Accurate　control　of　delta　phase　content,　distribution,　and　morphology　is　crucial　for　guiding　the　processing　schemes　to　form　an　alloy　with　desirable　properties.　Traditional　methods　for　identifying　the　delta　phase　are　challenging,　labor-intensive,　and　time-consuming.　Therefore,　we　proposed　an　end-to-end　convolutional　neural　network-based　framework　for　automatic　and　accurate　delta　phase　detection　and　area　estimation.　The　framework　consists　of　a　se-mantic　segmentation　network　and　a　regression　network.　This　model　was　trained　using　1966　scanning　electron　microscope　images　based　on　transfer　learning　and　resulted　in　1.968　of　mean　absolute　error　when　applied　to　the　unforeseen　images　with　the　compressive　strain　of　0.1.　As　such,　the　proposed　model,　performing　better　than　other　baseline　models,　was　helpful　for　segmenting　and　identifying　the　delta　phase　in　grain　boundaries　accurately　and　automatically.　The　introduced　approach　provides　an　alternative　route　for　analyzing　the　delta　phase,　which　allows　for　real-time　phase　monitoring　and　downstream　guiding　alloy　design.