Machine vision approaches for vegetable seedling growth measurement

Machine vision technique offers a convenient and non-destructive way for measurements of plant characteristics such as leaf area, stem diameter, petiole angle and canopy cover, etc. Recent development of microcomputer-based image processing system has boosted wider applications of this technique in agriculture. Leaf area is one of the most important parameter for plant growth assessment. The determined information can be used not only in growth analysis but also in plant modeling, estimation of evapotranspiration, and yield prediction. A brief review and comparisons of machine vision approaches for non-destructive plant assessment are presented. Projected profile image acquired from a plant was most commonly used to predict actual total leaf area. Highly correlated relationships were reported between projected area and plant fresh or dry weight. Therefore, plant fresh or dry weight can be indirectly estimated from projected plant image by calibration. However, the estimation error usually increases as the extent of overlapped leaves increases. To solve this problem, an image processing method was developed to determine the overlapped leaf area. The position and approximate shape of overlapped seedling leaves were initially located using elliptical Hough transform. Based on this information, the hidden leaf boundary can be further reconstructed and the total leaf area can be calculated without pre-determined calibration relationship. This image-processing algorithm is incorporated into a stereo machine vision system to dynamically measure selected vegetable seedlings. Besides leaf area, this machine vision system was capable of measuring several other seedling features such as: seedling height, span, leaf and stem node number, leaf area index, etc. The 3-dimensional architecture of a vegetable seedling can be further reconstructed and graphically displayed based on the determined features.