Computer　vision　systems　that　lack　color　sensitivity　may　encounter　challenges　in　object　identification,　particularly　when　the　target　object＇s　pattern　resembles　the　background.　To　enhance　neural　network　color　sensitivity　while　maintaining　low　power　consumption　and　fast　data　processing,　we　present　a　switchable　photodetection-synaptic　transistor　(SPST)　based　on　a　two-dimensional　(2D)　perovskite　ferroelectric　(PMA)2PbCl4　film　for　constructing　self-powered　optoelectronic　synapses　with　color　perception　capabilities,　with　applications　for　in-sensor　memory　and　neuromorphic　computing.　The　SPST　can　be　switched　to　photodetection　mode　or　optoelectronic　synapse　modes　by　controlling　its　gate　voltage,　and,　compared　with　the　neural　network　based　on　an　electrical　synapse,　the　processing　time　and　power　consumption　of　the　neural　network　based　on　the　SPST　are　reduced　by　56.52%　and　91.43%,　respectively.　Compared　to　the　traditional　architecture　based　on　electrical　synapses,　the　classification　accuracies　for　different　targets　improve　to　98.53%　(from　10.2%)　and　99.47%　(from　12.98%)　after　5　training　epochs.　©　2025　The　Author(s)