A　bolt　detection　and　precise　positioning　method　combining　neural　network　and　RGB-D　camera　is　proposed　for　the　disassembly　and　installation　task　of　distribution　line　insulators,　which　is　used　to　assist　the　robot　to　quickly　determine　the　accurate　location　of　insulator　bolts.　An　improved　YOLOv5s　network　is　used　to　detect　the　target　bolt,　after　which　the　bolt　end　center　is　predicted　by　a　bolt　end　center　prediction　network,　and　finally　the　three-dimensions　coordinates　of　the　bolt　end　center　are　obtained　by　combining　the　depth　information　of　the　RGB-D　camera.　The　improved　YOLOv5s　reduces　the　detection　time　of　a　single　image　by　25.3%　and　reduces　the　number　of　parameters　by　22.2%.　The　average　prediction　error　of　the　bolt　end　center　prediction　network　is　8.6　pixels,　and　the　maximum　prediction　error　is　12.4　pixels.　Combined　with　the　RGB-D　camera　for　testing,　the　XY　plane　positioning　error　is　no　more　than　4.2mm,　and　the　depth　error　is　no　more　than　2mm.　At　experimental　site,　the　method　is　tested　on　a　live　working　robot　and　meets　the　requirements　of　practical　operations.　©　2023　IEEE.