House building extraction from high-resolution remote sensing images based on IEU-Net

Abstract

æˆ¿å±‹å»ºç­‘ç‰©ä½œä¸ºäººç±»æ´»åŠ¨çš„ä¸»è¦åœºæ‰€ï¼Œå¿«é€Ÿå‡†ç¡®åœ°å°†å ¶ä»Žé«˜åˆ†è¾¨çŽ‡é¥æ„Ÿå½±åƒä¸­æå–å‡ºæ¥ï¼Œå¯¹ä¿ƒè¿›é¥æ„Ÿä¿¡æ¯åœ¨é˜²ç¾å‡ç¾ã€åŸŽé•‡ç®¡ç†ç­‰æ–¹é¢çš„åº”ç”¨å ·æœ‰é‡è¦æ„ä¹‰ã€‚æœ¬æ–‡åŸºäºŽæ·±åº¦å­¦ä¹ ï¼Œæå‡ºäº†é«˜åˆ†è¾¨çŽ‡é¥æ„Ÿå½±åƒæˆ¿å±‹å»ºç­‘ç‰©åƒç´ çº§ç²¾ç¡®æå–æ–¹æ³•ã€‚é¦–å ˆï¼Œé’ˆå¯¹æ ·æœ¬å›¾åƒè¾¹ç¼˜åƒç´ ç‰¹å¾ä¸è¶³çŽ°è±¡ï¼Œä»¥U-Net模型为基础提出IEU-Netæ¨¡åž‹ï¼Œè®¾è®¡äº†å ¨æ–°çš„å¿½ç•¥è¾¹ç¼˜äº¤å‰ç†µå‡½æ•°IELosså¹¶å°†å ¶ä½œä¸ºæŸå¤±å‡½æ•°ï¼Œå¦å¤–æ·»åŠ Dropout和BNå±‚åœ¨é¿å è¿‡æ‹Ÿåˆçš„åŒæ—¶æé«˜æ¨¡åž‹è®­ç»ƒé€Ÿåº¦å’Œé²æ£’æ€§ã€‚å ¶æ¬¡ï¼Œä¸ºè§£å†³æ¨¡åž‹ç‰¹å¾ä¸°å¯Œåº¦æœ‰é™çš„é—®é¢˜ï¼Œå¼•å ¥å½¢æ€å­¦å»ºç­‘ç‰©æŒ‡æ•°MBI，与遥感影像RGB波段一同参与到模型的分类过程。最后，在模型预测时与IELossç›¸å¯¹åº”é‡‡ç”¨å¿½ç•¥è¾¹ç¼˜é¢„æµ‹ç­–ç•¥ä»Žè€ŒèŽ·å¾—æœ€ä½³å»ºç­‘ç‰©æå–ç»“æžœã€‚å®žéªŒå¯¹æ¯”åˆ†æžè¡¨æ˜Žï¼šæœ¬æ–‡æ–¹æ³•èƒ½æœ‰æ•ˆå ‹æœæ ·æœ¬è¾¹ç¼˜åƒç´ ç‰¹å¾ä¸è¶³é—®é¢˜å¹¶æŠ‘åˆ¶é“è·¯ã€å»ºç­‘ç‰©é˜´å½±å¯¹ç»“æžœçš„å½±å“ï¼Œæå‡é«˜åˆ†è¾¨çŽ‡é¥æ„Ÿå½±åƒä¸­æˆ¿å±‹å»ºç­‘ç‰©çš„æå–ç²¾åº¦ã€‚