Efficient　in-sensor　computing　necessitates　linear,　bidirectional,　and　centrosymmetric　photoresponse　weight　updates;　however,　the　realization　of　these　attributes　poses　a　persistent　challenge,　with　most　photosensor　devices　achieving　linear　analog　weight　updates　while　falling　short　of　accomplishing　bidirectional　and　centrosymmetric　characteristics.　Here,　the　development　of　a　quantum　dot　(QD)-based　bulk　heterojunction　synaptic　transistor　(QBST)　with　multi-factor　modulation　through　surface　ligand　engineering　of　blend　QDs　is　reported.　By　controlling　the　charge　transmission　between　QDs　and　the　semiconductor,　the　QBST　device　enables　tunable　fading　memory,　which　transforms　linear　weight　updates　in　short-chain　devices　into　linear,　bidirectional,　and　unprecedented　centrosymmetric　optical　synaptic　responses　in　long-chain　devices.　Moreover,　through　the　synergy　of　chemical　and　electric　factors,　the　convolutional　kernel　of　QBSTs-based　convolutional　neural　network　realizes　enhanced　recognition　for　complex　noisy　fashion-costume　images,　achieving　an　impressive　90.3%　accuracy　in　the　long-chain　device,　highlighting　the　efficiency　of　centrosymmetric　weight　updates.　The　results　demonstrate　that　surface　ligand　engineering　offers　a　promising　approach　for　customizable　synaptic　modulation,　facilitating　energy-　and　time-efficient　in-sensor　computing.　By　modulating　the　ligand　chain　length　of　perovskite　QDs,　bulk　heterojunction　synaptic　transistors　can　achieve　multi-factor　optical　synaptic　modulation,　enabling　tunable　fading　memory.　Notably,　the　optical　synaptic　weight　transforms　linear　weight　updates　in　short-chain　devices　into　linear,　bidirectional,　and　unprecedented　centrosymmetric　optical　synaptic　responses　in　long-chain　devices,　showcasing　their　tremendous　potential　in　high-accuracy　in-sensor　computing　applications.　image