Vehicle　attribute　recognition　in　urban　traffic　monitoring　is　the　core　task　in　urban　intelligent　transportation　system,　Vehicle　attribute　recognition　mainly　includes　sub-tasks　such　as　vehicle　type　identification,　vehicle　color　recognition,　and　vehicle　brand　recognition.　Most　of　the　current　solutions　are　single-task　learning　based　on　a　single　attribute,　and　there　are　few　studies　on　complex　learning　tasks　with　multiple　attributes.　This　paper　constructs　a　vehicle　multi-attribute　data　set　for　the　multi-attribute　characteristics　of　vehicles,　and　based　on　the　training　mode　of　multitasking　learning,　Separate　vehicle　brand　recognition　network　and　vehicle　color　recognition　network　that　are　more　suitable　for　their　respective　characteristics,　and　integrate　vehicle　multi-attribute　identification　network　into　the　same　model　structure　for　training.　By　comparing　the　current　popular　neural　network　with　several　attributes,　the　final　experiment　shows　that　the　vehicle　multi-attribute　recognition　model　trained　by　the　algorithm　can　obtain　better　recognition　results　and　higher　accuracy.　©　2019　IEEE.