The invention discloses an anti-
noise image recognition method and
system, and relates to the technical field of neuromorphic calculation, image recognition and brain-like hardware control, and the
system comprises an input module, a
noise estimation module, a controller module, a synaptic array module, a read-out / classification module, a near-
infrared erasure reset module and a one-way electric rewrite-in calibration module. The synaptic array has the characteristics of electrical stimulation one-way
programming-near-
infrared light erasure reset; according to the method, the
noise density of an input image is calculated, and a result and confidence are obtained through initial reasoning of a synaptic array; when the noise density is larger than or equal to a preset noise density threshold value or the confidence coefficient is smaller than or equal to a preset confidence coefficient threshold value, triggering a near-
infrared light erasing reset array, applying single-polarity electric
pulse calibration in a
dark state, and then reasoning output, a selectable grading strategy and a voting mechanism; according to the scheme, closed-loop reset calibration is formed, recognition robustness and long-term stability under medium and high noise are improved, end side
energy consumption is reduced, and the method is suitable for a low-power-consumption
visual recognition scene.