基于改进U-Net模型的筏式养殖区提取分析

打开文本图片集

中图分类号：S967 文献标志码：A 文章编号：2096-9902（2025）23-0026-06

Abstract：Withtheincreaseinpopulationandtherisingdemandforseafood，marineaquaculture，especiallraftfarming， hasexpandedrapidly，whichhassubsequentlyledtomanyenvironmentalproblems.Thetraditionaldatacolectionmethodsfor aquacultureareineficientandcannotmetthedemands.Thisstudyaddressestheextractionprobleminofshoreraftfarming areasandproposesanextractionmethod basedontheU-Netdeplearning model.Taking Dalian Cityasanexample，basedon sateliteimages，thetransferlearingenhancesthegeneralizationabilityofthemodel，adoptsadynamicleaningrateadjustment mechanismtoimprove trainingeficiency，andintroducesanatentionmechanismtofocusthemodelonkeyareas.The experimentsshowthatthemodelperformsexcelentlyinextractingtheraftfarmingareasinDalianCity，withaDicecefficient of0.9572.TheIUandDicecoeficientsofthetestsetimagsarealsogoodverifyingthestrongadaptabilityandrobustsof themodelfordiferentfarmingdensityareas.Thisstudyprovidesaneficientandaccuratemethodformonitoringmarine aquaculture and offersa scientific reference for related research.

Keywords： deep learning; U-Net model; raft breeding area; transfer learning;attention mechanism

近年来，随着人口增长及人们对优质海产品需求的不断攀升，近海养殖作为渔业生产的关键构成部分，迎来了迅猛发展期。（剩余9213字）