Spectrometer-free time-division multiplexed NIR time-of-flight vision system for visually similar material recognition.

Conventional machine vision systems based on RGB cameras struggle to distinguish materials that appear visually identical, such as plastics of the same color and shape. To address this limitation, we present a spectrometer-free time-division multiplexed (TDM) near-infrared (NIR) time-of-flight (ToF) vision system that enables simultaneous acquisition of spectral and geometric information using dual-detector architecture composed of an avalanche photodiode (APD) for multispectral reflectance detection and a single-photon avalanche diode (SPAD) for ToF ranging. By extending TDM to multispectral NIR imaging, the proposed system temporally separates nanosecond laser pulses at 980 nm, 1450 nm, and 1650 nm for material discrimination, while an additional 905 nm channel provides high-precision ToF depth mapping. This architecture eliminates bulky spectrometers and dispersive optics, minimizing optical loss while maintaining compactness and scalability. The system successfully recognizes 12 visually similar materials, including six white plastics, three green rubbers, and three silver metals, based on their unique NIR reflectance fingerprints encoded into false-color RGB images. A convolutional neural network (CNN) trained on these images achieves near-perfect classification accuracy. Furthermore, a dual-domain experiment with a mannequin and a human subject demonstrates simultaneous reconstruction of surface geometry and material differentiation under realistic conditions. This spectrometer-free multispectral ToF vision approach establishes a compact and efficient sensing platform for high-precision robotic perception, intelligent manufacturing, and physical artificial intelligence systems requiring both spectral and spatial awareness.