王海超,宗哲英,張文霞,殷曉飛,王曉蓉,張海軍,劉艷秋,石 鑫,王春光
采用均值聚類和環(huán)形結構的狹葉錦雞兒木質部提取算法
王海超,宗哲英,張文霞,殷曉飛,王曉蓉,張海軍,劉艷秋,石 鑫,王春光※
(內(nèi)蒙古農(nóng)業(yè)大學能源與交通工程學院,呼和浩特 010018)
針對木質部交互統(tǒng)計誤差大、效率低、重現(xiàn)性差、勞動強度高和傳統(tǒng)圖像處理算法精度不理想等問題,該文以狹葉錦雞兒木質部切片圖像為研究對象,根據(jù)木質部特點提出基于均值聚類算法和環(huán)形結構提取算法相結合,實現(xiàn)木質部準確提取的方法。首先通過動態(tài)巴特沃斯同態(tài)濾波法對30幅供試圖像進行光照不均校正,然后采用均值聚類法對光照補償后圖像初分割,最后采用環(huán)形結構提取算法實現(xiàn)木質部提取計數(shù)。試驗結果表明:采用均值聚類算法對光照補償后的木質部圖像初分割分割誤差(section error,)、過分割誤差OR(over-segmentation error, OR)和欠分割誤差UR(under-segmentation error, UR)均值分別為5.15%、1.48%和6.46%,優(yōu)于未光照補償和3R-G-B算法;該文提出的環(huán)形結構提取算法對初分割后木質部圖像檢測的平均相對誤差為2.26%,比分水嶺法低11.69個百分點,比凹點匹配法低4.93個百分點。從速度上看,該算法平均耗時3.17 s,比分水嶺法快1.40 s,比凹點匹配法快4.88 s。該算法檢測的均方根誤差RMSE(root mean squared error, RMSE)為0.52%,約相當于分水嶺法的1/3,約相當于凹點匹配法的1/2,該算法優(yōu)于其他2種分割算法;在圖像結構復雜、光照不均勻、內(nèi)部分布不均等缺陷條件下,該文算法也能很好地實現(xiàn)木質部的分割和提取。該方法不僅能對狹葉錦雞兒木質部自動分割和提取,也可為其他植物木質部分割提取提供參考。
提??;算法;木質部;均值聚類;環(huán)形結構提取;狹葉錦雞兒
木質部是維管植物體內(nèi)重要的復合組織,負責水分及水分中離子運輸和支撐作用[1-2],其深入研究對揭示維管植物抗旱機制和不同條件下耐旱植物的選育具有重要意義。目前木質部統(tǒng)計常通過離析、切片等手段制成樣本,采用顯微鏡人工交互方式進行計數(shù),該方式存在人為誤差大、效率低、重現(xiàn)性差和勞動強度大等缺點,制約了該領域的深入研究[3-5]。
木質部多存在黏連情況,黏連細胞分割和統(tǒng)計是圖像處理領域一項基本而又十分關鍵的技術,一直是細胞統(tǒng)計學中研究難點和熱點問題。常用的黏連細胞分割方法有分水嶺算法、凹點匹配法、形態(tài)學法、橢圓建模法、水平集法和機器學習法[6-11],其中分水嶺算法、凹點匹配法因其實現(xiàn)簡單、高效,得到的應用最多,目前以這2種算法為框架,并出現(xiàn)了各種改進算法。Salim等[12]提出基于距離地形圖分水嶺變換分離黏連細胞,提高了正常白細胞和致密白血細胞病簇的分割精度;Miao等[13]提出一種標記控制分水嶺算法自動分割和統(tǒng)計血液中白細胞和紅細胞數(shù)量,該算法基于距離變換和邊緣梯度信息來獲取血細胞輪廓,通過分類獲得分段的白細胞和紅細胞,此法相較傳統(tǒng)分水領算法精度較高,但對先驗標記精度要求較高;Hasan等[14]提出2步驗證分水嶺匹配算法對腦腫瘤進行分割,其使用偽影去除、中值濾波和三邊濾波對圖像進行預處理,首先從MR圖像中分割出腫瘤區(qū)域,然后使將分割后的部分與驗證圖像進行匹配,從而準確分割腦腫瘤;Albayrak等[15]采用兩級超像素分割算法對腎癌細胞進行提取,該算法首先利用簡單線性迭代聚類法(simple linear iterative clustering,SLIC)將圖像分割為超像素圖像,然后采用基于密度的聚類算法(density-based spatial clustering of applications with noise,DBSCAN)對獲得的超像素進行聚類,找到組成細胞核的相似超像素,從而實現(xiàn)腎癌細胞準確分割;閆沫[16]結合梯度修正和區(qū)域歸并策略對傳統(tǒng)分水嶺算法進行改進,改善了分水嶺算法過分割現(xiàn)象;趙紅英等[17]采用基于水平集主動輪廓(active contour model,ACM)算法對宮頸癌細胞初分割,然后將歸一化后圖像與感興趣區(qū)域(region of interest,ROI)梯度圖像點乘來抑制無用梯度信息,最后運用標記分水嶺算法對感興趣區(qū)域細胞進行分割;廖慧司等[18]提出一種結合距離變換利用邊緣梯度的分水嶺血細胞顯微圖像分割算法,該算法由距離圖提取前景標記,將距離分水嶺變換所得的脊線作為梯度分水嶺變換的背景標記,能有效地分離黏連目標,但該方法魯棒性較差,對切片質量要求較高;張建華等[19]在H-minima分水嶺分割基礎上,結合最小二乘圓法誤差理論,提出了自適應H-minima分水嶺分割方法,實現(xiàn)了棉花葉部黏連病斑的準確分割,但當病斑黏連較緊密和大小病斑重疊在一起時會存在欠分割情況。Yao等[20]采用邊緣中心模態(tài)比例(edge center mode proportion,ECMP)法對水稻粒進行凹點匹配,在協(xié)同約束條件下進行分割,然后再用最小外接矩形計算其長度,從而精確識別出稻米粒,但該算法容易出現(xiàn)過分割情況;Zhang等[21]采用canny邊緣檢測和改進的凹點匹配算法對接觸種子進行分離,有效地提取了種子的位置和方向信息,有效地實現(xiàn)了種子自動挑選,Zhang等[22]利用凹點檢測和線性分組技術對重疊細胞進行自動分割,該算法主要包括輪廓提取、凹點檢測、輪廓段分組好橢圓擬合四個步驟,但模糊圖像凹點和邊界的準確定位仍是難點;楊輝華等[23]提出一種結合水平集輪廓提取的凹點區(qū)域檢測的黏連細胞分割方法,準確地分割了黏連細胞,但對于黏連嚴重情況分割精度不高,常出現(xiàn)過分割;王曉鵬等[24]提出一種基于形態(tài)學多尺度重建結合凹點匹配的枸杞圖像分割方法,結合枸杞顆粒的大小和形狀特點,實現(xiàn)黏連枸杞顆粒的分割和計數(shù);李毅念等[25]通過顏色空間轉換、去除細窄黏連、黏連判斷、凹點檢測等算法過程,實現(xiàn)了圖像中黏連麥穗的有效分割,依據(jù)麥穗和麥粒間關系,構建了產(chǎn)量預測模型,進一步得到了單位面積內(nèi)的小麥麥穗數(shù)量、總籽粒數(shù)及產(chǎn)量信息,但對于黏連麥穗存在部分過分割。
木質部存在多個不黏連和黏連形式,其顯微圖像具有紋理多、結構復雜、形狀不規(guī)則等特征,常存在低對比度、邊界模糊、內(nèi)部分布不均等缺陷,限制了細胞分割和統(tǒng)計的準確性,也對算法魯棒性提出了挑戰(zhàn),目前國內(nèi)外對其分割提取的研究鮮有報道。因此,本文以狹葉錦雞兒木質部圖像為研究對象,在分析總結前人算法和木質部圖像特點基礎上,首先對采集的木質部圖像采用動態(tài)巴特沃斯濾波器進行濾波,消除顯微圖像光照不均現(xiàn)象;然后采用均值聚類算法將木質部從原圖像中分離出來;最后采用本文提出的環(huán)形結構提取算法實現(xiàn)木質部提取和計數(shù)。
a. 第一組木質部圖像a. First set of xylem imageb. 第二組木質部圖像b. Second set of xylem imagec. 第三組木質部圖像c. Third set of xylem image
1.2.1 光照不均校正
顯微圖像常存在光照不均和光照不足現(xiàn)象,這會對后續(xù)圖像分割和特征提取準性造成較大影響,改善圖像分辨率和視覺效果是圖像處理中不可缺少的環(huán)節(jié)。本文采用HSV變換和動態(tài)巴特沃斯同態(tài)濾波算法對木質部圖像進行光線補償,該算法在不改變原圖色調和飽和度不變的前提下對亮度分量進行增強,圖像細節(jié)增強同時削弱低頻分量,改善圖像質量[27-29]。
1.2.2均值聚類
均值聚類算法是典型的無監(jiān)督硬聚類算法,其以歐式距離、漢明距離、閔可夫斯基距離和街區(qū)距離等作為相似度度量(默認采用歐式距離),以誤差平方和作為聚類準則,可實現(xiàn)類間相似度最低和類內(nèi)相似度最高,且局部最優(yōu),十分適合彩色圖像分割[30]。采用均值聚類算法對彩色圖像進行分割時,往往選用Lab顏色空間,Lab模型可近似使用球體結構表示,顏色空間是均勻的,過球心的笛卡爾三坐標對應各顏色分量,各任意色彩均可由以上亮度()、色度(,+表示紅色,-表示綠色)和色度(,+表示黃色,-表示藍色)3個分量疊加而成[31]。由于木質部圖像主要由紅色的木質部、白色液體膜和綠色韌皮部部分構成,故聚類中心數(shù)目為3,聚類后紅色區(qū)域為分割的目標區(qū)域。
1.2.3 初分割質量評價
為定量評價算法分割效果,本文在總結分析已有圖像分割評價法基礎上,選用分割誤差R(section error,R)、過分割誤差OR(over-segmentation error,OR)和欠分割誤差UR(under-segmentation error,UR)對分割結果進行評價。這3種評價指標值越低,表明圖像分割效果越好,目標提取精度越高。這3種評價指標均需要分割目標真實面積作為基準,目標真實面積采用Photoshop進行手動分割,擦除背景區(qū)域后剩余像素數(shù)作為目標真實尺寸。3種評價指標計算公式為
由木質部結構特點可知,其經(jīng)初分割后存在獨立、黏連和不閉合現(xiàn)象,且木質部呈環(huán)狀。從分割目標考慮,對木質部個數(shù)統(tǒng)計時,不必保證木質部結構完整性,只需保證個數(shù)準確即可。故本文在充分分析木質部細胞結構特點和前人算法基礎上,提出一種環(huán)形結構提取算法,從而實現(xiàn)木質部準確提取計數(shù)。該算法首先確定木質部圖像連通域,并對聯(lián)通域進行標記,剔除較小的雜質區(qū)域;然后采用定步長窗口掃描方式粗略估計出環(huán)形結構中心位置;最后通過圓心位置對其對應上、下、左、右4個進行檢測,若檢測出不少于2個方向上存在環(huán)形部分結構,則該圓心對應的環(huán)形結構即為1個木質部,具體過程如下:
1)連通域標記
2)環(huán)形結構圓心位置估計
注:圖中各個變量為點的坐標,1m為組數(shù),為圖像寬度。
Note: Variables in the graph are the coordinates of points,1mis the number of group,is the width of image.
圖2 環(huán)形結構圓心位置示意圖
Fig.2 Schematic diagram of circular structure’s center position
3)環(huán)形結構判定
①采用Sobel算子提取木質部邊緣,對木質部上、下、左、右4個方向進行檢測,若距圓心(,)>min和 ②當上、下、左、右4個方向上無法檢測出2個及以上環(huán)形結構時,可能木質部存在缺口,此時需將原檢測方向左右偏移45°,若有2個以上方向上存在環(huán)形結構,則認為存在1個木質部,如圖3b所示。 注:Rmax為最大外徑,rmin為最小內(nèi)徑,(a,b)為圓心,r為實際檢測半徑。 4)環(huán)形結構提取 以(,)為中心,用矩形框將環(huán)形結構標出并計數(shù),實現(xiàn)環(huán)形結構提取。 上述木質部提取流程如圖4所示。 圖4 算法流程 為驗證算法精度、穩(wěn)定性和速度等有效性,從已拍攝的木質部圖像中隨機選取木質部黏連程度各異的圖像30幅進行木質部分割提取。試驗采用Window7旗艦版64位系統(tǒng)、主頻2.40 GHz、8 G內(nèi)存Asus筆記本電腦,軟件采用MatlabR2014a,具體試驗分為4部分: 1)為驗證光照補償?shù)挠行?,從已拍攝的木質部圖像中選取木質部黏連程度各異的圖像30幅進行試驗,采用均值聚類算法分別對原始圖像和同態(tài)濾波后圖像ab分量進行聚類,分別采用分割誤差、過分割誤差OR和欠分割誤差UR對算法進行定量評價; 2)為驗證分割算法有效性,采用均值聚類算法和3R-G-B閾值分割算法[32]對同態(tài)濾波后木質部圖像進行分割,并對分割效果進行比較; 3)為檢驗本文環(huán)形提取算法性能,對初分割后的30幅木質部圖像進行提取,試驗軟件和硬件與木質部初分割使用相同。分別采用分水嶺法[33]、凹點匹配法[34]和本文算法對木質部進行提取,最后將各算法提取結果與實際木質部數(shù)量進行對比,從而對各算法性能進行評價。 1)采用聚類中心數(shù)目為3的均值聚類算法對30幅供試圖像處理,結果如圖5所示,聚類后紅色區(qū)域為目標區(qū)域,分割效果如表1所示。其中,圖5a為動態(tài)巴特沃斯同態(tài)濾波后圖像,可以看出,濾波后木質部圖像細節(jié)、紋理、對比度和視覺效果得到明顯改善,光照均勻度增強;圖5b是未進行光照補償直接采用均值聚類算法分割后效果,由于受光照不足和不均影響,存在較嚴重的過分割現(xiàn)象;圖5c為同態(tài)濾波光照補償后均值聚類算法分割后效果,可以發(fā)現(xiàn)分割效果得到明顯改善,木質部分割的更為完整。由表1知,采用均值聚類算法對未進行光照補償處理的木質部圖像分割誤差、過分割誤差OR和欠分割誤差UR均值分別為28.75%、9.23%和19.47%,同態(tài)濾波光照補償后,均值聚類算法分割誤差、過分割誤差OR和欠分割誤差UR均值分別為5.15%、1.48%和6.46%,較未進行光照補償分別降低了23.60、7.75和13.01個百分點。由此可以發(fā)現(xiàn),采用動態(tài)巴特沃斯同態(tài)濾波算法對木質部圖像光照補償后,不但能改善圖像質量和分割效果,而且還能夠提高分割算法分割精度; 圖5 光照補償前后不同分割算法分割結果示例 2)采用3R-G-B閾值分割算法對光照補償木質部細胞圖像分割結果如圖5d所示,分割效果客觀評價如表1所示??梢园l(fā)現(xiàn),雖然部分分割效果優(yōu)于均值聚類算法,但大部分分割存在較大誤分割,整體分割效果不如均值聚類算法。由表1知,3R-G-B閾值分割算法對光照補償后木質部細胞分割誤差、過分割誤差OR和欠分割誤差UR均值分別為15.58、6.06和11.42個百分點,較光照補償后均值聚類算法分別增加了10.43、4.58和4.96個百分點。由上述結果可以發(fā)現(xiàn),針對木質部細胞圖像,均值聚類算法分割效果優(yōu)于3R-G-B閾值分割算法,分割誤差、過分割誤差OR和欠分割誤差UR更低。 表1 本文算法與3R-G-B算法對測試圖像分割效果 注:、OR、UR分別為分割誤差、過分割誤差和欠分割誤差。 Note:stands for segmentation error; OR stands for over-segmentation error; UR stands for under-segmentation error. 圖6 本文算法與其他算法對測試圖像木質部分割結果 表2 不同算法對測試圖像木質部檢測結果對比 由圖6a和6b可以看出,當木質部黏連較簡單時,分水嶺法和凹點匹配法分割較準確,但當木質部黏連程度復雜時,分割效果較差,出現(xiàn)了較多誤分割。由圖6c可知,相較上述2種算法,本文提出的環(huán)形結構提取算法分割較準確。由表2可以看出,本文算法檢測木質部數(shù)目平均相對誤差為2.26%,比分水嶺法低11.69個百分點,比凹點匹配法低4.93個百分點;從速度上看,本文算法平均耗時3.17 s,比分水嶺法快1.40 s,比凹點匹配法快4.88 s,但本文算法與凹點匹配法耗時均隨木質部數(shù)目增多、黏連復雜度增高呈上升趨勢,分水嶺法耗時相對穩(wěn)定;本文算法檢測的均方根誤差RMSE(root mean squared error,RMSE)為0.52%,約相當于分水嶺法的1/3,約相當于凹點匹配法的1/2。綜合衡量,本文算法較好。 本文以狹葉錦雞兒木質部圖像為研究對象,針對黏連細胞分割問題,通過光照不均校正、K均值聚類初分割和木質部環(huán)形提取等算法,實現(xiàn)了圖像中木質部的有效分割和提取。通過試驗得出以下結論: 1)采用均值聚類算法對光照補償后的木質部圖像初分割誤差(section error,)、過分割誤差OR(over-segmentation error,OR)和欠分割誤差UR(under-segmentation error,UR)均值分別為5.15%、1.48%和6.46%,優(yōu)于3R-G-B閾值分割算法; 2)本文提出的環(huán)形結構提取算法能夠實現(xiàn)木質部準確提取計數(shù),對初分割后木質部圖像檢測的平均相對誤差為2.26%,比分水嶺法低11.69個百分點,比凹點匹配法低4.93個百分點。從速度上看,本文算法平均耗時3.17 s,比分水嶺法快1.40 s,比凹點匹配法快4.88 s,但本文算法與凹點匹配法耗時均隨著木質部數(shù)目增多、黏連復雜度增高呈上升趨勢,分水嶺法耗時相對穩(wěn)定。本文算法檢測的均方根誤差RMSE(root mean squared error,RMSE)為0.52,約相當于分水嶺法的1/3,約相當于凹點匹配法的1/2。綜合衡量,本文算法優(yōu)于上述2種算法。 當木質部黏連特別緊密和缺失嚴重時,本文方法存在部分欠分割現(xiàn)象,在今后進一步研究中,將結合深度學習中的語義分割和實例分割,提高黏連木質部分割精度,改善本文算法不足。 [1]董星光,曹玉芬,王昆,等. 中國3個主要梨砧木資源木質部導管分子結構及分布比較[J]. 植物學報,2015,50(2):227-233. Dong Xingguang, Cao Yufen, Wang Kun, et al. Comparison of the characters and distribution of vessel elements in xylem among three main pear rootstocks in China[J]. Bulletin of Botany, 2015, 50(2): 227-233. (in Chinese with English abstract) [2]Tombesi S, Johnson R S, Day K R, et al. Relationships between xylem vessel characteristics, calculated axial hydraulic conductance and size-controlling capacity of peach rootstocks[J]. Annals of Botany, 2010, 105(2): 327-331. [3]王明浩,張曉瑋,王婧如,等. 一種簡易準確測定木質部導水率的新方法[J]. 植物生理學報,2013,49(3):297-300. Wang Minghao, Zhang Xiaowei, Wang Jingru, et al. A simple and accurate method for measuring hydraulic conductivity in xylem[J]. Plant Physiology Journal, 2013, 49(3): 297-300. (in Chinese with English abstract) [4]Morris H, Plavcová L, Gorai M, et al. Vessel-associated cells in angiosperm xylem: Highly specialized living cells at the symplast-apoplast boundary[J]. American Journal of Botany, 2018, 105(2): 151-160. [5]Savi T, Petruzzellis F, Martellos S, et al. Vineyard water relations in a karstic area: deep roots and irrigation management[J]. Agriculture Ecosystems & Environment, 2018, 263(5): 53-59. [6]李葉舟,胡靜,朱少彰,等. 基于光照方向不一致的圖像盲取證技術[J]. 北京郵電大學學報,2011,34(3):26-30. Li Yezhou, Hu Jing, Zhu Shaozhang, et al. Exposing digital image forgeries by detecting inconsistence in light source direction[J]. Journal of Beijing University of Posts and Telecommunications, 2011, 34(3): 26-30. (in Chinese with English abstract) [7]Felix B, Carsten M, Michael S, et al. An automatic method for robust and fast cell detection in bright field images from high-throughput microscopy[J]. BMC Bioinformatic, 2013, 14(1): 297-310. [8]Chitade A Z, Katiyar S K. Colour based image segmentation using-means clustering[J]. International Journal of Engineering Science and Technology, 2010, 2(10): 5319-5325. [9]Dzyubachyk O, Van Cappellen W, Essers J, et al. Advanced level set-based cell tracking in time-lapse fluorescence microscopy[J]. IEEE Transactions on Medical Imaging, 2010, 29(3): 852-867. [10]Duartea M, Alvarenga A, Azevedo C, et al. Evaluating geodesic active contours in microcalcifications segmentation on mammograms[J]. Computer Methods Programs Biomedical, 2015, 122(3): 304-315. [11]宋懷波,張傳棟,潘景朋,等. 基于凸殼的重疊蘋果目標分割與重建算法[J]. 農(nóng)業(yè)工程學報,2013,29(3):163-168. Song Huaibo, Zhang Chuandong, Pan Jingpeng, et al. Segmentation and reconstruction of overlapped apple images based on convex hull[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2013, 29(3): 163-168. (in Chinese with English abstract) [12]Salim A, Emel O, Cigdem G D. A color and shape based algorithm for segmentation of white blood cells in peripheral blood and bone marrow images[J]. International Society for Advancement of Cytometry, 2014, 85(1): 480-490. [13]Miao H, Xiao C. Simultaneous segmentation of leukocyte and erythrocyte in microscopic images using a marker-controlled watershed algorithm [J]. Computational and Mathematical Methods in Medicine, 2018, 2018(9): 1-9. [14]Hasan S M K, Ahmad M. Two-step verification of brain tumor segmentation using watershed-matching algorithm[J]. Brain Informatics, 2018, 5(2): 1-11. [15]Albayrak A, Bilgin G. Automatic cell segmentation in histopathological images via two-staged superpixel-based algorithms[J]. Medical and Biological Engineering and Computing, 2019, 57(3): 653-665. [16]閆沫. 基于組件樹濾波及快速區(qū)域合并的分水嶺分割算法[J]. 計算機科學,2013,40(1):282-285. Yan Mo. Watershed algorithm based on image filtering by using component tree and fast region merging[J]. Computer Science, 2013, 40(1): 282-285. (in Chinese with English abstract) [17]趙紅英,周中順,孫存杰,等. 一種新的用于宮頸癌粘連細胞圖像分割的分水嶺算法[J]. 軍事醫(yī)學,2014,32(12):972-975. Zhao Hongying, Zhou Zhongshun, Sun Cunjie, et al. A new modified watershed algorithm for image segmentation in cervical cancer cell adhesion[J]. Military Medical Sciences, 2014, 32(12): 972-975. (in Chinese with English abstract) [18]繆慧司,梁光明,劉任任,等. 結合距離變換與邊緣梯度的分水嶺血細胞分割[J]. 中國圖象圖形學報,2016,21(2):192-198. Liao Huisi, Liang Guangming, Liu Renren, et al. Watershed algorithm using edge gradient combined with distance transformation for segmentation of blood cells[J]. Journal of Image and Graphics, 2016, 21(2): 192-198. (in Chinese with English abstract) [19]張建華,韓書慶,翟治芬,等. 改進自適應分水嶺方法分割棉花葉部粘連病斑[J]. 農(nóng)業(yè)工程學報,2018,34(24):165-174. Zhang Jianhua, Han Shuqing, Zhai Zhifen, et al. Improved adaptive watershed method for segmentation of cotton leaf adhesion lesions[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2018, 34(24): 165-174. (in Chinese with English abstract) [20]Yao Y, Wu W, Yang T, et al. Head rice rate measurement based on concave point matching[J]. Scientific Reports, 2017, 7: 41353. [21]Zhang G, Sun Q, Jiang S, et al. Automatic seed picking for brachytherapy postimplant validation with 3D CT images[J]. International Journal of Computer Assisted Radiology and Surgery, 2017, 12(4): 1-9. [22]Zhang W, Li H. Automated segmentation of overlapped nuclei using concave point detection and segment grouping[J]. Pattern Recognition, 2017, 71: 349-360. [23]楊輝華,趙玲玲,潘細朋,等. 基于水平集和凹點區(qū)域檢測的粘連細胞分割方法[J]. 北京郵電大學學報,2016,39(6):11-16. Yang Huihua, Zhao Lingling, Pan Xipeng, et al. Overlapping cell segmentation based on level set and concave area detection[J]. Journal of Beijing University of Posts and Telecommunications, 2016, 39(6): 11-16. (in Chinese with English abstract) [24]王小鵬,姚麗娟,文昊天,等. 形態(tài)學多尺度重建結合凹點匹配分割枸杞圖像[J]. 農(nóng)業(yè)工程學報,2018,34(2):212-218. Wang Xiaopeng, Yao Lijuan, Wen Haotian, et al. Wolfberry image segmentation based on morphological multi-scale reconstruction and concave points matching[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2018, 34(2): 212-218. (in Chinese with English abstract) [25]李毅念,杜世偉,姚敏,等. 基于小麥群體圖像的田間麥穗計數(shù)及產(chǎn)量預測方法[J]. 農(nóng)業(yè)工程學報,2018,34(21):185-194. Li Yinian, Du Shiwei, Yao Min, et al. Method for wheatear counting and yield predicting based on image of wheatear population in field[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2018, 34(21): 185-194. (in Chinese with English abstract) [26]王海超,王春光,宗哲英,等. 基于噪聲類型及強度估計的狹葉錦雞兒葉切片圖像盲去噪[J]. 農(nóng)業(yè)工程學報,2017,33(10):229-238. Wang Haichao, Wang Chunguang, Zong Zheying, et al. Blind image denoising of microscopic slices image of Caragana stenophylla Pojark based on noise type and intensity estimation[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2017, 33(10): 229-238. (in Chinese with English abstract) [27]焦竹青,徐保國. HSV變換和同態(tài)濾波的彩色圖像光照補償[J]. 計算機工程與應用,2010,46(30):142-144. Jiao Zhuqing, Xu Baoguo. Color image illumination compensation based on HSV transform and homomorphic filtering[J]. Computer Engineering and Applications, 2010, 46(30): 142-144. (in Chinese with English abstract) [28]張亞飛,謝明鴻. 基于HSI和局部同態(tài)濾波的彩色圖像增強算法[J]. 計算機應用與軟件,2013,30(12):303-307. Zhang Yafei, Xie Minghong. Colour image enhancement algorithm based on HIS and local homomorphic filtering[J]. Computer Applications and Software, 2013, 30(12): 303-307. (in Chinese with English abstract) [29]焦竹青,徐保國. 基于同態(tài)濾波的彩色圖像光照補償方法[J]. 光電子,激光,2010,21(4):602-605. Jiao Zhuqing, Xu Baoguo. Color image illumination compensation based on homomorphic filtering[J]. Journal of Optoelectronics. Laser, 2010, 21(4): 602-605. (in Chinese with English abstract) [30]徐黎明,呂繼東. 基于同態(tài)濾波和均值聚類算法的楊梅圖像分割[J]. 農(nóng)業(yè)工程學報,2015,31(14):202-208. Xu Liming, Lü Jidong. Bayberry image segmentation based on homomorphic filtering and-means clustering algorithm[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2015, 31(14): 202-20. (in Chinese with English abstract) [31]Azetsu T, Suetake N. Hue-preserving image enhancement in CIELAB color space considering color gamut[J]. Optical Review, 2019, 26(12): 1-12. [32]宋懷波,屈衛(wèi)鋒,王丹丹,等. 基于光照無關圖理論的蘋果表面陰影去除方法[J]. 農(nóng)業(yè)工程學報,2014,30(24):168-176. Song Huaibo, Qu Weifeng, Wang Dandan, et al. Shadow removal method of apples based on illumination invariant image[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2014, 30(24): 168-176. (in Chinese with English abstract) [33]張紅民,王一博. 一種改進的細胞圖像分水嶺分割方法[J]. 重慶理工大學學報,2012,26(11):59-62. Zhang Hongmin, Wang Yibo. An improved method of cell images segmentation based on watershed[J]. Journal of Chongqing University of Technology, 2012, 26(11): 59-62. (in Chinese with English abstract) [34]陶德威. 基于凹點匹配和分水嶺變換的車輛圖像分割算法研究[D]. 南昌:南昌大學,2016. Tao Dewei. Research on the Vehicle Image Segmentation Algorithm Based on Concave Points Matching and Watershed Transformation[D]. Nanchang: Nanchang University, 2016. (in Chinese with English abstract) An extraction xylem images ofPojark based on-means clustering and circle structure extraction algorithm Wang Haichao, Zong Zheying, Zhang Wenxia, Yin Xiaofei, Wang Xiaorong, Zhang Haijun, Liu Yanqiu, Shi Xin, Wang Chunguang※ (,010018,) In the slice images of the xylem ofPojarkthis paper proposed a novel algorithm that combined the-means clustering and circle structure extraction algorithm, to achieve much more accurate information data of the xylem than that from the traditional image processing algorithms. Firstly, the dynamic Butterworth homomorphic filtering can be used to compensate for illumination variations on V components in the 30 imagesofPojark xylem in a HSV color space; then the-means clustering can be used to initially segment theandcomponents of the pre-processed xylem images under the Lab color space with a cluster of 3,finally, the circle structure extraction algorithm can be used to accurately cluster and extract the specific feature of the xylem images. The processing results showed that the Butterworth homomorphic filtering have a good effect on the illumination compensation for the various illumination variations in a series of different images, indicating some high resolution information in detail, texture, contrast and visual effect of the images. After being initially segmented by-means clustering, the illumination compensated xylem images had an average section error () of 5.15%, over-segmentation error (OR) of 1.48% and under-segmentation error (UR) of 6.46%, respectively, which decreased by 23.60, 7.75 and 13.01 percentage points, respectively compared to the xylem images before the illumination compensation. The segmentation accuracy was enhanced significantly, which decreased 10.43 percentage points in, 4.58 percentage points in OR and 4.96 percentage points in UR to 3R-G-B threshold value clustering algorithm after the illumination compensation. The average mean error of the circle structure extraction for the xylem images after the initial segment reached 2.26%, which was 11.69 percentage points lower than that of the watershed method, and 4.93 percentage points lower than that of pit matching method. The average duration of the algorithm in this case was 3.66 s on each image, saving 0.95 and 4.78 s compared to that of the watershed and pit matching method, respectively. The root mean squared error (RMSE) of the algorithm was 0.52%, one third of that from the watershed and half of that from the pit matching. The proposed combined algorithm can automatically segment and extract the xylem information data fromPojark, particularly on some images with the complex xylem structure, uneven illumination and uneven internal distribution, indicating better than the other two types of segmentation algorithms. These findings can provide fundamental reference for the promising extraction algorithm and the image processing of the xylem from other plants. extract; algorithm; xylem;-means clustering; circle structure extraction;Pojark 王海超,宗哲英,張文霞,殷曉飛,王曉蓉,張海軍,劉艷秋,石 鑫,王春光. 采用均值聚類和環(huán)形結構的狹葉錦雞兒木質部提取算法[J]. 農(nóng)業(yè)工程學報,2020,36(1):193-199.doi:10.11975/j.issn.1002-6819.2020.01.022 http://www.tcsae.org Wang Haichao, Zong Zheying, Zhang Wenxia, Yin Xiaofei, Wang Xiaorong, Zhang Haijun, Liu Yanqiu, Shi Xin, Wang Chunguang. An extraction xylem images ofPojark based on-means clustering and circle structure extraction algorithm[J]. Transactions of the Chinese Society of Agricultural Engineering (Transactions of the CSAE), 2020, 36(1): 193-199. (in Chinese with English abstract) doi:10.11975/j.issn.1002-6819.2020.01.022 http://www.tcsae.org 2019-08-22 2019-12-26 內(nèi)蒙古農(nóng)業(yè)大學高層次人才科研啟動項目(NDYB201857);內(nèi)蒙古自治區(qū)自然科學基金項目(2019BS06003,2017MS0514,2017MS0361);教育部“云數(shù)融合科教創(chuàng)新”基金項目(2017A10019);內(nèi)蒙古自治區(qū)博士研究生科研創(chuàng)新項目(B20151012902Z);實驗室開放項目(20180104) 王海超,博士,講師,研究方向:荒漠草原典型植物切片圖像特征與草原早期退化相關性研究。Email:wanghaichao1129@163.com 王春光,教授,博士生導師,研究方向:圖像與數(shù)字化研究。Email:jdwcg@imau.edu.cn 10.11975/j.issn.1002-6819.2020.01.022 TP391.41 A 1002-6819(2020)-01-0193-072 試驗與結果分析
2.1 試驗方法
2.2 結果與分析
3 結 論