Abstract
Mid-infrared (MIR) intelligent sensing technology is essential for precise identification and tracking for dynamic target detection in challenging and low-visibility environments. However, existing MIR vision systems based on traditional von Neumann architecture face significant delays and inefficiencies due to the separation of sensing, memory, and processing units. Neuromorphic motion devices offer better tracking capabilities, but most studies are limited to the near-infrared spectrum. Inspired by the fire beetle's MIR sensing capabilities, we have developed an MIR neuromorphic device using a 2D inorganic/organic heterostructure. The device exhibits biological synaptic behavior in the MIR region (up to 4.25 μm) based on the persistent photoconductivity (PPC) effect, successfully realizing the function of dynamic trajectories memorization with real-time hardware implementation. Additionally, a reservoir computing (RC) system trained on an MIR flame motion dataset achieves a recognition accuracy of 94.79% in classifying flame motion direction. While the research on MIR neuromorphic devices is limited, this study underscores the potential of such devices to advance MIR-based machine vision applications.