基于凸包算法的鸡蛋尺寸形状在线视觉高通量检测方法

High-throughput online detection method of egg size and shape based on convex hull algorithm

鸡蛋的尺寸形状是鸡蛋包装和销售以及种蛋挑选中需要考察的重要指标。目前鸡蛋的商品化处理需要高通量在线检测,然而检测速度和效率在高通量检测中要求较高。为了能够实现鸡蛋尺寸形状的高通量在线检测分级,该文在30000枚/h的传送装置上动态采集群体鸡蛋图像,采取有效的图像处理方法消除高速传输对鸡蛋图像的影响,结合应用凸包算法,快速准确提取出群体鸡蛋图像上的特征参数(长短轴表征尺寸大小、蛋形指数表征形状扁圆程度),最后按照尺寸大小与扁圆程度进行分级,其正确率分别为90.5%和89.3%,表明该方法对鸡蛋尺寸形状的高通量在线检测分级可行。

Size and shape of egg are the important indices which need to be investigated in egg packaging, marketing and selection of fertile eggs. Currently, high-throughput online detection is required in commercialization processing of egg. In order to achieve high-throughput online detection and grading for the size and shape of eggs, industrial camera（IDS company, UI-2210RE-C-HQ） was used to capture group egg images dynamically on transport mechanism, which reached 30 000 eggs per hour about its throughput in this study. There were 6 eggs that needed to be processed in each picture. Firstly, some light leaks produced by the clearances of transport mechanism were eliminated by deducting double green component from red component. Then Otsu＇s method was used for threshold segmentation based on the different information characteristics of each egg image. After binaryzation, hole filling was performed, the 2 2 circle structure elements were used to remove burrs on edge of egg, and then small area removing method was adopted in image pretreatment. Good effect was obtained in most of egg images, which presented the whole shape of egg. However,incomplete shape existed in some images after pre-processing due to the color similarity between light leak and part area of the egg. Therefore, the method of least- squares ellipse fitting was used to supplement the outline of egg, but unsatisfactory effect was showed from the fitting result. Concave had a bad effect on the fitting result and was required to remove. In order to remove the concave points of the outline better, the convex hull algorithm in the computational geometry was applied. Better effect was showed in the image with the method of least-squares ellipse fitting combined with convex hull algorithm, and the fitting outline of egg was more close to the actual outline. Size of egg was described by major and minor axis, and oblate degree was described by egg shape index. Sixty eggs were selected randomly to get actual axis sizes and the number of axis pixels. The correlation coefficient between the size of actual major axis and the number of fitting major axis pixels obtained by direct least-squares ellipse fitting was 0.8228, and 0.8620 for the minor axis. The correlation coefficient by the method of least-squares ellipse fitting combined with convex hull algorithm was0.9566 and 0.9439 respectively for the major and the minor axis. By comparing the correlation coefficient from 2methods, the method combined with convex hull algorithm was more proper. For testing the accuracy of grading, a request of size grading with the Z8- X type egg packing box was formulated, and the shape grading was based on the influence of egg shape index on hatchability. Eighty-four eggs were selected randomly to classify, and the accuracy of size and shape reached 90.5% and 89.3%, respectively. The experiment result shows that the image processing methods have a good effect on extracting the characteristic parameters of egg size and shape quickly and accurately, and among the methods, convex hull algorithm is of great significance. This method is believed to be feasible for the highthroughput online detection of egg size and shape, which can fulfill the requirements of production.