A multi-direction bending sensor based on spot pattern demodulation of a dual-hole fiber (DHF) is proposed. By using the interference and scattering in a DHF, the related multidirectional variations can be captured by the optical field. Furthermore, the multi-directional bending characteristics of the fiber are quantitatively described by the pattern of the output light spot, achieving multidirectional bending sensing. In addition, considering the subtle changes in the deformation patterns over time, a convolutional neural network (CNN) model based on deep learning is introduced for accurate recognition and prediction of the bending angle. The experimental results show that the sensor can perceive different bending angles in four directions. These outstanding results indicate that the multi-directional bending sensor based on dual-hole interference pattern decoding has potential applications in multi-directional quantitative sensing and artificial intelligence perception.

We demonstrate a deep-learning-based fiber imaging system that can transfer real-time artifact-free cell images through a meter-long Anderson localizing optical fiber. The cell samples are illuminated by an incoherent LED light source. A deep convolutional neural network is applied to the image reconstruction process. The network training uses data generated by a setup with straight fiber at room temperature (～20 ° C) but can be utilized directly for high-fidelity reconstruction of cell images that are transported through fiber with a few degrees bend or fiber with segments heated up to 50°C. In addition, cell images located several millimeters away from the bare fiber end can be transported and recovered successfully without the assistance of distal optics. We provide evidence that the trained neural network is able to transfer its learning to recover images of cells featuring very different morphologies and classes that are never “seen” during the training process.