基于机器视觉的白掌组培苗在线分级方法

白掌在观叶类花卉中占有很大比例,其育苗多采用组织栽培法,且组培苗生产具有规模化。为提高成苗出苗品质,需要在组培苗炼苗前对其分级,而目前常用分级法不能有效解决自然状态下水平放置的白掌组培苗存在的叶片扭曲和重叠问题,因此该文提出一种基于机器视觉实现白掌组培苗在线分级的方法,通过对自然状态下水平放置的白掌组培苗的叶片面积、苗高、地径以及投影面积的分析,得到其投影面积与叶片面积呈线性关系,相关度为0.9344;投影面积与地径呈多项式函数关系,相关性为0.9067,故确定组培苗投影面积和苗高为实际生产中的分级指标。该文采用基于颜色模板匹配算法测量组培苗投影面积,得到的叶片面积和地径与实际叶片面积和地径的变异系数相对误差分别为0.35%和7.95%;利用最小外接矩形法(MBR,minimum bounding rectangle)测量苗高,得到的苗高和实际苗高变异系数相对误差为1.44%。通过整机分级试验发现在输送间距为0.25 m,输送速度为0.5 m/s,分级级别为3级的条件下,该分级装置的分级成功率可达96%,对应生产率为7 200株/h。

Online grading method for tissue culture seedlings of Spathiphyllum floribundum based on machine vision

Abstract: At present, most of young plants of Spathiphyllum floribundum are breeding by the technique of tissue culture. Due to absence of grading machine specially designed for primary-growth plants that is small, irregular and young, the grading of tissue culture seedlings are normally handled manually. In this paper, we proposed an automated online grading method for Spathiphyllum floribundum tissue culture seedlings based on the technique of machine vision. Since Spathiphyllum floribundum is a foliage flower, the leaf area is one of the most important parameters in grading, along with seedling height and diameter. Direct measurement not only would do damage to young plant because of its tenderness, but also the manpower productivity would decreased significantly. In our study, first, we grabbed the image of young plant under the natural state, and studied the relationship of actual parameters of Spathiphyllum floribundum tissue culture seedlings and the parameters of the image of Spathiphyllum floribundum tissue culture seedlings. By analyzing that leaf area of tissue culture seedlings image between actual leaf area and projection area of tissue culture seedlings, there was a linear relationship, and the regression coefficient R2 was 0.9344. It was time-consuming to measure diameter on ground by machine vision considering plants diversity, uncertain position and rotation angle. By analyzing, the relationship between projection area and diameter on ground we found that actual diameter on ground had a polynomial function with projection area, and the regression coefficient R2 was 0.9067. We also found that correlation of projection area and actual seedling height of tissue culture was insignificant. Ultimately, we reached the conclusion that the tissue culture seedlings can be graded by projection area and seedling height of the young plant image. The second task of this paper was to use machine vision to realize automatic grading algorithms according to the above conclusions. Considering the influence of shadow, a color image matching algorithm was executed for extracting projection area. Using the color of leaf, stem and root as a template, projection area could be intact segmented from background when darker was equal to 0.4, highlight was equal to 1.5 and hue was equal to 2.0. Based on the functional relationship between leaf area and projection area, leaf area could be calculated directly. The same procedure may be easily adapted to obtain diameter on ground. The ultimate positions of tissue culture seedling is stochastic in the field of camera view, thus the main difficulty for seedling height depended on how the measurement position was accurately determined. In this article, we adopted the method of seedling height based on MBR (minimum bounding rectangle). In the first step, color-image was preprocessed with gray scaling, and then binarization was implemented on grayscale image. Image matching algorithm had the most effective result when binary threshold value was equal to 103. Finally, a MBR of binarization image was obtained. Because tissue culture seedlings were shaped like strips, the length of MBR could be used to determine seedling height. By the testing of vision algorithm it was found that the relative error of coefficient of variation of leaf area, diameter in ground and seedling height was 0.35%, 7.95% and 1.44% respectively. Lastly, an online grading machine was made to test the grading precision and the productivity. The machine consisted of conveyor, machine vision detection device, grading unit and control unit. Test results revealed that besides the effectiveness of vision algorithms, the factors which determined the success rate of grading machine also included the distance between two young plants and the speed of the conveyor belt. By orthogonal experiment and range analysis, the results showed that the distance between two tissue culture seedlings had the most greatly influence to the success rate of grading machine. The second factor was the speed of the conveyor belt, the size of young plant had the least influence. In the end, the conclusion was that when the distance between two tissue culture seedlings was 0.25 m, the conveyor belt speed was 0.5 m/s the grading machine can get the highest success rate with minimal time consumption. In this condition, the typical speed of grading machine can reach 7 200 plants/h when hierarchical level for three levels, the grading precision can reach above 96%.