基于图像拼接的苗期玉米植株缺失数量自动测量方法

Automatic measurement of numbers of maize seedlings based on mosaic imaging

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。

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.