Object recognition through a multi-mode fiber
Object recognition through a multi-mode fiber
复制标题
DOI：
10.1007/s10043-017-0303-5
复制
发表时间：
2017-04-01
期刊：
影响因子：
1.2
通讯作者：
Tanida, Jun
中科院分区：
文献类型：
--
作者：
Takagi, Ryosuke;Horisaki, Ryoichi;Tanida, Jun
我们提出了一种通过多模光纤识别物体的方法。通过多模光纤传输的许多散斑图案被提供给基于机器学习的分类器。我们通过实验证明了基于该方法的人脸和非人脸目标的二元分类。实验装置的测量过程是随机和非线性的，因为多模光纤是典型的强散射介质，并且我们的装置中没有使用任何参考光。还提供了支持向量机、自适应增强和神经网络这三种监督学习方法之间的比较。所有这些学习方法都实现了约 90% 的分类准确率。这里提出的方法可以实现紧凑且智能的光学传感器。它对于医疗应用（例如内窥镜检查）实际上很有用。此外，我们的研究还表明，人工智能在降低光学传感系统中的光学和计算成本方面有着快速发展的前景。
We present a method of recognizing an object through a multi-mode fiber. A number of speckle patterns transmitted through a multi-mode fiber are provided to a classifier based on machine learning. We experimentally demonstrated binary classification of face and non-face targets based on the method. The measurement process of the experimental setup was random and nonlinear because a multi-mode fiber is a typical strongly scattering medium and any reference light was not used in our setup. Comparisons between three supervised learning methods, support vector machine, adaptive boosting, and neural network, are also provided. All of those learning methods achieved high accuracy rates at about 90% for the classification. The approach presented here can realize a compact and smart optical sensor. It is practically useful for medical applications, such as endoscopy. Also our study indicated a promising utilization of artificial intelligence, which has rapidly progressed, for reducing optical and computational costs in optical sensing systems.