Aiming　at　the　problems　in　automobile　height　regulator　production　that　manual　defect　detection　is　labor-intensive　and　time-consuming,　and　traditional　diagnosis　method　has　poor　applicability,　an　intelligent　detection　method　based　on　improved　convolution　network　is　proposed　using　deep　learning.　In　this　method,　convolution　network　is　used　to　extract　features,　and　residual　network　structure　and　separable　convolution　are　added　to　the　network,　which　improves　the　accuracy　of　deep　network　and　reduces　the　parameter　calculation　amount.　The　improved　structure　mainly　uses　convolution　layer,　pooling　layer,　batch　standardization　layer　and　softmax　layer,　and　introduces　residual　network　structure　and　separable　convolution.　The　experiment　results　show　that　the　defect　detection　method　for　automobile　height　regulator　based　on　improved　convolution　network　has　good　recognition　accuracy.　In　the　detection　experiment　on　multiple　kinds　of　defects　for　automobile　height　regulator,　the　accuracy　is　above　99%,　which　is　superior　to　that　of　the　classical　convolution　network　VGG16.　©　2020,　Science　Press.　All　right　reserved.