基于机器视觉的红掌检测分级方法

摘要

With the raising of our consumption, the sales of flowers keep on expanding. Anthurium, as a prominent example, is facing some problems. Productivity and identical grading standards are restricted by the traditional method of grading in the actual process of production. These issues affect the quantity and quality of the product in China. In this paper, an online determination and grading method based on machine vision was developed. The variety of test materials was“flame,”which is popular in China. In accordance with the national standard, 30 pots of red anthurium of grade A, 30 pots of grade B and 30 pots of grade C, a total of 90 pots, were chosen to be test materials. After integrating national standards, the Ministry of Agriculture standards, and local standards about flowering potted plant, new standards were drawn out which are suitable for machine vision grading. In the standards, height, crown diameter, spathe quantity, and spathe diameter were to be the four items to be determined for every potted red anthurium. A horizontal camera and a vertical camera were used in this method to capture the side view and top view photos of 90 Anthurium one by one at the same time. After binarization and edge-based segmentation, the side contour, top contour, and spathe contour were extracted out. The side view photo was scanned in, the first point of side contour was obtained, and this point was regarded as the top point of this plant. The height could be calculated out through the coordinate of the top point. The result obtained by the traditional method, measured with a tapeline, was taken as the normal height. Experimental results show that the measurement error of height was less than 5.4 mm by the machine vision method. To reflect the growth state of the plant more accurately, this paper put forward an equivalent diameter method to use instead of crown diameter. In the top view photo, the crown area of each red anthurium was drawn out from the top contour, the top contour was transformed to an equivalent circle with the same area;and the diameter of equivalent circle was the equivalent diameter of the crown. There always may be some empty holes or gaps between leaves for a big-leaf plant in the top view photo. Therefore, the equivalent diameter method can avoid the effect of the empty holes and gap on the growth state of the plant. This method is more precise than using crown diameter. To count out the number of spathe in one red anthurium, an identification model of spathe was trained, which is complete and symmetry, and arranged all the identified spathe by similarity rate with the identification model from spathe contour. The result of this experiment showed that the correct rate for counting was up to 100%for the 90 pots red anthurium. Taking certain spathe with the highest similarity rate as the target, the widest diameter that is perpendicular with the main nervure was measured. Taking the result by traditional method, measurement was obtained with a tapeline as the normal spathe diameter. The maximum relative error of the spathe diameter was 6.5%. The result of the validate experiment showed that the success rate was up to 97.8%for grading 90 pots of red anthurium. This means that the method based on machine vision fulfilled the objective, which is online determination and grading for potted red anthurium.%中国盆花销售量随着消费水平的提高而迅速攀升，四大盆花之一的红掌在规模化生产中存在分级标准掌控不统一、分级结果不稳定以及效率低下等问题，制约了红掌产品的规模化生产，影响了红掌的出品质量。该研究提出基于机器视觉技术的红掌分级检测方法，对90株红掌的水平和竖直方向图像采集，经过二值化等处理，从侧视轮廓、俯视轮廓及佛焰苞轮廓特征三方面获取红掌植株的高度、冠幅、佛焰苞片数、苞片横径4项指标的信息，针对每一项指标提出检测方法。试验结果表明，该方法对植株高度测量误差小于5.4 mm；针对植株冠幅提出了当量直径测量方法，佛焰苞片数测量正确率可达98.9%，苞片横径相对误差最大为6.52%。针对试验所选90盆红掌分级成功率达到97.8%。研究表明，利用机器视觉技术能够很好地实现对红掌盆花的在线检测分级。