With　the　rapid　development　of　artificial　intelligence　technology,　traditional　machine　vision　systems　are　facing　challenges　due　to　high　power　consumption,　high　latency,　and　low　efficiency　of　redundant　data　processing.　On　the　other　hand,　with　traditional　vision　sensors　it　is　difficult　to　obtain　clear　images　in　very　bright　or　dark　conditions.　In　this　paper,　we　propose　a　bionic　vision　sensor　based　on　a　back-to-back　structure,　which　realizes　efficient　optical　information　processing　by　simulating　the　adaptive　mechanism　of　biological　vision　system.　The　device　combines　the　carrier　trapping　property　of　polymethyl　methacrylate　with　heterogeneous　interface　and　the　photosensitive　property　of　quantum　dots　to　simulate　the　biological　synaptic　behaviors,　such　as　excitatory　postsynaptic　current,　paired　pulse　facilitation,　short-term　plasticity,　and　long-term　plasticity.　Under　dynamic　light　intensities　(198.6　mu　W　cm-2　to　8.32　mW　cm-2),　the　device　exhibits　remarkable　visual　adaptive　behaviors:　enhancement　of　the　current　response　by　carrier　accumulation　in　low　light,　and　desensitization　by　inhibition　of　carrier　migration　through　the　interfacial　electric　field　in　strong　light,　with　an　adaptation　rate　(＜40　s)　significantly　better　than　that　of　biological　systems　(2-30　min).　The　device　array　further　enhances　the　contrast　of　its　images　in　extreme　light　and　the　recognition　accuracy　in　combination　with　artificial　neural　network　is　improved　to　73.7%　(dark-adapted)　and　87.5%　(light-adapted).　The　device　provides　a　prospective　strategy　for　the　development　of　next　generation　of　bionic　vision　devices,　which　have　great　potential　for　applications　in　fields　such　as　smart　driving　and　brain-like　computing.