农业工程学报
農業工程學報
농업공정학보
2014年
12期
148-153
,共6页
王传宇%郭新宇%肖伯祥%杜建军%吴升
王傳宇%郭新宇%肖伯祥%杜建軍%吳升
왕전우%곽신우%초백상%두건군%오승
农作物%图像处理%测量%图像序列%图像拼接%苗期玉米%缺苗数量
農作物%圖像處理%測量%圖像序列%圖像拼接%苗期玉米%缺苗數量
농작물%도상처리%측량%도상서렬%도상병접%묘기옥미%결묘수량
crops%image processing%measurements%image sequences%image mosaic%corn seedling%number of seedlings deficiency
为自动测量苗期玉米植株缺失数量,研究一种基于图像拼接的玉米早期缺苗数量自动测量方法。该方法首先在田间光照条件下,从植株顶部沿行向获取玉米图像序列,并将图像序列注册到同一坐标系下拼接为行向图像,然后将植株像素从土壤背景中分割出来,在植株细化骨架上标识茎秆中心点。最后以行向图像上各茎秆中心点拟合行向直线,将茎秆中心点向行向直线投影,从相邻投影点的距离计算植株平均株距,缺苗数量可由平均株距和两相邻植株的距离计算。在3个不同密度的试验小区上对比该方法与人工测量,每个小区进行10次重复,在低密度和中密度小区两种方法具有较高的相关性,在高密度小区两种方法的相关性有所下降。该方法可以替代人工测量,从而减少时间和人力投入,提高玉米早期植株缺苗数调查的自动化程度。
為自動測量苗期玉米植株缺失數量,研究一種基于圖像拼接的玉米早期缺苗數量自動測量方法。該方法首先在田間光照條件下,從植株頂部沿行嚮穫取玉米圖像序列,併將圖像序列註冊到同一坐標繫下拼接為行嚮圖像,然後將植株像素從土壤揹景中分割齣來,在植株細化骨架上標識莖稈中心點。最後以行嚮圖像上各莖稈中心點擬閤行嚮直線,將莖稈中心點嚮行嚮直線投影,從相鄰投影點的距離計算植株平均株距,缺苗數量可由平均株距和兩相鄰植株的距離計算。在3箇不同密度的試驗小區上對比該方法與人工測量,每箇小區進行10次重複,在低密度和中密度小區兩種方法具有較高的相關性,在高密度小區兩種方法的相關性有所下降。該方法可以替代人工測量,從而減少時間和人力投入,提高玉米早期植株缺苗數調查的自動化程度。
위자동측량묘기옥미식주결실수량,연구일충기우도상병접적옥미조기결묘수량자동측량방법。해방법수선재전간광조조건하,종식주정부연행향획취옥미도상서렬,병장도상서렬주책도동일좌표계하병접위행향도상,연후장식주상소종토양배경중분할출래,재식주세화골가상표식경간중심점。최후이행향도상상각경간중심점의합행향직선,장경간중심점향행향직선투영,종상린투영점적거리계산식주평균주거,결묘수량가유평균주거화량상린식주적거리계산。재3개불동밀도적시험소구상대비해방법여인공측량,매개소구진행10차중복,재저밀도화중밀도소구량충방법구유교고적상관성,재고밀도소구량충방법적상관성유소하강。해방법가이체대인공측량,종이감소시간화인력투입,제고옥미조기식주결묘수조사적자동화정도。
The missing amount of planted corn seedlings plays an important role in corn yield, to acquire it automatically, a new system based on machine vision has been developed. System hardware includes: one Industrial Personal Computer, a Central Processing nit:Intel (r) CPU i5@3.4GHz, 4 Gb of memory, one mvc3000 high speed Industrial camera (24FPS), and one Pentax len 8.5 mm f/1.5. The software development environment includes:Win7 Operating System, Microsoft Visual Studio 2010 Professional, and Open CV1.0. The core of the system is the image processing method. Firstly, image sequences obtained along plant rows from a top view under in-field lighting conditions were registered to the uniform coordinate system. Secondly, plant pixel (vegetation) was segmented from the background with a pixel classifier trained by a neural network. The segmentation method employed a decision surface in color space that was defined by only three parameters. This surface was a at runcated ellipsoidal surface which was robust in outdoor field images under varying lighting conditions. A simple parallel algorithm working on 8-connectivity was implemented, whereby skeletonization extracts a network of thin curves that describe the overall shape or“skeleton”of objects in a binary image. Due to limitations in camera resolution and non-ideal lighting conditions, the minimum gray level point along the plant skeleton is the best estimation of the actual stem location. The minimum gray pixel area was searched along the plant skeleton, and the center of minimum gray pixel area was marked as the stem center. Finally, a plant row line was fitted by stem centers;a model that predicts a linear relationship between the stem centers and the corn plant row was defined, and the parameters of linear function was estimated by a least-squares fit. Stem centers were projected onto the row line, and the average plant spacing was calculated by a projected point. The number of missing plants between two neighbored seedlings has a linear relationship of plant average spacing. On three varieties of 10 repeats each, a 10m long row field experiment was performed, In a low density experiment, measurement results of the method agree with manual measurements of 7 in 10 and 3 in 10 have a difference of one plant. In a high density experiment, measurement results of the method agree with a manual measurement 6 in 10 and 4 in 10 have a difference up to two plants. Comparison with a manual measurement and our method, a high correlation on the two methods was found; this method can replace manual measurement, reduce time cost and human labor effort, and improve the degree of automation of the corn seedling missing survey.