基于图像特征R通道检测玉米粒霉变及霉变程度的方法

Method for detecting corn mold mildew and mildew degree based on image feature R channel

霉变玉米粒对食品业、畜牧业和工业加工都有影响,因此快速识别霉变玉米粒具有重要应用意义。通过试验发现,正常玉米粒与霉变玉米粒在R通道上的像素值界限分明,G通道次之,而在B通道上两者的像素值大致相同,据此提出一种基于R通道来检测霉变玉米粒的方法。该方法利用正常/霉变玉米粒R通道上像素值的特征,对测试集中的玉米粒进行数理统计分析,获得正常/霉变玉米粒R通道像素值的取值区间。通过玉米粒图像处理得到总像素点数和霉变像素点数,并定义霉变像素点数与总像素点数之比为占比,通过非参数估测方法获取正常/霉变玉米粒占比交点作为判断玉米粒是否霉变的阈值。本文还探讨霉变玉米粒的霉变程度,即将霉变像素值的区间划分为三个子区间,分别得到重度、中度和轻度霉变像素点的数量,并分别依据占比将霉变玉米粒划分为重度、中度和轻度霉变。试验结果表明,所提出的方法对玉米粒是否属于正常还是霉变,检测的正确率都达到99%;对同一批霉变玉米粒进行不同位置重复试验,也能较稳定地识别出霉变玉米粒的霉变程度。

Moldy corn kernels have an impact on the food industry,animal husbandry,and industrial processing,so it is important to quickly identify moldy corn kernels.Through experiments,it was found that the pixel values of normal corn kernels and moldy corn kernels on R channel were clearly defined,followed by G channel,while the pixel values of both on B channel were roughly the same.Therefore,a method for detecting moldy corn kernels based on R channel was proposed.In this method,the characteristics of the pixel values on the R channel of normal/moldy corn kernels were used to perform mathematical statistical analysis on the kernels in the test set,and the value range of the pixel values on the R channel of normal/moldy corn kernels was obtained.The total number of pixels and the number of moldy pixels were obtained by image processing of corn kernels,and the ratio of moldy pixel points to total number of pixels was defined as the proportion,and the intersection point of normal/moldy corn kernels was obtained by non-parametric estimation method as the threshold to judge whether corn kernels were moldy or not.This paper also discusses the degree of mildew of mildewed corn kernels,which divides the mildewed pixel value into three sub-intervals to obtain the number of severely mildewed,moderately mildewed and mildly mildewed pixel points,and divides mildewed corn kernels into severely mildewed,moderately mildewed and mildly mildewed kernels according to their proportions.The experimental results show that the proposed method is 99%accurate in determining whether the corn kernels are normal or moldy.The mildew degree of the same batch of mildew corn kernels can also be identified stably by repeated experiments at different locations.