论文标题
具有人工神经网络的磁性纳米颗粒的光谱识别
Spectral Recognition of Magnetic Nanoparticles with Artificial Neural Networks
论文作者
论文摘要
铁磁共振（FMR）光谱是一种量化纳米颗粒中内部磁各向异性场的强大方法，这在广泛的生物医学和存储应用中很重要。但是，FMR光谱的解释只能通过使用适当的模型来实现，并且无需使用逆方法来从FMR光谱中提取内部字段。在这里，我们介绍了人工神经网络对光谱识别的使用，即从FMR光谱中识别内部磁各向异性场。我们通过馈送具有相应的各向异性场标记的预计FMR光谱的网络来培训了两种不同类型的网络，一个卷积神经网络和多层感知器。用看不见的光谱对训练有素的网络进行测试表明，它们成功地预测了正确的各向异性领域，而且令人惊讶的是，这些网络在超出其训练范围的数据方面表现良好。这些结果表明了使用人工神经网络加速对磁性材料和纳米结构的高通量分析的希望。例如，他们可以在自动化和优化探索任务中使用，在这些探索任务中，通常将纳米磁信号用作代理。
Ferromagnetic resonance (FMR) spectroscopy is a powerful method for quantifying internal magnetic anisotropy fields in nanoparticles, which is important in a wide range of biomedical and storage applications. The interpretation of FMR spectra, however, can only be achieved with the use of an appropriate model, and no inverse methods are available to extract internal fields from FMR spectra. Here, we present the use of artificial neural networks for spectral recognition, i.e., to identify the internal magnetic anisotropy field from the FMR spectrum. We trained two different types of networks, a convolutional neural network and a multi-layer perceptron, by feeding the networks pre-computed FMR spectra labeled with the corresponding anisotropy fields. Testing of the trained networks with unseen spectra showed that they successfully predict the correct anisotropy fields and, surprisingly, the networks performed well for data that was beyond their training range. These results show the promise of using artificial neural networks for accelerated high-throughput analysis of magnetic materials and nanostructures; for example they could serve in automatizing and optimizing exploration missions where nanomagnetic signals are often used as proxies.
