安徽快三

本所声明  |    |    |    |  English  |    |  

上海光机所在智能计算成像研究方面取得新进展

来源: 发布时间:2020-05-06【字体:

  近期,中国科学院上海光学精密机械研究所信息光学与光电技术实验室与德国斯图加特大学应用光学研究所、美国麻省理工学院合作,提出并实验验证了一种基于物理模型和深度神经网络的新型计算成像方法,无需大量带标签的数据来完成神经网络训练,将能有力促进人工智能技术在计算成像中的广泛应用。相关成果于5月6日在线发表于《光:科学与应用》(Light: Science & Applications)。

  jinnianlai,jiyushenduxuexidefangfabeiguangfanyingyongyujisuanchengxiangzhong,zaixiangweihuifu、shuziquanxi、danxiangsuchengxiang、sanshechengxiangdengzhongduolingyuqudeleyixilielingrenzhumudechengguo。raner,chuantongjiyushenduxuexidejisuanchengxiangfangfadadoucaiyongjianduxuexidecelve,yincixuyaoyuxianhuoqudaliangdaibiaoqiandeshujulaixunlianshenjingwangluo,qiesuohuoqushujudeshulianghezhiliangduisuodemoxingdexingnengjuyouhendayingxiang,erzaishijiyingyongzhongzheiyitiaojianwangwangshinanyimanzude。jinguanciqianbenwenjinzhandeketizuyanjiubiaoming,dangchengxiangxitongdezhengxiangwulimoxingyizhishikeyitongguofangzhenshengchengxunlianshuju,danshishenjingwangluodefanhuaxingzongshiyouxiande,suodemoxingzhinengduileisixunlianjidechangjingdedaojiaohaodejieguo。

  zhenduijiyushenduxuexidejisuanchengxiangfangfazhongxunlianshujunanyihuoquhemoxingfanhuaxingyouxiandewenti,yanjiurenyuantichujiangwulimoxingyushenjingwangluoxiangjiehedefangfa(physics-enhanced deep neural network, physennet),liyongwulimoxingtidaixunlianshujulaiqudongwangluocanshudeyouhua。xiangbichuantongshujuqudongdeduandaoduanshenduxuexifangfa,physennetwuxuhuoquxunlianshujuqieshiyizhongjuyoupushixingdefangfa。xiangbimoxingqudongdeyouhuasuanfa,physennetwuxushiyongxianshizhengzexiangjiunengyongyubingtainiwenti(congtancedaodewuliceliangzhonghuifuyuanshiwutixinxi,zaitancejieduancunzaizhuruxiangweidengxinxidiushi)deqiujie。

  yanjiurenyuanyijisuanchengxiangzhongdejingdianlizi——xiangweichengxiang——laiyanzhenglegaifangfadeyouxiaoxing,tongguobuduandiedaishishenjingwangluoshuchujieguojingyanshechuanboheceliangguocheng(wulimoxing)houjisuandedaodeyansheqiangdutuzhujianbijinshijiceliangdeyansheqiangdutu,suizhediedaidejinxing,shenjingwangluoshuchujieguoyezhujianbijinshijidaiqiuxiangweiwuti(tu1)。shiyanjieguo(tu2)biaoming,zaijinshiyongdanzhangyansheqiangdutushi,physennetdehuifuxiaoguoyouyuxuzaiduogelijiaomianzhijianlaihuidiedaidegerchberg-saxton(gs)suanfadehuifuxiaoguo,qiejiejinshuziquanxifangfadehuifuxiaoguo。gaifangfakeyingyongyuzhongduozhengxiangwulimoxingyizhidejisuanchengxiangfangfazhong。

  gaixianggongzuodedaolezhongguokexueyuanqianyankexuezhongdianyanjiujihua,zhongdezhongxin“zhongdehezuoxiaozu”heguojiazirankexuejijinzhongdaxiangmudezhichi。(xinxiguangxueyuguangdianjishushiyanshigonggao)

  

tu1 physennetyuanlitu

tu2 shiyanjieguo (a)shiyanzhuangzhitu;(b)he(g)lianggexiangweixingwutideyansheqiangdutufenbiewei(c)he(h); liyongphysennet、shuziquanxi、gsfangfahuifudejieguofenbiewei(d)he(i)、(e)he(j)、(f)he(k)

附件下载: