基于高分辨率遙感圖像的荔枝林樹(shù)冠信息提取方法研究

姜仁榮　汪春燕　沈利強(qiáng)　王培法

(1.深圳市規(guī)劃國(guó)土發(fā)展研究中心， 深圳 518040； 2.國(guó)土資源部城市土地資源監(jiān)測(cè)與仿真重點(diǎn)實(shí)驗(yàn)室， 深圳 518040；>3.南京信息工程大學(xué)地理與遙感學(xué)院， 南京 210044)

?

基于高分辨率遙感圖像的荔枝林樹(shù)冠信息提取方法研究

姜仁榮1,2汪春燕1沈利強(qiáng)1王培法3

(1.深圳市規(guī)劃國(guó)土發(fā)展研究中心， 深圳 518040； 2.國(guó)土資源部城市土地資源監(jiān)測(cè)與仿真重點(diǎn)實(shí)驗(yàn)室， 深圳 518040；>3.南京信息工程大學(xué)地理與遙感學(xué)院， 南京 210044)

為有效提取荔枝林樹(shù)冠信息，解決局部最大值法窗口選擇和區(qū)域生長(zhǎng)法在樹(shù)冠相互連接時(shí)的過(guò)度生長(zhǎng)問(wèn)題，將水文分析和區(qū)域生長(zhǎng)融合方法用于荔枝單木探測(cè)和樹(shù)冠描繪。首先將均值濾波方法平滑后的全色圖像進(jìn)行反轉(zhuǎn)完成圖像預(yù)處理；然后對(duì)預(yù)處理后圖像提取洼地和洼地貢獻(xiàn)區(qū)域，接著剔除錯(cuò)提洼地，合并樹(shù)冠分支洼地的貢獻(xiàn)區(qū)域，從而提取樹(shù)頂位置，完成單木探測(cè)；最后以單木探測(cè)結(jié)果為種子點(diǎn)，采用區(qū)域生長(zhǎng)方法對(duì)樹(shù)冠進(jìn)行描繪，種子生長(zhǎng)被限定在洼地貢獻(xiàn)區(qū)域內(nèi)，在閾值控制下進(jìn)行生長(zhǎng)，最終完成單木樹(shù)冠描繪。采用遙感分類精度評(píng)價(jià)指標(biāo)對(duì)提取結(jié)果進(jìn)行評(píng)價(jià)，單木探測(cè)總體精度為87.75%，用戶精度為80.69%，生產(chǎn)者精度為96.06%；單木樹(shù)冠描繪總體精度為78.69%，用戶精度為71.32%，生產(chǎn)者精度為87.76%。

荔枝; 高分辨率遙感圖像; 單木探測(cè); 單木樹(shù)冠描繪; 水文分析; 區(qū)域增長(zhǎng)

引言

荔枝是熱帶亞熱帶水果，是廣東省特色農(nóng)作物，高分辨遙感圖像的出現(xiàn)使精細(xì)監(jiān)測(cè)荔枝種植成為可能。在高分辨率遙感圖像上，樹(shù)木樹(shù)冠清晰可辨，通過(guò)提取單木樹(shù)冠信息可獲取包括種植數(shù)量、種植密度、樹(shù)冠面積、郁閉度、估測(cè)胸徑和生物量等指標(biāo)，但單木樹(shù)冠信息提取仍然是一個(gè)難題[1]。

單木樹(shù)冠信息的半自動(dòng)或自動(dòng)提取可追溯到20世紀(jì)90年代[2]，目前已有多種提取方法。通常樹(shù)冠信息提取由2步構(gòu)成，首先是單木探測(cè)，獲取樹(shù)頂或樹(shù)的位置；然后是單木樹(shù)冠描繪，自動(dòng)獲取樹(shù)冠邊界。單木探測(cè)方法主要有局部最大值法[3-7]、多尺度分析法[8-11]和模板匹配法[12-14]; 單木樹(shù)冠描繪方法主要有谷地跟蹤法[15-17]、區(qū)域增長(zhǎng)[5,18-21]、分水嶺分割[22-24]和局部射線法[4,25]。隨著數(shù)據(jù)獲取和信息提取技術(shù)的發(fā)展，有研究者將激光雷達(dá)和高分辨率雷達(dá)數(shù)據(jù)引入單木樹(shù)冠信息提取中[26-27]，也有研究者直接基于聚類分割提取樹(shù)冠信息[28]。

荔枝樹(shù)冠通常表現(xiàn)為中心高、四周低的近半圓球形，在高分辨率遙感圖像上表現(xiàn)為中心值高、向陽(yáng)面值高、背陰面值低的特征。荔枝的樹(shù)冠特征適合利用局部最大值法進(jìn)行單木探測(cè)，以及利用區(qū)域增長(zhǎng)法進(jìn)行單木樹(shù)冠描繪。但由于荔枝樹(shù)齡和種植密度不同，荔枝樹(shù)冠常存在大小不一、樹(shù)冠相互連接的現(xiàn)象，這導(dǎo)致使用局部最大值法進(jìn)行單木探測(cè)時(shí)難以確定最優(yōu)搜索窗口，使用區(qū)域增長(zhǎng)法進(jìn)行樹(shù)冠描繪時(shí)容易導(dǎo)致過(guò)度生長(zhǎng)，使若干相互連接的樹(shù)冠錯(cuò)誤地被提取為單木樹(shù)冠。因此本文針對(duì)上述問(wèn)題，將水文分析和區(qū)域生長(zhǎng)融合用于荔枝單木樹(shù)冠探測(cè)和樹(shù)冠描繪。

1　實(shí)驗(yàn)區(qū)及數(shù)據(jù)

實(shí)驗(yàn)區(qū)位于深圳市寶安區(qū)鐵崗水庫(kù)旁的一處道路分隔的荔枝園，果園中心坐標(biāo)為22°38′24.00″N、113°53′19.60″E，面積為30 183.42 m2。實(shí)驗(yàn)采用法國(guó)Pléiades高分辨率衛(wèi)星于2012年11月6日獲取的全色波段數(shù)據(jù)，分辨率為0.5 m，圖像尺寸324像素×483像素，如圖1所示。圖中四周較亮柵格為道路，果園中較亮柵格則為建筑物和園內(nèi)小路，荔枝表現(xiàn)為顆粒狀近半圓球形，樹(shù)冠大小不一，且有部分荔枝樹(shù)冠連接在一起，與背景相比，荔枝樹(shù)的光譜反射值較高，荔枝樹(shù)陰影的光譜反射值較低。

圖1　研究區(qū)圖像Fig.1　Image of study area

2　提取方法

荔枝樹(shù)冠相比非樹(shù)冠的背景區(qū)域，在遙感圖像上表現(xiàn)為較高像素值，比道路、裸土類地物像素值低，樹(shù)冠樹(shù)頂區(qū)域像素值相對(duì)較高。針對(duì)荔枝在全色圖像上的光譜響應(yīng)特點(diǎn)，將水文分析方法和區(qū)域增長(zhǎng)融合方法用于荔枝單木樹(shù)冠探測(cè)和樹(shù)冠描繪。首先對(duì)圖像進(jìn)行平滑濾波后反轉(zhuǎn)，將樹(shù)頂提取的局部最大值問(wèn)題轉(zhuǎn)換為局部最小值問(wèn)題，將圖像看作地面高程數(shù)據(jù)，從而利用水文分析洼地提取方法進(jìn)行樹(shù)頂位置的提取,完成單木探測(cè)；然后采用區(qū)域生長(zhǎng)方法進(jìn)行樹(shù)冠描繪，將單木探測(cè)結(jié)果作為樹(shù)冠描繪的種子點(diǎn),對(duì)應(yīng)的洼地貢獻(xiàn)區(qū)域作為區(qū)域生長(zhǎng)時(shí)的生長(zhǎng)范圍限定，根據(jù)給定的閾值進(jìn)行樹(shù)冠描繪，最后進(jìn)行精度評(píng)價(jià)。提取方法流程如圖2所示，首先是圖像預(yù)處理，然后是單木探測(cè)，最后是單木樹(shù)冠描繪。

圖2　研究方法流程圖Fig.2　Flow chart of the proposed method

2.1圖像預(yù)處理

圖像預(yù)處理包括圖像平滑和圖像反轉(zhuǎn)，具體步驟如下：

(1)為避免圖像高值噪聲對(duì)樹(shù)頂提取的影響，采用3×3窗口對(duì)圖像進(jìn)行均值濾波。

(2)然后利用濾波后的圖像最大值減去圖像原值將圖像反轉(zhuǎn)，使樹(shù)頂亮值變?yōu)榫植堪抵?，將?shù)頂?shù)木植孔畲笾堤崛?wèn)題轉(zhuǎn)換為局部最小值提取問(wèn)題。

將反轉(zhuǎn)后圖像看作是地面高程數(shù)據(jù)，利用水文分析的洼地提取方法解決局部最小值提取問(wèn)題，完成樹(shù)頂信息的提取。

2.2單木探測(cè)

通過(guò)探測(cè)樹(shù)頂位置來(lái)完成單木探測(cè)，利用水文分析的洼地提取方法初步提取樹(shù)頂位置，然后剔除掉非樹(shù)冠地物導(dǎo)致的誤提取樹(shù)頂位置和由于樹(shù)冠分支導(dǎo)致的多提取樹(shù)頂位置，從而獲得最終的樹(shù)頂位置，完成單木探測(cè)，同時(shí)獲取洼地貢獻(xiàn)區(qū)域用作后續(xù)利用區(qū)域生長(zhǎng)方法進(jìn)行樹(shù)冠描繪時(shí)的生長(zhǎng)范圍限定。單木探測(cè)的具體過(guò)程如下：

(1)樹(shù)頂位置初步提取

樹(shù)頂位置的初步提取是利用水文分析的洼地提取方法完成，包括水流方向計(jì)算和洼地提取。從反轉(zhuǎn)后圖像獲取局部最小值，即洼地所在像素，需對(duì)圖像進(jìn)行水流方向計(jì)算，獲取水流方向柵格。水流方向是水流離開(kāi)每一個(gè)柵格單元的指向，其計(jì)算采用D8單流向算法[29]，通過(guò)計(jì)算中心柵格與鄰域柵格的最大距離權(quán)落差(指中心柵格與鄰域柵格的高程差除以兩柵格間的距離)來(lái)確定。洼地是水流無(wú)法流出的區(qū)域，根據(jù)水流方向計(jì)算結(jié)果則可提取洼地位置。利用ArcGIS水文分析模塊的洼地提取功能提取的洼地如圖3中黃色、黑色和紅色符號(hào)所示，獲取的洼地既有單柵格，也有多個(gè)相鄰柵格。這些提取的洼地對(duì)應(yīng)原始圖像上的局部最大值點(diǎn)，即樹(shù)頂位置，提取的洼地位置則是初步提取的樹(shù)頂位置。通過(guò)反轉(zhuǎn)圖像提取洼地來(lái)獲取樹(shù)頂位置的方法，避免了局部最大值方法中的搜索窗口問(wèn)題。

(2)非樹(shù)頂位置洼地剔除

通過(guò)洼地提取方法初步確定了樹(shù)頂位置，但荔枝林中經(jīng)常會(huì)存在看護(hù)管理用房和道路，且選擇的實(shí)驗(yàn)圖像也有這些地物，由于這些地物具有較強(qiáng)的光譜反射能力，在原始圖像上表現(xiàn)為高亮柵格，當(dāng)圖像反轉(zhuǎn)后則是洼地區(qū)域，會(huì)被誤提取為樹(shù)頂信息。為此需要把這些非樹(shù)頂位置的洼地(由于這些洼地在原始圖像中為高亮柵格，以下簡(jiǎn)稱這些洼地為高亮洼地)剔除。首先計(jì)算洼地柵格所對(duì)應(yīng)原始圖像的像素值，根據(jù)道路和房屋地物對(duì)應(yīng)洼地的像素值特征設(shè)定閾值，然后將大于閾值的作為高亮洼地予以刪除。經(jīng)過(guò)對(duì)實(shí)驗(yàn)圖像分析，確定閾值為42，刪除的高亮洼地基本位于四周道路、林內(nèi)建筑物和林內(nèi)道路上，如圖3中紅色符號(hào)所示。

(3)樹(shù)冠分支樹(shù)頂剔除

有些荔枝樹(shù)樹(shù)冠較大，分支較多，導(dǎo)致一個(gè)樹(shù)冠提取了多個(gè)局部最大值，即反轉(zhuǎn)圖像后的洼地。為去除小的樹(shù)冠分支對(duì)應(yīng)的洼地，并限定后續(xù)單木樹(shù)冠生長(zhǎng)范圍，提取了洼地貢獻(xiàn)區(qū)域。洼地貢獻(xiàn)區(qū)域是將提取的洼地作為流域出水口，所有水流流向洼地的區(qū)域，這個(gè)區(qū)域則是包含提取目標(biāo)樹(shù)冠和目標(biāo)背景的區(qū)域，是洼地的匯水區(qū)域，對(duì)原始圖像來(lái)說(shuō)則是樹(shù)頂所對(duì)應(yīng)局部最大值的區(qū)域。由于樹(shù)冠分支通常距離較近，計(jì)算洼地與相鄰?fù)莸氐淖钹徑嚯x，將此距離小于閾值的看作是樹(shù)冠分支洼地，將分支洼地中原始像素均值大的洼地(即樹(shù)冠樹(shù)頂)保留，像素均值小的洼地(即樹(shù)冠分支頂部)刪除，同時(shí)合并對(duì)應(yīng)的洼地貢獻(xiàn)區(qū)域用于后續(xù)樹(shù)冠描繪。經(jīng)過(guò)計(jì)算，洼地間平均最鄰近距離為3.89 m，荔枝的建議種植密度為3 m×3 m，因此設(shè)定洼地間距離小于3 m的洼地為分支洼地，將分支洼地中像素均值小的洼地刪除，刪除的洼地如圖3中黑色面狀符號(hào)所示。利用ArcGIS水文分析模塊和制圖綜合工具集完成了上述處理，提取的單木樹(shù)頂位置如圖3中黃色面狀符號(hào)所示，合并后的洼地貢獻(xiàn)區(qū)域如圖3中中空面狀符號(hào)所示。

圖3　單木探測(cè)結(jié)果Fig.3　Result of individual treetop detection

2.3單木樹(shù)冠描繪

采用改進(jìn)的區(qū)域生長(zhǎng)方法完成單木樹(shù)冠描繪，即樹(shù)冠輪廓描繪。區(qū)域生長(zhǎng)是一種根據(jù)事先定義的準(zhǔn)則將像素或子區(qū)域聚合成更大區(qū)域的過(guò)程，通常以一組種子點(diǎn)開(kāi)始，將與種子性質(zhì)相似的相鄰像素合并到種子上。區(qū)域生長(zhǎng)法用于樹(shù)冠描繪時(shí)，因樹(shù)冠相互連接容易導(dǎo)致區(qū)域過(guò)度生長(zhǎng)，需要對(duì)區(qū)域生長(zhǎng)進(jìn)行生長(zhǎng)范圍空間限定，因此提出生長(zhǎng)范圍限定的區(qū)域生長(zhǎng)算法，算法步驟如下：

(1)將前述探測(cè)的單木位置像素，即最終提取的洼地作為區(qū)域生長(zhǎng)的種子點(diǎn)，將對(duì)應(yīng)的洼地貢獻(xiàn)區(qū)域作為種子生長(zhǎng)的限定范圍。

(2)種子點(diǎn)周圍鄰域柵格作為生長(zhǎng)候選柵格，對(duì)每一候選柵格進(jìn)行判斷。若候選柵格在生長(zhǎng)范圍內(nèi)且符合設(shè)定的樹(shù)冠生長(zhǎng)閾值，則將此柵格合并到種子點(diǎn)中，若不能同時(shí)滿足上述條件，則對(duì)此柵格進(jìn)行標(biāo)記，不合并到種子點(diǎn)中去，且不參與后續(xù)的生長(zhǎng)過(guò)程，重復(fù)以上過(guò)程，直到遍歷所有候選柵格。

(3)重復(fù)步驟(2)的生長(zhǎng)過(guò)程，直到?jīng)]有符合條件的柵格為止，則完成了一個(gè)種子點(diǎn)的生長(zhǎng)過(guò)程，實(shí)現(xiàn)了一個(gè)樹(shù)冠的描繪。

(4)對(duì)每一種子點(diǎn)重復(fù)步驟(2)、(3)，直到完成所有種子點(diǎn)的生長(zhǎng)，即完成了全部樹(shù)冠描繪。

在ArcGIS中用VBA實(shí)現(xiàn)了上述算法，對(duì)樹(shù)冠與樹(shù)頂像素的數(shù)值差異進(jìn)行分析，設(shè)定差值7為生長(zhǎng)閾值，單木樹(shù)冠描繪結(jié)果如圖4所示。

圖4　單木樹(shù)冠描繪結(jié)果Fig.4　Result of individual tree-crown delineation

3　實(shí)驗(yàn)結(jié)果及精度評(píng)價(jià)

實(shí)驗(yàn)結(jié)果如圖3、4所示，從圖3可以看出提取的樹(shù)頂位置基本位于樹(shù)冠亮度大的區(qū)域，受剔除亮值洼地的影響，亮值地物旁有些荔枝位置未被提??；從圖4樹(shù)冠描繪結(jié)果可以看出，描繪的樹(shù)冠基本能夠覆蓋荔枝樹(shù)冠范圍，但提取的樹(shù)冠形狀不規(guī)則。為定量評(píng)價(jià)提取精度，參考實(shí)驗(yàn)區(qū)全色波段和多光譜波段融合數(shù)據(jù)，對(duì)樹(shù)冠信息進(jìn)行了數(shù)字化處理，數(shù)字化結(jié)果如圖5所示。

圖5　單木樹(shù)冠描繪參考數(shù)據(jù)Fig.5　Reference data of tree-crown

3.1精度評(píng)價(jià)方法

從單木探測(cè)和單木描繪兩方面，將遙感分類精度評(píng)價(jià)方法應(yīng)用于樹(shù)冠描繪信息提取評(píng)價(jià)中，主要評(píng)價(jià)指標(biāo)為用戶精度、錯(cuò)提誤差、生產(chǎn)者精度、漏提誤差和總體精度，計(jì)算公式分別為

Au=Nuc/Ne×100%

(1)

Ec=Nuw/Ne×100%=1-Au

(2)

Ap=Npc/Nr×100%

(3)

Eo=Npo/Nr×100% = 1-Ap

(4)

Ao= (Nuc+Npc)/(Ne+Nr)×100%

(5)

式中Au——用戶精度，%

Nuc——用戶角度正確提取數(shù)量

Ne——提取數(shù)量Ec——錯(cuò)提誤差，%

Nuw——用戶角度錯(cuò)誤提取數(shù)量

Ap——生產(chǎn)者精度，%

Npc——生產(chǎn)者角度正確提取數(shù)量

Nr——參考數(shù)量Eo——漏提誤差，%

Npo——生產(chǎn)者角度未提取數(shù)量

Ao——總體精度，%

對(duì)單木探測(cè)精度評(píng)價(jià)而言，Nuc和Nuw分別指提取的單木樹(shù)冠包含和未包含參考單木位置點(diǎn)的樹(shù)冠數(shù)量；Npc和Npo分別指參考的單木樹(shù)冠范圍內(nèi)包含和未包含提取的單木位置的樹(shù)冠數(shù)量，Ne是本文方法提取的樹(shù)數(shù)量，Nr指數(shù)字化的單木數(shù)量。以上各指標(biāo)用于單木探測(cè)精度評(píng)價(jià)時(shí)單位為棵。

對(duì)單木樹(shù)冠描繪精度評(píng)價(jià)而言，提取樹(shù)冠面積為Nc，若提取樹(shù)冠與參考樹(shù)冠相交，則重疊面積為Nuc，提取樹(shù)冠中未重疊樹(shù)冠面積為Nuw。類似地，參考樹(shù)冠面積為Nr，若參考樹(shù)冠與提取樹(shù)冠相交，則重疊面積為Npc，參考樹(shù)冠中未重疊區(qū)域面積為Npo。以上各指標(biāo)用于樹(shù)冠描繪精度評(píng)價(jià)時(shí)單位為m2。

3.2精度評(píng)價(jià)結(jié)果

將提取的數(shù)據(jù)和參考數(shù)據(jù)作空間統(tǒng)計(jì)分析，得到單木探測(cè)精度評(píng)價(jià)如表1所示，單木樹(shù)冠描繪精度評(píng)價(jià)如表2所示。

表1　單木探測(cè)精度評(píng)價(jià)Tab.1　Accuracy assessment of individual tree detection

從表1可看出：總共提取的單木數(shù)量為777棵，正確提取的數(shù)量為627棵，其中9個(gè)樹(shù)冠描繪過(guò)大，每個(gè)樹(shù)冠包含了2個(gè)參考樹(shù)冠的范圍；參考樹(shù)冠對(duì)象為660個(gè)，其中漏提的是26棵，可以正確識(shí)別的是634棵。但有47個(gè)參考樹(shù)冠對(duì)象包含多于一個(gè)的單木數(shù)量，這主要是由于荔枝樹(shù)冠分支導(dǎo)致的多個(gè)樹(shù)頂?shù)奶崛?。從總體上而言，總體精度為87.75%，說(shuō)明總體上單木探測(cè)精度較高。

從表2可以看出：本文方法描繪的樹(shù)冠面積大于數(shù)字化的樹(shù)冠面積，導(dǎo)致用戶角度錯(cuò)誤提取數(shù)量較大，說(shuō)明方法在描繪樹(shù)冠時(shí)比實(shí)際樹(shù)冠稍大些。從總體上來(lái)說(shuō)，圖像中78.69%的樹(shù)冠可被有效描繪。

表2　單木樹(shù)冠描繪精度評(píng)價(jià)Tab.2　Accuracy assessment of individual tree-crown delineation

總體上荔枝單木樹(shù)冠信息提取精度較好，但也存在漏提和錯(cuò)提現(xiàn)象，將提取結(jié)果和參考數(shù)據(jù)進(jìn)行對(duì)比分析發(fā)現(xiàn)：?jiǎn)文咎綔y(cè)結(jié)果中的漏提主要是由于剔除高亮洼地而導(dǎo)致在高亮洼地貢獻(xiàn)區(qū)域內(nèi)的荔枝無(wú)法提取，錯(cuò)提主要是由于荔枝林中空地引起；樹(shù)冠描述中的漏提主要是單木探測(cè)漏提導(dǎo)致，錯(cuò)提主要有兩方面原因：一方面來(lái)自于單木探測(cè)的錯(cuò)提導(dǎo)致進(jìn)行了樹(shù)冠描繪，另一方面來(lái)自于本文方法樹(shù)冠描繪邊界和參考樹(shù)冠數(shù)據(jù)的差異。

4　結(jié)束語(yǔ)

將水文分析和區(qū)域生長(zhǎng)融合方法用于單木探測(cè)和樹(shù)冠描繪，避免了局部最大值法的窗口選擇，解決了區(qū)域生長(zhǎng)法在樹(shù)冠相互連接時(shí)的過(guò)度生長(zhǎng)問(wèn)題。將反轉(zhuǎn)圖像看作地面高程數(shù)據(jù)，使局部最大值問(wèn)題轉(zhuǎn)換為局部最小值問(wèn)題。利用水文分析的洼地提取方法進(jìn)行單木探測(cè)；將洼地貢獻(xiàn)區(qū)域作為樹(shù)冠描繪區(qū)域生長(zhǎng)時(shí)的生長(zhǎng)范圍限定，提取的洼地作為種子點(diǎn)，根據(jù)給定閾值，用區(qū)域生長(zhǎng)法對(duì)樹(shù)冠進(jìn)行描繪。結(jié)果單木探測(cè)的總體精度為87.75%，單木樹(shù)冠描繪總體精度為78.69%，總體上本文方法提取效果較好，可有效探測(cè)單木和描繪樹(shù)冠。

1劉曉雙，黃建文，鞠洪波. 高空間分辨率遙感的單木樹(shù)冠自動(dòng)提取方法與應(yīng)用[J]. 浙江林學(xué)院學(xué)報(bào)，2010，27(1): 126-133.

LIU Xiaoshuang, HUANG Jianwen, JU Hongbo. Research progreee in the methods and applications of individual tree crown’s automatic extraction by high spatial resolution remote sensing[J]. Journal of Zhejiang Forestry College,2010，27(1): 126-133.(in Chinese)

2KE Y, QUACKENBUSH L J. A review of methods for automatic individual tree-crown detection and delineation from passive remote sensing[J]. International Journal of Remote Sensing, 2011, 32(17): 4725-4747.

3WULDER M, NIEMANN K O, GOODENOUGH D G. Local maximum filtering for the extraction of tree locations and basal area from high spatial resolution imagery[J]. Remote Sensing of Environment, 2000, 73(1): 103-114.

4POULIOT D A, KING D J, BELL F W, et al. Automated tree crown detection and delineation in high-resolution digital camera imagery of coniferous forest regeneration[J]. Remote Sensing of Environment, 2002, 82(2): 322-334.

5CULVENOR D S. TIDA: an algorithm for the delineation of tree crowns in high spatial resolution remotely sensed imagery[J]. Computers & Geosciences, 2002, 28(1): 33-44.

6WANG L, GONG P, BIGING G S. Individual tree-crown delineation and treetop detection in high-spatial-resolution aerial imagery[J]. Photogrammetric Engineering & Remote Sensing, 2004, 70(3): 351-357.

8BRANDTBERG T, WALTER F. Automated delineation of individual tree crowns in high spatial resolution aerial images by multiple-scale analysis[J]. Machine Vision and Applications, 1998, 11(2): 64-73.

9BRANDTBERG T, WARNER T A, LANDENBERGER R E, et al. Detection and analysis of individual leaf-off tree crowns in small footprint, high sampling density lidar data from the eastern deciduous forest in North America[J]. Remote Sensing of Environment, 2003, 85(3): 290-303.

10HAY G J, CASTILLA G, WULDER M A, et al. An automated object-based approach for the multiscale image segmentation of forest scenes[J]. International Journal of Applied Earth Observation and Geoinformation, 2005, 7(4): 339-359.

11POULIOT D A, KING D J, PITT D G. Development and evaluation of an automated tree detection delineation algorithm for monitoring regenerating coniferous forests[J]. Canadian Journal of Forest Research, 2005, 35(10): 2332-2345.

12POLLOCK R J. The automatic recognition of individual trees in aerial images of forests based on a synthetic tree crown image model[D]. Montreal: Concordia University, 1996.

13LARSEN M, RUDEMO M. Optimizing templates for finding trees in aerial photographs[J]. Pattern Recognition Letters, 1998, 19(12): 1153-1162.

14TARP-JOHANSEN M J. Automatic stem mapping in three dimensions by template matching from aerial photographs[J]. Scandinavian Journal of Forest Research, 2002, 17(4): 359-368.

15GOUGEON F A. A crown-following approach to the automatic delineation of individual tree crowns in high spatial resolution aerial images[J]. Canadian Journal of Remote Sensing, 1995, 21(3): 274-284.

16LECKIE D G, GOUGEON F A, WALSWORTH N, et al. Stand delineation and composition estimation using semi-automated individual tree crown analysis[J]. Remote Sensing of Environment, 2003, 85(3): 355-369.

17GOUGEON F A, LECKIE D G. The individual tree crown approach applied to Ikonos images of a coniferous plantation area[J]. Photogrammetric Engineering & Remote Sensing, 2006, 72(11): 1287-1297.

18POULIOT D A, KING D J, BELL F W, et al. Automated tree crown detection and delineation in high-resolution digital camera imagery of coniferous forest regeneration[J]. Remote Sensing of Environment, 2002, 82(2): 322-334.

19ERIKSON M. Segmentation of individual tree crowns in colour aerial photographs using region growing supported by fuzzy rules[J]. Canadian Journal of Forest Research, 2003, 33(8): 1557-1563.

20BUNTING P, LUCAS R. The delineation of tree crowns in Australian mixed species forests using hyperspectral compact airborne spectrographic imager (CASI) data[J]. Remote Sensing of Environment, 2006, 101(2): 230-248.

21HIRSCHMUGL M, OFNER M, RAGGAM J, et al. Single tree detection in very high resolution remote sensing data[J]. Remote Sensing of Environment, 2007, 110(4): 533-544.

22WANG L, GONG P, BIGING G S. Individual tree-crown delineation and treetop detection in high-spatial-resolution aerial imagery[J]. Photogrammetric Engineering & Remote Sensing, 2004, 70(3): 351-357.

23LAMAR W R, MCGRAW J B, WARNER T A. Multitemporal censusing of a population of eastern hemlock (TsugacanadensisL.) from remotely sensed imagery using an automated segmentation and reconciliation procedure[J]. Remote Sensing of Environment, 2005, 94(1): 133-143.

24JING L, HU B, NOLAND T, et al. An individual tree crown delineation method based on multi-scale segmentation of imagery[J]. ISPRS Journal of Photogrammetry and Remote Sensing, 2012, 70: 88-98.

25熊軼群, 吳健平. 基于高分辨率遙感圖像的樹(shù)冠面積提取方法[J]. 地理與地理信息科學(xué), 2007, 23(6): 30-33.

XIONG Yiqun, WU Jianping. Tree-crown area detection algorithm for high spatial resolution remote-sensing image[J]. Geography and Geo-Information Science, 2007, 23(6): 30-33. (in Chinese)

26LIU J, SHEN J, ZHAO R, et al. Extraction of individual tree crowns from airborne LiDAR data in human settlements[J]. Mathematical and Computer Modelling, 2013, 58(3): 524-535.

27SCHMITT M, SHAHZAD M, ZHU X X. Reconstruction of individual trees from multi-aspect TomoSAR data[J]. Remote Sensing of Environment, 2015, 165: 175-185.

28張凝, 張曉麗, 葉栗. 基于改進(jìn)爬峰法高分辨率遙感影像分割的樹(shù)冠提取[J]. 農(nóng)業(yè)機(jī)械學(xué)報(bào), 2014, 45(12): 294-300.

ZHANG Ning, ZHANG Xiaoli, YE Li. Tree crown extraction based on segmentation of high-resolution remote sensing image improved peak-climbing algorithm[J]. Transactions of the Chinese Society for Agricultural Machinery， 2014，45(12): 294-300. (in Chinese)

29JENSON S K, DOMINGUE J O. Extracting topographic structure from digital elevation data for geographic information system analysis[J]. Photogrammetric Engineering and Remote Sensing, 1988, 54(11): 1593-1600.

A Method for Lichee’s Tree-crown Information Extraction Based on High Spatial Resolution Image

Jiang Renrong1,2Wang Chunyan1Shen Liqiang1Wang Peifa3

(1.ShenzhenUrbanPlanningandLandResourceResearchCenter,Shenzhen518040,China2.KeyLaboratoryofUrbanLandResourcesMonitoringandSimulation,MinistryofLandandResource,Shenzhen518040,China3.SchoolofGeographyandRemoteSensing,NanjingUniversityofInformationScienceandTechnology,Nanjing210044,China)

As high spatial resolution remotely sensed image be acquired more easily, there is a great potential for obtaining forest inventory automatically and cost-efficiently. A method was proposed to detect the lichee’s treetop and delineate tree-crown. The method can be divided into three steps. In the first step, a 3×3 mean filter was utilized to smooth image, and then the image was inverted through subtracting image from the maximum of the filtered image. The second step was individual tree detection, namely treetop detection. The inverted image can be viewed as a topographic surface, the flow direction grid was built and then the depressions grid was extracted. The depressions distributed on roads and constructions were deleted according to the predefined threshold. Watersheds were delineated to obtain the contributing area of depressions viewing depressions as the pour point. For solving that the multiple depressions were erroneously identified within the same crown, the depressions were deleted if the distance to the nearest depression was less than threshold and the mean value of depression in the filtered image was not the maximum in multiple depressions, the watersheds of multiple depressions were merged. The remaining depressions were viewed as the detected treetop. The third step was to delineate the tree-crown by using region growing method. The remaining depressions were used for seed points, crown regions were expanded from depression to surrounding pixels until the difference between the pixel and mean value of depression exceeded the predefined threshold or to the boundary of depression watershed. A 324 pixel×483 pixel Pléiades image with 0.5 m resolution was employed to test the method. A promising agreement between the detected results and manual delineation results was achieved in counting the number of trees and the area of delineating tree crowns. For individual tree detection, the overall accuracy was 87.75%, user’s accuracy was 80.69%, producer’s accuracy was 96.06%; for individual tree-crow delineation, the overall accuracy was 78.69%, user’s accuracy was 71.32%, producer’s accuracy was 87.76%.

lichee; high spatial resolution remotely sensed image; individual tree detection; individual tree-crown delineation; hydrological analysis; region growing

10.6041/j.issn.1000-1298.2016.09.003

2015-11-16

2016-03-29

國(guó)土資源部公益性行業(yè)專項(xiàng)(201411014-4)、深圳市基本生態(tài)控制線專項(xiàng)調(diào)查和深圳市2012年測(cè)繪地籍工程計(jì)劃項(xiàng)目

姜仁榮(1981—)，男，高級(jí)工程師，博士，主要從事地圖學(xué)與地理信息系統(tǒng)研究，E-mail： jiangrenrong@126.com

王培法(1980—)，男，講師，博士，主要從事高分辨率遙感圖像信息提取研究，E-mail： wangpeifa1980@163.com

TP751.1; S758

A

1000-1298(2016)09-0017-06