3D　reconstruction　is　crucial　in　computer　vision,　especially　in　medical　imaging　and　human-computer　interaction.　However,　traditional　reconstruction　methods　face　challenges　like　energy　inefficiency　and　memory　limitations　due　to　the　storage-computation-separated　architecture.　Neuromorphic　devices,　inspired　by　the　brain＇s　architecture,　offer　a　solution　for　efficient　data　processing.　Electrical　output　synapse　devices　for　3D　reconstruction　face　delays　in　coloring　point　clouds　after　depth　processing,　leading　to　errors.　In　this　work,　a　co-planar　quantum　dot　(QD)　light-emitting　synapse　is　proposed　for　high-precision　3D　reconstruction.　By　using　the　light-emitting　synapse,　handwritten　digit　recognition　achieved　92.35%　accuracy　in　just　20　epochs.　Depth　and　grayscale　information　are　independently　processed　through　electrical　and　optical　outputs,　allowing　for　parallel　processing　that　enhances　reconstruction　quality.　This　method　decreases　losses　by　46.3%　and　reduces　the　reconstruction　pixel　error　rate　by　over　21%　in　comparison　to　the　single　output　approach.　This　study　demonstrates　significant　potential　of　light-emitting　synapses　in　computer　vision　applications.