The human brain efficiently processes only a fraction of visual
information, a phenomenon termed attentional control, resulting in
energy savings and heightened adaptability. Translating this mechanism
into artificial visual neurons holds promise for constructing energy-efficient,
bioinspired visual systems. Here, we propose a self-rectifying artificial
visual neuron (SEVN) based on a NiO/Ga2O3 bipolar
heterojunction with attentional control on patterns with a target
color. The device exhibits short-term potentiation (STP) with quantum
point contact (QPC) traits at low bias and transitions to long-term
potentiation (LTP) at high bias, particularly facilitated by electron
capture in deep defects upon ultraviolet (UV) exposure. With the utilization
of two wavelengths of light upon the target and interference part
of CAPTCHA to simulate top-down attentional control, the recognition
accuracy is enhanced from 74 to 84%. These findings have the potential
to augment the visual capability of neuromorphic systems with implications
for diverse applications, including cybersecurity, healthcare, and
machine vision.