Search

• All-optical tactile sensor realized with only one fiber. • Accurate 2D location + force recognition with fine resolution. • Combines laser multimode interference with AI-enabled imaging processing. • Promising for large-area, fine-resolution yet low-cost touch pads. A 2D tactile sensor is demonstrated using a single winding fiber embedded in a soft, elastic silicone substrate of 6 mm thickness. Laser light at 1550 nm is injected from a single mode fiber into a spliced multimode fiber and causes multimode interference. The formed image is susceptible to any disturbance along the fiber path. These images at the output facet are captured during the automatic scanning/probing loops by a spring needle. The collected data are fed into a convolutional neural network for training, validation and test. Results show that with only a few hundred images, over 98% accuracy can be achieved in recognizing the probe spatial position within a 0.5 mm × 0.5 mm resolution. In addition, close to 100% accuracy is reached for force sensing in the third dimension at a resolution of 3 g. The presented technology may lead to the design of large-area, fine-resolution yet low-cost touch pads, thanks to its extremely simple structure and powerful interrogation method. [ABSTRACT FROM AUTHOR]

Machine learning overfitting caused by data scarcity greatly limits the application of chemical artificial intelligence in membrane materials. As the original data for thin film polyamide nanofiltration membranes is limited, here we propose to extract the natural features of monomer molecular structures and rationally distort them to augment the data availability. This few-shot learning method allows a chemical engineering project to leverage the powerful fit of deep learning without big data at the outset, which is advantageous over traditional machine learning models. The rejection and flux predictions of polyamide nanofiltration membranes are practiced by the molecular augmentation in deep learning. Convergence of loss function indicates that the model is effectively optimized. Correlation coefficients over 0.80 and the mean relative error below 5% prove an accurate prediction of nanofiltration performance. The success of predicting nanofiltration membrane performances is widely instructive for molecule and material science. Chemical distortion data augmentation renders effective deep learning for polyamide thin film composite nanofiltration membranes with limited raw dataset. Image 1 • Neural network was deployed to fit the relationship from polyamide membrane synthesis to nanofiltration performances. • Representative chemical augmentation could tackle the data scarcity that hinders deep learning of polyamide TFCMs. • In fitting TFCMs, spatial-based modeling with Neural Network is more effective than knowledge-based one with Random Forest. [ABSTRACT FROM AUTHOR]

• For the first time, neural networks were applied to denoise in spatial heterodyne spectroscopy, and significant results were achieved. • Four corresponding network models were constructed for denoising different brightness and noise levels. • The denoising ability of DnCNN in spatial heterodyne interferograms were comprehensively evaluated. • The reasonable selection of four neural network schemes were discussed. Spatial heterodyne spectroscopy is a hyperspectral remote sensing technology. With the continuous improvement of signal detection accuracy, new methods are urgently needed to further reduce the interference of noise on the information contained in spatial heterodyne interferograms. Convolutional neural networks, as an emerging field, have achieved good results in image denoising in recent years. This paper applies convolutional neural networks to the field of spatial heterodyne spectroscopy, constructs and trains four deep convolutional neural network models for denoising spatial heterodyne interferograms under different brightness and Gaussian noise conditions, and compares their performance with other algorithms. The results show that deep convolutional neural networks have significant advantages in denoising spatial heterodyne interferograms. Finally, a comparison of four neural networks was conducted to explore how to reasonably select network models. This work provides new ideas and effective solutions for reducing the interference of noise on spatial heterodyne spectral information and improving signal detection accuracy. [ABSTRACT FROM AUTHOR]

• Achieved arbitrary power ratio splitter with only 4 microheaters on a multimode waveguide. • Construct a nonlinear equivalent neural network to represent the thermally controlled multimode interference device. • Using "offline" inverse design to achieve splitting functions without tedious numerical simulations. A 3-port beam splitter with arbitrary power ratio is developed on a multimode waveguide by effectively manipulating the multimode interference through 4 locally placed microheaters. For matched interference, different light powers are coupled to the three output waveguides but the sum does not undergo extra loss, whereas the splitter integrates also the variable attenuation function, from zero to maximum, to any of the outputs. Different from the conventional design approach using extensive simulations, an equivalent neural network is established to represent such a multimode waveguide system based on experimental data only. From this neural network, an optimization program is developed to inverse the output and input. For a target power splitting ratio, the required heater currents can be readily calculated on a computer and then loaded onto the waveguide chip. The experimental results match well with the targets. This work demonstrates that the nonlinear activation is inherently integrated in the thermally controlled multimode waveguide though the optical material itself is linear. This feature can be used to construct complex deep learning networks. It may lead to the development of advanced integrated optical switches and computing networks. [ABSTRACT FROM AUTHOR]