- 标题
- 摘要
- 关键词
- 实验方案
- 产品
-
Deep learning enabled inverse design in nanophotonics
摘要： Deep learning has become the dominant approach in artificial intelligence to solve complex data-driven problems. Originally applied almost exclusively in computer-science areas such as image analysis and nature language processing, deep learning has rapidly entered a wide variety of scientific fields including physics, chemistry and material science. Very recently, deep neural networks have been introduced in the field of nanophotonics as a powerful way of obtaining the nonlinear mapping between the topology and composition of arbitrary nanophotonic structures and their associated functional properties. In this paper, we have discussed the recent progress in the application of deep learning to the inverse design of nanophotonic devices, mainly focusing on the three existing learning paradigms of supervised-, unsupervised-, and reinforcement learning. Deep learning forward modelling i.e. how artificial intelligence learns how to solve Maxwell’s equations, is also discussed, along with an outlook of this rapidly evolving research area.
关键词： forward modelling,inverse design,nanophotonics,artificial intelligence,metamaterials,machine learning
更新于2025-09-23 15:19:57
-
Determination of minor metal elements in steel using laser-induced breakdown spectroscopy combined with machine learning algorithms
摘要： The properties of a steel are crucially influenced by the contained minor elements, including metals, such as Mn, Cr and Ni. The determination of their concentrations using laser-induced breakdown spectroscopy (LIBS) represents a great help in many application scenarios, especially with in situ and online measurement requirements. Such determination can be significantly perturbed by spectral interferences with Fe I and Fe II lines which is particularly dense in the VIS and near UV ranges. Univariate regression can sometimes, lead to calibration models with modest analytical performances. In this work, multivariate calibration models are developed using a machine learning approach. We first show the regression results with univariate models. The development of multivariate models is then briefly presented, in successive steps of data pretreatment, feature selection with SelectKBest algorithm and regression model training with back-propagation neural network (BPNN). The analytical performances obtained with the developed multivariate models are compared with those obtained with the univariate models. We demonstrate in such way, the efficiency of the machine learning approach in the development of multivariate models for calibration and prediction with LIBS spectra acquired from steel samples. In particular, the prediction trueness (relative error of prediction) and precision (relative standard deviation) for the determination of the above mentioned metal elements in steel reach the respective values of 1.13%, 2.85%, 7.20% (for Mn, Cr, Ni) and 6.68%, 3.96%, 6.52% (for Mn, Cr, Ni) with the used experimental condition and measurement protocol.
关键词： Minor metal elements,LIBS,Machine learning,Multivariate regression,Steels,Spectral interference
更新于2025-09-23 15:19:57
-
Three-dimensional vectorial holography based on machine learning inverse design
摘要： The three-dimensional (3D) vectorial nature of electromagnetic waves of light has not only played a fundamental role in science but also driven disruptive applications in optical display, microscopy, and manipulation. However, conventional optical holography can address only the amplitude and phase information of an optical beam, leaving the 3D vectorial feature of light completely inaccessible. We demonstrate 3D vectorial holography where an arbitrary 3D vectorial field distribution on a wavefront can be precisely reconstructed using the machine learning inverse design based on multilayer perceptron artificial neural networks. This 3D vectorial holography allows the lensless reconstruction of a 3D vectorial holographic image with an ultrawide viewing angle of 94° and a high diffraction efficiency of 78%, necessary for floating displays. The results provide an artificial intelligence–enabled holographic paradigm for harnessing the vectorial nature of light, enabling new machine learning strategies for holographic 3D vectorial fields multiplexing in display and encryption.
关键词： manipulation,microscopy,artificial neural networks,inverse design,optical display,machine learning,3D vectorial holography
更新于2025-09-23 15:19:57
-
Photo-responsible Synapse using Ge Synaptic Transistors and GaAs Photodetectors
摘要： Current neural networks are accumulating accolades for their performance on a variety of real-world computational tasks including recognition, classification, regression, and prediction, yet there are few scalable architectures that have emerged to address the challenges posed by their computation. This paper introduces Minitaur, an event-driven neural network accelerator, which is designed for low power and high performance. As an field-programmable gate array-based system, it can be integrated into existing robotics or it can offload computationally expensive neural network tasks from the CPU. The version presented here implements a spiking deep network which achieves 19 million postsynaptic currents per second on 1.5 W of power and supports up to 65 K neurons per board. The system records 92% accuracy on the MNIST handwritten digit classification and 71% accuracy on the 20 newsgroups classification data set. Due to its event-driven nature, it allows for trading off between accuracy and latency.
关键词： Deep belief networks,spiking neural networks,field programmable arrays,restricted Boltzmann machines,neural networks,machine learning
更新于2025-09-23 15:19:57
-
Supervised Learning of Natural-Terrain Traversability with Synthetic 3D Laser Scans
摘要： Autonomous navigation of ground vehicles on natural environments requires looking for traversable terrain continuously. This paper develops traversability classifiers for the three-dimensional (3D) point clouds acquired by the mobile robot Andabata on non-slippery solid ground. To this end, different supervised learning techniques from the Python library Scikit-learn are employed. Training and validation are performed with synthetic 3D laser scans that were labelled point by point automatically with the robotic simulator Gazebo. Good prediction results are obtained for most of the developed classifiers, which have also been tested successfully on real 3D laser scans acquired by Andabata in motion.
关键词： 3D laser scanner,field robotics,sensor simulation,traversability,supervised machine learning
更新于2025-09-23 15:19:57
-
Organic Photovoltaics: Relating Chemical Structure, Local Morphology, and Electronic Properties
摘要： Substantial enhancements in the efficiencies of bulk-heterojunction (BHJ) organic solar cells (OSCs) have come from largely trial-and-error-based optimizations of the morphology of the active layers. Further improvements, however, require a detailed understanding of the relationships among chemical structure, morphology, electronic properties, and device performance. On the experimental side, characterization of the local (i.e., nanoscale) morphology remains challenging, which has called for the development of robust computational methodologies that can reliably address those aspects. In this review, we describe how a methodology that combines all-atom molecular dynamics (AA-MD) simulations with density functional theory (DFT) calculations allows the establishment of chemical structure–local morphology–electronic properties relationships. We also provide a brief overview of coarse-graining methods in an effort to bridge local to global (i.e., mesoscale to microscale) morphology. Finally, we give a few examples of machine learning (ML) applications that can assist in the discovery of these relationships.
关键词： Machine Learning,Density Functional Theory,Organic Photovoltaics,Organic Solar Cells,Bulk-Heterojunction,Electronic Properties,Coarse-Graining Methods,Local Morphology,Chemical Structure,All-Atom Molecular Dynamics
更新于2025-09-23 15:19:57
-
A deep learning-based model for defect detection in laser-powder bed fusion using in-situ thermographic monitoring
摘要： Additive manufacturing of metal components with laser-powder bed fusion is a very complex process, since powder has to be melted and cooled in each layer to produce a part. Many parameters influence the printing process; however, defects resulting from suboptimal parameter settings are usually detected after the process. To detect these defects during the printing, different process monitoring techniques such as melt pool monitoring or off-axis infrared monitoring have been proposed. In this work, we used a combination of thermographic off-axis imaging as data source and deep learning-based neural network architectures, to detect printing defects. For the network training, a k-fold cross validation and a hold-out cross validation were used. With these techniques, defects such as delamination and splatter can be recognized with an accuracy of 96.80%. In addition, the model was evaluated with computing class activation heatmaps. The architecture is very small and has low computing costs, which means that it is suitable to operate in real time even on less powerful hardware.
关键词： Additive manufacturing,Convolutional neural networks,Machine learning,Quality assurance
更新于2025-09-23 15:19:57
-
Detection of cleaning interventions on photovoltaic modules with machine learning
摘要： Soiling losses are a major concern for remote power systems that rely on photovoltaic energy. Power loss analysis is efficient for the monitoring of large power plants and for developing an optimal cleaning schedule, but it is not adapted for remote monitoring of standalone photovoltaic systems that are used in rural and poor regions. Indeed, this technique relies on a costly and dirt sensitive irradiance sensor. This paper investigates the possibility of a low-cost monitoring of cleaning interventions on photovoltaic modules during daytime. We believe that it can be helpful to know whether the soiling is regularly removed or not, and to decide if it is necessary to carry out additional cleaning operations. The problem is formulated as a classification task to automatically identify the occurrence of a cleaning intervention using a time window of temperature, voltage and current measurements of a photovoltaic array. We investigate machine learning tools based on Logistic Regression, Support Vector Machines, Artificial Neural Networks and Random Forest to achieve such classification task. In addition, we study the influence of the temporal resolution of the signals and the feature extraction on the classification performance. The experiments are conducted on a real dataset and show promising results with classification accuracy of up to 95%. Based on the results, three implementation strategies addressing different practical needs are proposed. The results may be particularly useful for non-governmental organizations, governments and energy service companies to improve the maintenance level of their photovoltaic facilities.
关键词： Detection,Machine learning,Photovoltaics,Maintenance,Monitoring,Soiling
更新于2025-09-23 15:19:57
-
Combining multimodal imaging and treatment features improves machine learning-based prognostic assessment in patients with glioblastoma multiforme
摘要： Background: For Glioblastoma (GBM), various prognostic nomograms have been proposed. This study aims to evaluate machine learning models to predict patients’ overall survival (OS) and progression‐free survival (PFS) on the basis of clinical, pathological, semantic MRI‐based, and FET‐PET/CT‐derived information. Finally, the value of adding treatment features was evaluated. Methods: One hundred and eighty‐nine patients were retrospectively analyzed. We assessed clinical, pathological, and treatment information. The VASARI set of semantic imaging features was determined on MRIs. Metabolic information was retained from preoperative FET‐PET/CT images. We generated multiple random survival forest prediction models on a patient training set and performed internal validation. Single feature class models were created including "clinical," "pathological," "MRI‐based," and "FET‐PET/CT‐based" models, as well as combinations. Treatment features were combined with all other features. Results: Of all single feature class models, the MRI‐based model had the highest prediction performance on the validation set for OS (C‐index: 0.61 [95% confidence interval: 0.51‐0.72]) and PFS (C‐index: 0.61 [0.50‐0.72]). The combination of all features did increase performance above all single feature class models up to C‐indices of 0.70 (0.59‐0.84) and 0.68 (0.57‐0.78) for OS and PFS, respectively. Adding treatment information further increased prognostic performance up to C‐indices of 0.73 (0.62‐0.84) and 0.71 (0.60‐0.81) on the validation set for OS and PFS, respectively, allowing significant stratification of patient groups for OS. Conclusions: MRI‐based features were the most relevant feature class for prognostic assessment. Combining clinical, pathological, and imaging information increased predictive power for OS and PFS. A further increase was achieved by adding treatment features.
关键词： prognostic model,machine learning,VASARI,glioblastoma,FET‐PET,biomarker,MRI
更新于2025-09-23 15:19:57
-
Remote sensing bio-control damage on aquatic invasive alien plant species
摘要： Aquatic Invasive Alien Plant (AIAP) species are a major threat to freshwater ecosystems, placing great strain on South Africa’s limited water resources. Bio-control programmes have been initiated in an effort to mitigate the negative environmental impacts associated with their presence in non-native areas. Remote sensing can be used as an effective tool to detect, map and monitor bio-control damage on AIAP species. This paper reconciles previous and current research concerning the application of remote sensing to detect and map bio-control damage on AIAP species. Initially, the spectral characteristics of bio-control damage are described. Thereafter, the potential of remote sensing chlorophyll content and chlorophyll fluorescence as pre-visual indicators of bio-control damage are reviewed and synthesised. The utility of multispectral and hyperspectral sensors for mapping different severities of bio-control damage are also discussed. Popular machine learning algorithms that offer operational potential to classify bio-control damage are proposed. This paper concludes with the challenges of remote sensing bio-control damage as well as proposes recommendations to guide future research to successfully detect and map bio-control damage on AIAP species.
关键词： machine learning algorithms,multispectral sensors,chlorophyll content,Aquatic Invasive Alien Plant (AIAP) species,chlorophyll fluorescence,hyperspectral sensors,Remote sensing,bio-control damage
更新于2025-09-23 15:19:57