Bioinspired Adaptive UV-NIR Cross-Band Convolutional Fusion for Motion and Shape Recognition in Adverse Environments

Emulating biological vision remains challenging due to the absence of integrated multispectral detection, nonvolatile memory, and real-time processing in existing technologies. We propose the Bioinspired NeuroFusion Motion Intelligence (BNMI) platform, which unifies multispectral sensing, in-sensor memory, and on-device computing for reliable perception from dim light to glare. The system employs dual-electrolyte-gated In₂O₃ transistors that emulate biological photoreceptors, providing stable, tunable, bidirectional photoresponses. Its all-integrated cross-band hardware convolution kernel (AICB-Conv) performs direct ultraviolet/infrared convolution, eliminating digitization delays while preserving spectral fidelity. Hardware-level fusion yields feature-aware representations resilient to spectral degradation such as UV overexposure or NIR attenuation. Demonstrated on M3SVD, MNIST, and custom video data sets, BNMI achieves efficient feature extraction and accurate motion decoding via integrated neural readouts. These results position BNMI as a neuromorphic multispectral vision paradigm for autonomous navigation, intelligent surveillance, and edge AI.