has attracted significant interest with the wide application of deep learning methods. This paper proposes a deep neural network (DNN) to realize forward prediction and inverse design of reconfigurable 3D SMAs. In the inverse design, a polarization-insensitive broadband SMA with an absorption bandwidth of 2.7 μm and an average absorption rate of 97.6% with an adjustable bandwidth range of 369 nm is successfully designed. The design of SMAs with different structures is also realized by a transfer learning method to improve the training speed and performance further. Using the transfer learning approach, the training speed of the neural network target model can be accelerated, and its training performance can be improved on small datasets by utilizing the trained neural network source model. Meanwhile, using the trained inverse design target model, a polarization-insensitive broadband SMA was designed with an absorption bandwidth of 2.7 μm, an average absorption of 97.9%, and an adjustable bandwidth range of 141 nm. Finally, we verified the solar energy harvesting capability of the designed broadband SMAs under real-world conditions using air mass (AM) 1.5, and they were calculated to be capable of harvesting the vast majority of the energy. The method is instructive in the design process of SMAs and can be effectively used to explore multifunctional complex nanophotonic devices. • Designing reconfigurable solar absorbers with material selectivity using deep learning methods. • Design of ultra-wideband solar metmaterial absorbers with different shapes through the transfer learning approach. • The designed solar metamaterial absorber has high solar energy harvesting performance. [ABSTRACT FROM AUTHOR]

Detecting road cracks is essential for inspecting and assessing the integrity of concrete pavement structures. Traditional image-based methods often require complex preprocessing to extract crack features, making them challenging when dealing with noisy concrete surfaces in diverse real-world scenarios, such as autonomous vehicle road detection. This study introduces an image-based crack detection approach that combines a Random Forest machine learning classifier with a deep convolutional neural network (CNN) to address these challenges. Three state-of-the-art models, namely MobileNet, InceptionV3, and Xception, were employed and trained using a dataset of 30,000 images to build an effective CNN. A systematic comparison of validation accuracy across various base learning rates identified a base learning rate of 0.001 as optimal, achieving a maximum validation accuracy of 99.97%. This optimal learning rate was then applied in the subsequent testing phase. The robustness and flexibility of the trained models were evaluated using 6,000 test photos, each with a resolution of 224 × 224 pixels, which were not part of the training or validation sets. The outstanding results, boasting a remarkable 99.95% accuracy, 99.95% precision, 99.94% recall, and a matching 99.94% F1 Score, unequivocally affirm the efficacy of the proposed technique in precisely identifying road fractures in photographs taken on real concrete surfaces. [ABSTRACT FROM AUTHOR]