Neuromorphic optoelectronic sensors with in-sensor computing architecture hold great promise for applications that require processing large amounts of redundant data, such as the Internet of Things, robotics, and environmental sciences, due to their advantages of low time latency and energy efficiency. Halide perovskites, known for their extraordinary optoelectronic properties and stimuli-responsive characteristics, offer excellent opportunities for developing switchable visual sensors with high sensitivity, fast response, low energy consumption, and wide adaptive range. In this work, we successfully realized a MAPbBr3-PdSe2 heterojunction-based optoelectronic sensor, demonstrating a responsivity of 28 mA/W and a high specific detectivity of 5.2 × 1011 Jones. A fast response time of ∼ 25 μs has also been achieved. Additionally, we investigated the role of voltage-induced ion migration in actively adjusting the device's photoresponse capacity. Under 0 V bias, the device exhibited a wide switchable range of optical responsivity from 137.5% to 27000%, significantly surpassing pure perovskite-based devices in previous reports. Moreover, we demonstrated the device's adaptive learning capabilities and reproducible switching characteristics by simulating Pavlovian classical conditioned reflex experiments using electrical modulation. These findings open up exciting possibilities for next-generation artificial vision systems that are energy-efficient, adaptive, and capable of learning and effectively responding to varying visual conditions.